Last year, China saw a boom in foundation models, the do-everything large language models that underpin the AI revolution. This year, the focus has shifted to AI agents—systems that are less about responding to users’ queries and more about autonomously accomplishing things for them.  There are now a host of Chinese startups building these general-purpose digital tools, which can answer emails, browse the internet to plan vacations, and even design an interactive website. Many of these have emerged in just the last two months, following in the footsteps of Manus—a general AI agent that sparked weeks of social media frenzy for invite codes after its limited-release launch in early March.  These emerging AI agents aren’t large language models themselves. Instead, they’re built on top of them, using a workflow-based structure designed to get things done. A lot of these systems also introduce a different way of interacting with AI. Rather than just chatting back and forth with users, they are optimized for managing and executing multistep tasks—booking flights, managing schedules, conducting research—by using external tools and remembering instructions.  China could take the lead on building these kinds of agents. The country’s tightly integrated app ecosystems, rapid product cycles, and digitally fluent user base could provide a favorable environment for embedding AI into daily life.  For now, its leading AI agent startups are focusing their attention on the global market, because the best Western models don’t operate inside China’s firewalls. But that could change soon: Tech giants like ByteDance and Tencent are preparing their own AI agents that could bake automation directly into their native super-apps, pulling data from their vast ecosystem of programs that dominate many aspects of daily life in the country.  As the race to define what a useful AI agent looks like unfolds, a mix of ambitious startups and entrenched tech giants are now testing how these tools might actually work in practice—and for whom. Set the standard It’s been a whirlwind few months for Manus, which was developed by the Wuhan-based startup Butterfly Effect. The company raised million in a funding round led by the US venture capital firm Benchmark, took the product on an ambitious global roadshow, and hired dozens of new employees.  Even before registration opened to the public in May, Manus had become a reference point for what a broad, consumer‑oriented AI agent should accomplish. Rather than handling narrow chores for businesses, this “general” agent is designed to be able to help with everyday tasks like trip planning, stock comparison, or your kid’s school project.  Unlike previous AI agents, Manus uses a browser-based sandbox that lets users supervise the agent like an intern, watching in real time as it scrolls through web pages, reads articles, or codes actions. It also proactively asks clarifying questions, supports long-term memory that would serve as context for future tasks. "Manus represents a promising product experience for AI agents," says Ang Li, cofounder and CEO of Simular, a startup based in Palo Alto, California, that’s building computer use agents, AI agents that control a virtual computer. “I believe Chinese startups have a huge advantage when it comes to designing consumer products, thanks to cutthroat domestic competition that leads to fast execution and greater attention to product details.” In the case of Manus, the competition is moving fast. Two of the most buzzy follow‑ups, Genspark and Flowith, for example, are already boasting benchmark scores that match or edge past Manus’s.  Genspark, led by former Baidu executives Eric Jing and Kay Zhu, links many small “super agents” through what it calls multi‑component prompting. The agent can switch among several large language models, accepts both images and text, and carries out tasks from making slide decks to placing phone calls. Whereas Manus relies heavily on Browser Use, a popular open-source product that lets agents operate a web browser in a virtual window like a human, Genspark directly integrates with a wide array of tools and APIs. Launched in April, the company says that it already has over 5 million users and over million in yearly revenue. Flowith, the work of a young team that first grabbed public attention in April 2025 at a developer event hosted by the popular social media app Xiaohongshu, takes a different tack. Marketed as an “infinite agent,” it opens on a blank canvas where each question becomes a node on a branching map. Users can backtrack, take new branches, and store results in personal or sharable “knowledge gardens”—a design that feels more like project management softwarethan a typical chat interface. Every inquiry or task builds its own mind-map-like graph, encouraging a more nonlinear and creative interaction with AI. Flowith’s core agent, NEO, runs in the cloud and can perform scheduled tasks like sending emails and compiling files. The founders want the app to be a “knowledge marketbase”, and aims to tap into the social aspect of AI with the aspiration of becoming “the OnlyFans of AI knowledge creators”. What they also share with Manus is the global ambition. Both Genspark and Flowith have stated that their primary focus is the international market. A global address Startups like Manus, Genspark, and Flowith—though founded by Chinese entrepreneurs—could blend seamlessly into the global tech scene and compete effectively abroad. Founders, investors, and analysts that MIT Technology Review has spoken to believe Chinese companies are moving fast, executing well, and quickly coming up with new products.  Money reinforces the pull to launch overseas. Customers there pay more, and there are plenty to go around. “You can price in USD, and with the exchange rate that’s a sevenfold multiplier,” Manus cofounder Xiao Hong quipped on a podcast. “Even if we’re only operating at 10% power because of cultural differences overseas, we’ll still make more than in China.” But creating the same functionality in China is a challenge. Major US AI companies including OpenAI and Anthropic have opted out of mainland China because of geopolitical risks and challenges with regulatory compliance. Their absence initially created a black market as users resorted to VPNs and third-party mirrors to access tools like ChatGPT and Claude. That vacuum has since been filled by a new wave of Chinese chatbots—DeepSeek, Doubao, Kimi—but the appetite for foreign models hasn’t gone away.  Manus, for example, uses Anthropic’s Claude Sonnet—widely considered the top model for agentic tasks. Manus cofounder Zhang Tao has repeatedly praised Claude’s ability to juggle tools, remember contexts, and hold multi‑round conversations—all crucial for turning chatty software into an effective executive assistant. But the company’s use of Sonnet has made its agent functionally unusable inside China without a VPN. If you open Manus from a mainland IP address, you’ll see a notice explaining that the team is “working on integrating Qwen’s model,” a special local version that is built on top of Alibaba’s open-source model.  An engineer overseeing ByteDance’s work on developing an agent, who spoke to MIT Technology Review anonymously to avoid sanction, said that the absence of Claude Sonnet models “limits everything we do in China.” DeepSeek’s open models, he added, still hallucinate too often and lack training on real‑world workflows. Developers we spoke with rank Alibaba’s Qwen series as the best domestic alternative, yet most say that switching to Qwen knocks performance down a notch. Jiaxin Pei, a postdoctoral researcher at Stanford’s Institute for Human‑Centered AI, thinks that gap will close: “Building agentic capabilities in base LLMs has become a key focus for many LLM builders, and once people realize the value of this, it will only be a matter of time." For now, Manus is doubling down on audiences it can already serve. In a written response, the company said its “primary focus is overseas expansion,” noting that new offices in San Francisco, Singapore, and Tokyo have opened in the past month. A super‑app approach Although the concept of AI agents is still relatively new, the consumer-facing AI app market in China is already crowded with major tech players. DeepSeek remains the most widely used, while ByteDance’s Doubao and Moonshot’s Kimi have also become household names. However, most of these apps are still optimized for chat and entertainment rather than task execution. This gap in the local market has pushed China’s big tech firms to roll out their own user-facing agents, though early versions remain uneven in quality and rough around the edges.  ByteDance is testing Coze Space, an AI agent based on its own Doubao model family that lets users toggle between “plan” and “execute” modes, so they can either directly guide the agent’s actions or step back and watch it work autonomously. It connects up to 14 popular apps, including GitHub, Notion, and the company’s own Lark office suite. Early reviews say the tool can feel clunky and has a high failure rate, but it clearly aims to match what Manus offers. Meanwhile, Zhipu AI has released a free agent called AutoGLM Rumination, built on its proprietary ChatGLM models. Shanghai‑based Minimax has launched Minimax Agent. Both products look almost identical to Manus and demo basic tasks such as building a simple website, planning a trip, making a small Flash game, or running quick data analysis. Despite the limited usability of most general AI agents launched within China, big companies have plans to change that. During a May 15 earnings call, Tencent president Liu Zhiping teased an agent that would weave automation directly into China’s most ubiquitous app, WeChat.  Considered the original super-app, WeChat already handles messaging, mobile payments, news, and millions of mini‑programs that act like embedded apps. These programs give Tencent, its developer, access to data from millions of services that pervade everyday life in China, an advantage most competitors can only envy. Historically, China’s consumer internet has splintered into competing walled gardens—share a Taobao link in WeChat and it resolves as plaintext, not a preview card. Unlike the more interoperable Western internet, China’s tech giants have long resisted integration with one another, choosing to wage platform war at the expense of a seamless user experience. But the use of mini‑programs has given WeChat unprecedented reach across services that once resisted interoperability, from gym bookings to grocery orders. An agent able to roam that ecosystem could bypass the integration headaches dogging independent startups. Alibaba, the e-commerce giant behind the Qwen model series, has been a front-runner in China's AI race but has been slower to release consumer-facing products. Even though Qwen was the most downloaded open-source model on Hugging Face in 2024, it didn’t power a dedicated chatbot app until early 2025. In March, Alibaba rebranded its cloud storage and search app Quark into an all-in-one AI search tool. By June, Quark had introduced DeepResearch—a new mode that marks its most agent-like effort to date.  ByteDance and Alibaba did not reply to MIT Technology Review’s request for comments. “Historically, Chinese tech products tend to pursue the all-in-one, super-app approach, and the latest Chinese AI agents reflect just that,” says Li of Simular, who previously worked at Google DeepMind on AI-enabled work automation. “In contrast, AI agents in the US are more focused on serving specific verticals.” Pei, the researcher at Stanford, says that existing tech giants could have a huge advantage in bringing the vision of general AI agents to life—especially those with built-in integration across services. “The customer-facing AI agent market is still very early, with tons of problems like authentication and liability,” he says. “But companies that already operate across a wide range of services have a natural advantage in deploying agents at scale.” #manus #has #kickstarted #agent #boom

AGI will be embedded into humanoid robots, which makes white-collar and blue-collar jobs a target ... More for walking/talking automation.getty In today’s column, I examine the highly worrisome qualms expressed that the advent of artificial general intelligenceis likely to usurp white-collar jobs. The stated concern is that since AGI will be on par with human intellect, any job that relies principally on intellectual pursuits such as typical white-collar work will be taken over via the use of AGI. Employers will realize that rather than dealing with human white-collar workers, they can more readily get the job done via AGI. This, in turn, has led to a rising call that people should aim toward blue-collar jobs, doing so becausethose forms of employment will not be undercut via AGI. Sorry to say, that misses the bigger picture, namely that AGI when combined with humanoid robots is coming not only for white-collar jobs but also blue-collar jobs too. It is a proverbial double-whammy when it comes to the attainment of AGI. Let’s talk about it. This analysis of an innovative AI breakthrough is part of my ongoing Forbes column coverage on the latest in AI, including identifying and explaining various impactful AI complexities. Heading Toward AGI And ASI First, some fundamentals are required to set the stage for this weighty discussion. There is a great deal of research going on to further advance AI. The general goal is to either reach artificial general intelligenceor maybe even the outstretched possibility of achieving artificial superintelligence. AGI is AI that is considered on par with human intellect and can seemingly match our intelligence. ASI is AI that has gone beyond human intellect and would be superior in many if not all feasible ways. The idea is that ASI would be able to run circles around humans by outthinking us at every turn. For more details on the nature of conventional AI versus AGI and ASI, see my analysis at the link here. We have not yet attained AGI. In fact, it is unknown as to whether we will reach AGI, or that maybe AGI will be achievable in decades or perhaps centuries from now. The AGI attainment dates that are floating around are wildly varying and wildly unsubstantiated by any credible evidence or ironclad logic. ASI is even more beyond the pale when it comes to where we are currently with conventional AI. AGI Problem Only Half Seen Before launching into the primary matter at hand in this discussion, let’s contemplate a famous quote attributed to Charles Kettering, a legendary inventor, who said, “A problem well-stated is a problem half-solved.” I bring this up because those loud clamors right now about the assumption that AGI will replace white-collar workers are only seeing half of the problem. The problem as they see it is that since AGI is intellectually on par with humans, and since white-collar workers mainly use intellect in their work endeavors, AGI is going to be used in place of humans for white-collar work. I will in a moment explain why that’s only half of the problem and there is a demonstrative need to more carefully and fully articulate the nature of the problem. Will AGI Axiomatically Take White-Collar Jobs On a related facet, the belief that AGI will axiomatically replace white-collar labor makes a number of other related key assumptions. I shall briefly explore those and then come back to why the problem itself is only half-baked. The cost of using AGI for doing white-collar work will need to be presumably a better ROI choice over human workers. If not, then an employer would be wiser to stick with humans rather than employing AGI. There seems to often be an unstated belief that AGI is necessarily going to be a less costly route than employing humans. We don’t know yet what the cost of using AGI will be. It could be highly expensive. Indeed, some are worried that the world will divide into the AGI haves and AGI have-nots, partially due to the exorbitant cost that AGI might involve. If AGI is free to use, well, that would seem to be the nail in the coffin related to using human workers for the same capacity. Another angle is that AGI is relatively inexpensive in comparison to human labor. In that case, the use of AGI is likely to win over human labor usage. But if the cost of AGI is nearer to the cost of human labor, or more so, then employers would rationally need to weigh the use of one versus the other. Note that when referring to the cost of human labor, there is more to that calculation than simply the dollar-hour labor rate per se. There are lots of other less apparent costs, such as the cost to manage human labor, the cost of dealing with HR-related issues, and many other factors that come into the weighty matter. Thus, an AGI versus human labor ROI will be more complex than it might seem at an initial glance. In addition, keep in mind that AGI would seemingly be readily switched on and off, and have other capacities that human labor would not equally tend to allow. The Other Half Is Coming Too Assume that by and large the advent of AGI will decimate the need for white-collar human labor. The refrain right now is that people should begin tilting toward blue-collar jobs as an alternative to white-collar jobs. This is a logical form of thinking in the sense that AGI as an intellectual mechanism would be unable to compete in jobs that involve hands-on work. A plumber needs to come to your house and do hands-on work to fix your plumbing. This is a physicality that entails arriving at your physical home, physically bringing and using tools, and physically repairing your faulty home plumbing. A truck driver likewise needs to sit in the cab of a truck and drive the vehicle. These are physically based tasks. There is no getting around the fact that these are hands-on activities. Aha, yes, those are physical tasks, but that doesn’t necessarily mean that only human hands can perform them. The gradual emergence of humanoid robots will provide an alternative to human hands. A humanoid robot is a type of robot that is built to resemble a human in form and function. You’ve undoubtedly seen those types of robots in the many online video recordings showing them walking, jumping, grasping at objects, and so on. A tremendous amount of active research and development is taking place to devise humanoid robots. They look comical right now. You watch those videos and laugh when the robot trips over a mere stick lying on the ground, something that a human would seldom trip over. You scoff when a robot tries to grasp a coffee cup and inadvertently spills most of the coffee. It all seems humorous and a silly pursuit. Keep in mind that we are all observing the development process while it is still taking place. At some point, those guffaws of the humanoid robots will lessen. Humanoid robots will be as smooth and graceful as humans. This will continue to be honed. Eventually, humanoid robots will be less prone to physical errors that humans make. In a sense, the physicality of a humanoid robot will be on par with humans, if not better, due to its mechanical properties. Do not discount the coming era of quite physically capable humanoid robots. AGI And Humanoid Robots Pair Up You might remember that in The Wonderful Wizard of Oz, the fictional character known as The Strawman lacked a brain. Without seeming to anthropomorphize humanoid robots, the current situation is that those robots typically use a form of AI that is below the sophistication level of modern generative AI. That’s fine for now due to the need to first ensure that the physical movements of the robots get refined. I have discussed that a said-to-be realm of Physical AI is going to be a huge breakthrough with incredible ramifications, see my analysis at the link here. The idea underlying Physical AI is that the AI of today is being uplifted by doing data training on the physical world. This also tends to include the use of World Models, consisting of broad constructions about how the physical world works, such as that we are bound to operate under conditions of gravity, and other physical laws of nature, see the link here. The bottom line here is that there will be a close pairing of robust AI with humanoid robots. Imagine what a humanoid robot can accomplish if it is paired with AGI. I’ll break the suspense and point out that AGI paired with humanoid robots means that those robots readily enter the blue-collar worker realm. Suppose your plumbing needs fixing. No worries, a humanoid robot that encompasses AGI will be sent to your home. The AGI is astute enough to carry on conversations with you, and the AGI also fully operates the robot to undertake the plumbing tasks. How did the AGI-paired humanoid robot get to your home? Easy-peasy, it drove a car or truck to get there. I’ve previously predicted that all the work on devising autonomous vehicles and self-driving cars will get shaken up once we have suitable humanoid robots devised. There won’t be a need for a vehicle to contain self-driving capabilities. A humanoid robot will simply sit in the driver’s seat and drive the vehicle. This is a much more open-ended solution than having to craft components that go into and onto a vehicle to enable self-driving. See my coverage at the link here. Timing Is Notable One of the reasons that many do not give much thought to the pairing of AGI with humanoid robots is that today’s humanoid robots seem extraordinarily rudimentary and incapable of performing physical dexterity tasks on par with human capabilities. Meanwhile, there is brazen talk that AGI is just around the corner. AGI is said to be within our grasp. Let’s give the timing considerations a bit of scrutiny. There are three primary timing angles: Option 1: AGI first, then humanoid robots. AGI is attained before humanoid robots are sufficiently devised. Option 2: Humanoid robots first, then AGI. Humanoid robots are physically fluently adept before AGI is attained. Option 3: AGI and humanoid robots arrive about at the same time. AGI is attained and at the same time, it turns out that humanoid robots are fluently adept too, mainly by coincidence and not due to any cross-mixing. A skeptic would insist that there is a fourth possibility, consisting of the possibility that we never achieve AGI and/or we fail to achieve sufficiently physically capable humanoid robots. I am going to reject that possibility. Perhaps I am overly optimistic, but it seems to me that we will eventually attain AGI, and we will eventually attain physically capable humanoid robots. I shall next respectively consider each of the three genuinely reasonable possibilities. Option 1: AGI First, Then Humanoid Robots What if we manage to attain AGI before we manage to achieve physically fluent humanoid robots? That’s just fine. We would indubitably put AGI to work as a partner with humans in figuring out how we can push along the budding humanoid robot development process. It seems nearly obvious that with AGI’s capable assistance, we would overcome any bottlenecks and soon enough arrive at top-notch physically adept humanoid robots. At that juncture, we would then toss AGI into the humanoid robots and have ourselves quite an amazing combination. Option 2: Humanoid Robots First, Then AGI Suppose that we devise very physically adept humanoid robots but have not yet arrived at AGI. Are we in a pickle? Nope. We could use conventional advanced AI inside those humanoid robots. The combination would certainly be good enough for a wide variety of tasks. The odds are that we would need to be cautious about where such robots are utilized. Nonetheless, we would have essentially walking, talking, and productive humanoid robots. If AGI never happens, oh well, we end up with pretty good humanoid robots. On the other hand, once we arrive at AGI, those humanoid robots will be stellar. It’s just a matter of time. Option 3: AGI And Humanoid Robots At The Same Time Let’s consider the potential of AGI and humanoid robots perchance being attained around the same time. Assume that this timing isn’t due to an outright cross-mixing with each other. They just so happen to advance on a similar timeline. I tend to believe that’s the most likely of the three scenarios. Here’s why. First, despite all the hubris about AGI being within earshot, perhaps in the next year or two, which is a popular pronouncement by many AI luminaries, I tend to side with recent surveys of AI developers that put the date around the year 2040. Some AI luminaires sneakily play with the definition of AGI in hopes of making their predictions come true sooner, akin to moving the goalposts to easily score points. For my coverage on Sam Altman’s efforts of moving the cheese regarding AGI attainment, see the link here. Second, if you are willing to entertain the year 2040 as a potential date for achieving AGI, that’s about 15 years from now. In my estimation, the advancements being made in humanoid robots will readily progress such that by 2040 they will be very physically adept. Probably be sooner, but let’s go with the year 2040 for ease of contemplation. In my view, we will likely have humanoid robots doing well enough that they will be put into use prior to arriving at AGI. The pinnacle of robust humanoid robots and the attainment of AGI will roughly coincide with each other. Two peas in a pod.Impact Of Enormous Consequences In an upcoming column posting, I will examine the enormous consequences of having AGI paired with fully physically capable humanoid robots. As noted above, this will have a humongous impact on white-collar work and blue-collar work. There will be gargantuan economic impacts, societal impacts, cultural impacts, and so on. Some final thoughts for now. A single whammy is already being hotly debated. The debates currently tend to be preoccupied with the loss of white-collar jobs due to the attainment of AGI. A saving grace seems to be that at least blue-collar jobs are going to be around and thriving, even once AGI is attained. The world doesn’t seem overly gloomy if you can cling to the upbeat posture that blue-collar tasks remain intact. The double whammy is a lot more to take in. But the double whammy is the truth. The truth needs to be faced. If you are having doubts as a human about the future, just remember the famous words of Vince Lombardi: “Winners never quit, and quitters never win.” Humankind can handle the double whammy. Stay tuned for my upcoming coverage of what this entails. #doublewhammy #when #agi #embeds #with

Last year, China saw a boom in foundation models, the do-everything large language models that underpin the AI revolution. This year, the focus has shifted to AI agents—systems that are less about responding to users’ queries and more about autonomously accomplishing things for them.  There are now a host of Chinese startups building these general-purpose digital tools, which can answer emails, browse the internet to plan vacations, and even design an interactive website. Many of these have emerged in just the last two months, following in the footsteps of Manus—a general AI agent that sparked weeks of social media frenzy for invite codes after its limited-release launch in early March.  These emerging AI agents aren’t large language models themselves. Instead, they’re built on top of them, using a workflow-based structure designed to get things done. A lot of these systems also introduce a different way of interacting with AI. Rather than just chatting back and forth with users, they are optimized for managing and executing multistep tasks—booking flights, managing schedules, conducting research—by using external tools and remembering instructions.  China could take the lead on building these kinds of agents. The country’s tightly integrated app ecosystems, rapid product cycles, and digitally fluent user base could provide a favorable environment for embedding AI into daily life.  For now, its leading AI agent startups are focusing their attention on the global market, because the best Western models don’t operate inside China’s firewalls. But that could change soon: Tech giants like ByteDance and Tencent are preparing their own AI agents that could bake automation directly into their native super-apps, pulling data from their vast ecosystem of programs that dominate many aspects of daily life in the country.  As the race to define what a useful AI agent looks like unfolds, a mix of ambitious startups and entrenched tech giants are now testing how these tools might actually work in practice—and for whom. Set the standard It’s been a whirlwind few months for Manus, which was developed by the Wuhan-based startup Butterfly Effect. The company raised million in a funding round led by the US venture capital firm Benchmark, took the product on an ambitious global roadshow, and hired dozens of new employees.  Even before registration opened to the public in May, Manus had become a reference point for what a broad, consumer‑oriented AI agent should accomplish. Rather than handling narrow chores for businesses, this “general” agent is designed to be able to help with everyday tasks like trip planning, stock comparison, or your kid’s school project.  Unlike previous AI agents, Manus uses a browser-based sandbox that lets users supervise the agent like an intern, watching in real time as it scrolls through web pages, reads articles, or codes actions. It also proactively asks clarifying questions, supports long-term memory that would serve as context for future tasks. "Manus represents a promising product experience for AI agents," says Ang Li, cofounder and CEO of Simular, a startup based in Palo Alto, California, that’s building computer use agents, AI agents that control a virtual computer. “I believe Chinese startups have a huge advantage when it comes to designing consumer products, thanks to cutthroat domestic competition that leads to fast execution and greater attention to product details.” In the case of Manus, the competition is moving fast. Two of the most buzzy follow‑ups, Genspark and Flowith, for example, are already boasting benchmark scores that match or edge past Manus’s.  Genspark, led by former Baidu executives Eric Jing and Kay Zhu, links many small “super agents” through what it calls multi‑component prompting. The agent can switch among several large language models, accepts both images and text, and carries out tasks from making slide decks to placing phone calls. Whereas Manus relies heavily on Browser Use, a popular open-source product that lets agents operate a web browser in a virtual window like a human, Genspark directly integrates with a wide array of tools and APIs. Launched in April, the company says that it already has over 5 million users and over million in yearly revenue. Flowith, the work of a young team that first grabbed public attention in April 2025 at a developer event hosted by the popular social media app Xiaohongshu, takes a different tack. Marketed as an “infinite agent,” it opens on a blank canvas where each question becomes a node on a branching map. Users can backtrack, take new branches, and store results in personal or sharable “knowledge gardens”—a design that feels more like project management softwarethan a typical chat interface. Every inquiry or task builds its own mind-map-like graph, encouraging a more nonlinear and creative interaction with AI. Flowith’s core agent, NEO, runs in the cloud and can perform scheduled tasks like sending emails and compiling files. The founders want the app to be a “knowledge marketbase”, and aims to tap into the social aspect of AI with the aspiration of becoming “the OnlyFans of AI knowledge creators”. What they also share with Manus is the global ambition. Both Genspark and Flowith have stated that their primary focus is the international market. A global address Startups like Manus, Genspark, and Flowith—though founded by Chinese entrepreneurs—could blend seamlessly into the global tech scene and compete effectively abroad. Founders, investors, and analysts that MIT Technology Review has spoken to believe Chinese companies are moving fast, executing well, and quickly coming up with new products.  Money reinforces the pull to launch overseas. Customers there pay more, and there are plenty to go around. “You can price in USD, and with the exchange rate that’s a sevenfold multiplier,” Manus cofounder Xiao Hong quipped on a podcast. “Even if we’re only operating at 10% power because of cultural differences overseas, we’ll still make more than in China.” But creating the same functionality in China is a challenge. Major US AI companies including OpenAI and Anthropic have opted out of mainland China because of geopolitical risks and challenges with regulatory compliance. Their absence initially created a black market as users resorted to VPNs and third-party mirrors to access tools like ChatGPT and Claude. That vacuum has since been filled by a new wave of Chinese chatbots—DeepSeek, Doubao, Kimi—but the appetite for foreign models hasn’t gone away.  Manus, for example, uses Anthropic’s Claude Sonnet—widely considered the top model for agentic tasks. Manus cofounder Zhang Tao has repeatedly praised Claude’s ability to juggle tools, remember contexts, and hold multi‑round conversations—all crucial for turning chatty software into an effective executive assistant. But the company’s use of Sonnet has made its agent functionally unusable inside China without a VPN. If you open Manus from a mainland IP address, you’ll see a notice explaining that the team is “working on integrating Qwen’s model,” a special local version that is built on top of Alibaba’s open-source model.  An engineer overseeing ByteDance’s work on developing an agent, who spoke to MIT Technology Review anonymously to avoid sanction, said that the absence of Claude Sonnet models “limits everything we do in China.” DeepSeek’s open models, he added, still hallucinate too often and lack training on real‑world workflows. Developers we spoke with rank Alibaba’s Qwen series as the best domestic alternative, yet most say that switching to Qwen knocks performance down a notch. Jiaxin Pei, a postdoctoral researcher at Stanford’s Institute for Human‑Centered AI, thinks that gap will close: “Building agentic capabilities in base LLMs has become a key focus for many LLM builders, and once people realize the value of this, it will only be a matter of time." For now, Manus is doubling down on audiences it can already serve. In a written response, the company said its “primary focus is overseas expansion,” noting that new offices in San Francisco, Singapore, and Tokyo have opened in the past month. A super‑app approach Although the concept of AI agents is still relatively new, the consumer-facing AI app market in China is already crowded with major tech players. DeepSeek remains the most widely used, while ByteDance’s Doubao and Moonshot’s Kimi have also become household names. However, most of these apps are still optimized for chat and entertainment rather than task execution. This gap in the local market has pushed China’s big tech firms to roll out their own user-facing agents, though early versions remain uneven in quality and rough around the edges.  ByteDance is testing Coze Space, an AI agent based on its own Doubao model family that lets users toggle between “plan” and “execute” modes, so they can either directly guide the agent’s actions or step back and watch it work autonomously. It connects up to 14 popular apps, including GitHub, Notion, and the company’s own Lark office suite. Early reviews say the tool can feel clunky and has a high failure rate, but it clearly aims to match what Manus offers. Meanwhile, Zhipu AI has released a free agent called AutoGLM Rumination, built on its proprietary ChatGLM models. Shanghai‑based Minimax has launched Minimax Agent. Both products look almost identical to Manus and demo basic tasks such as building a simple website, planning a trip, making a small Flash game, or running quick data analysis. Despite the limited usability of most general AI agents launched within China, big companies have plans to change that. During a May 15 earnings call, Tencent president Liu Zhiping teased an agent that would weave automation directly into China’s most ubiquitous app, WeChat.  Considered the original super-app, WeChat already handles messaging, mobile payments, news, and millions of mini‑programs that act like embedded apps. These programs give Tencent, its developer, access to data from millions of services that pervade everyday life in China, an advantage most competitors can only envy. Historically, China’s consumer internet has splintered into competing walled gardens—share a Taobao link in WeChat and it resolves as plaintext, not a preview card. Unlike the more interoperable Western internet, China’s tech giants have long resisted integration with one another, choosing to wage platform war at the expense of a seamless user experience. But the use of mini‑programs has given WeChat unprecedented reach across services that once resisted interoperability, from gym bookings to grocery orders. An agent able to roam that ecosystem could bypass the integration headaches dogging independent startups. Alibaba, the e-commerce giant behind the Qwen model series, has been a front-runner in China's AI race but has been slower to release consumer-facing products. Even though Qwen was the most downloaded open-source model on Hugging Face in 2024, it didn’t power a dedicated chatbot app until early 2025. In March, Alibaba rebranded its cloud storage and search app Quark into an all-in-one AI search tool. By June, Quark had introduced DeepResearch—a new mode that marks its most agent-like effort to date.  ByteDance and Alibaba did not reply to MIT Technology Review’s request for comments. “Historically, Chinese tech products tend to pursue the all-in-one, super-app approach, and the latest Chinese AI agents reflect just that,” says Li of Simular, who previously worked at Google DeepMind on AI-enabled work automation. “In contrast, AI agents in the US are more focused on serving specific verticals.” Pei, the researcher at Stanford, says that existing tech giants could have a huge advantage in bringing the vision of general AI agents to life—especially those with built-in integration across services. “The customer-facing AI agent market is still very early, with tons of problems like authentication and liability,” he says. “But companies that already operate across a wide range of services have a natural advantage in deploying agents at scale.” #manus #has #kickstarted #agent #boom

Last year, China saw a boom in foundation models, the do-everything large language models that underpin the AI revolution. This year, the focus has shifted to AI agents—systems that are less about responding to users’ queries and more about autonomously accomplishing things for them.  There are now a host of Chinese startups building these general-purpose digital tools, which can answer emails, browse the internet to plan vacations, and even design an interactive website. Many of these have emerged in just the last two months, following in the footsteps of Manus—a general AI agent that sparked weeks of social media frenzy for invite codes after its limited-release launch in early March.  These emerging AI agents aren’t large language models themselves. Instead, they’re built on top of them, using a workflow-based structure designed to get things done. A lot of these systems also introduce a different way of interacting with AI. Rather than just chatting back and forth with users, they are optimized for managing and executing multistep tasks—booking flights, managing schedules, conducting research—by using external tools and remembering instructions.  China could take the lead on building these kinds of agents. The country’s tightly integrated app ecosystems, rapid product cycles, and digitally fluent user base could provide a favorable environment for embedding AI into daily life.  For now, its leading AI agent startups are focusing their attention on the global market, because the best Western models don’t operate inside China’s firewalls. But that could change soon: Tech giants like ByteDance and Tencent are preparing their own AI agents that could bake automation directly into their native super-apps, pulling data from their vast ecosystem of programs that dominate many aspects of daily life in the country.  As the race to define what a useful AI agent looks like unfolds, a mix of ambitious startups and entrenched tech giants are now testing how these tools might actually work in practice—and for whom. Set the standard It’s been a whirlwind few months for Manus, which was developed by the Wuhan-based startup Butterfly Effect. The company raised million in a funding round led by the US venture capital firm Benchmark, took the product on an ambitious global roadshow, and hired dozens of new employees.  Even before registration opened to the public in May, Manus had become a reference point for what a broad, consumer‑oriented AI agent should accomplish. Rather than handling narrow chores for businesses, this “general” agent is designed to be able to help with everyday tasks like trip planning, stock comparison, or your kid’s school project.  Unlike previous AI agents, Manus uses a browser-based sandbox that lets users supervise the agent like an intern, watching in real time as it scrolls through web pages, reads articles, or codes actions. It also proactively asks clarifying questions, supports long-term memory that would serve as context for future tasks. “Manus represents a promising product experience for AI agents,” says Ang Li, cofounder and CEO of Simular, a startup based in Palo Alto, California, that’s building computer use agents, AI agents that control a virtual computer. “I believe Chinese startups have a huge advantage when it comes to designing consumer products, thanks to cutthroat domestic competition that leads to fast execution and greater attention to product details.” In the case of Manus, the competition is moving fast. Two of the most buzzy follow‑ups, Genspark and Flowith, for example, are already boasting benchmark scores that match or edge past Manus’s.  Genspark, led by former Baidu executives Eric Jing and Kay Zhu, links many small “super agents” through what it calls multi‑component prompting. The agent can switch among several large language models, accepts both images and text, and carries out tasks from making slide decks to placing phone calls. Whereas Manus relies heavily on Browser Use, a popular open-source product that lets agents operate a web browser in a virtual window like a human, Genspark directly integrates with a wide array of tools and APIs. Launched in April, the company says that it already has over 5 million users and over million in yearly revenue. Flowith, the work of a young team that first grabbed public attention in April 2025 at a developer event hosted by the popular social media app Xiaohongshu, takes a different tack. Marketed as an “infinite agent,” it opens on a blank canvas where each question becomes a node on a branching map. Users can backtrack, take new branches, and store results in personal or sharable “knowledge gardens”—a design that feels more like project management softwarethan a typical chat interface. Every inquiry or task builds its own mind-map-like graph, encouraging a more nonlinear and creative interaction with AI. Flowith’s core agent, NEO, runs in the cloud and can perform scheduled tasks like sending emails and compiling files. The founders want the app to be a “knowledge marketbase”, and aims to tap into the social aspect of AI with the aspiration of becoming “the OnlyFans of AI knowledge creators”. What they also share with Manus is the global ambition. Both Genspark and Flowith have stated that their primary focus is the international market. A global address Startups like Manus, Genspark, and Flowith—though founded by Chinese entrepreneurs—could blend seamlessly into the global tech scene and compete effectively abroad. Founders, investors, and analysts that MIT Technology Review has spoken to believe Chinese companies are moving fast, executing well, and quickly coming up with new products.  Money reinforces the pull to launch overseas. Customers there pay more, and there are plenty to go around. “You can price in USD, and with the exchange rate that’s a sevenfold multiplier,” Manus cofounder Xiao Hong quipped on a podcast. “Even if we’re only operating at 10% power because of cultural differences overseas, we’ll still make more than in China.” But creating the same functionality in China is a challenge. Major US AI companies including OpenAI and Anthropic have opted out of mainland China because of geopolitical risks and challenges with regulatory compliance. Their absence initially created a black market as users resorted to VPNs and third-party mirrors to access tools like ChatGPT and Claude. That vacuum has since been filled by a new wave of Chinese chatbots—DeepSeek, Doubao, Kimi—but the appetite for foreign models hasn’t gone away.  Manus, for example, uses Anthropic’s Claude Sonnet—widely considered the top model for agentic tasks. Manus cofounder Zhang Tao has repeatedly praised Claude’s ability to juggle tools, remember contexts, and hold multi‑round conversations—all crucial for turning chatty software into an effective executive assistant. But the company’s use of Sonnet has made its agent functionally unusable inside China without a VPN. If you open Manus from a mainland IP address, you’ll see a notice explaining that the team is “working on integrating Qwen’s model,” a special local version that is built on top of Alibaba’s open-source model.  An engineer overseeing ByteDance’s work on developing an agent, who spoke to MIT Technology Review anonymously to avoid sanction, said that the absence of Claude Sonnet models “limits everything we do in China.” DeepSeek’s open models, he added, still hallucinate too often and lack training on real‑world workflows. Developers we spoke with rank Alibaba’s Qwen series as the best domestic alternative, yet most say that switching to Qwen knocks performance down a notch. Jiaxin Pei, a postdoctoral researcher at Stanford’s Institute for Human‑Centered AI, thinks that gap will close: “Building agentic capabilities in base LLMs has become a key focus for many LLM builders, and once people realize the value of this, it will only be a matter of time.” For now, Manus is doubling down on audiences it can already serve. In a written response, the company said its “primary focus is overseas expansion,” noting that new offices in San Francisco, Singapore, and Tokyo have opened in the past month. A super‑app approach Although the concept of AI agents is still relatively new, the consumer-facing AI app market in China is already crowded with major tech players. DeepSeek remains the most widely used, while ByteDance’s Doubao and Moonshot’s Kimi have also become household names. However, most of these apps are still optimized for chat and entertainment rather than task execution. This gap in the local market has pushed China’s big tech firms to roll out their own user-facing agents, though early versions remain uneven in quality and rough around the edges.  ByteDance is testing Coze Space, an AI agent based on its own Doubao model family that lets users toggle between “plan” and “execute” modes, so they can either directly guide the agent’s actions or step back and watch it work autonomously. It connects up to 14 popular apps, including GitHub, Notion, and the company’s own Lark office suite. Early reviews say the tool can feel clunky and has a high failure rate, but it clearly aims to match what Manus offers. Meanwhile, Zhipu AI has released a free agent called AutoGLM Rumination, built on its proprietary ChatGLM models. Shanghai‑based Minimax has launched Minimax Agent. Both products look almost identical to Manus and demo basic tasks such as building a simple website, planning a trip, making a small Flash game, or running quick data analysis. Despite the limited usability of most general AI agents launched within China, big companies have plans to change that. During a May 15 earnings call, Tencent president Liu Zhiping teased an agent that would weave automation directly into China’s most ubiquitous app, WeChat.  Considered the original super-app, WeChat already handles messaging, mobile payments, news, and millions of mini‑programs that act like embedded apps. These programs give Tencent, its developer, access to data from millions of services that pervade everyday life in China, an advantage most competitors can only envy. Historically, China’s consumer internet has splintered into competing walled gardens—share a Taobao link in WeChat and it resolves as plaintext, not a preview card. Unlike the more interoperable Western internet, China’s tech giants have long resisted integration with one another, choosing to wage platform war at the expense of a seamless user experience. But the use of mini‑programs has given WeChat unprecedented reach across services that once resisted interoperability, from gym bookings to grocery orders. An agent able to roam that ecosystem could bypass the integration headaches dogging independent startups. Alibaba, the e-commerce giant behind the Qwen model series, has been a front-runner in China’s AI race but has been slower to release consumer-facing products. Even though Qwen was the most downloaded open-source model on Hugging Face in 2024, it didn’t power a dedicated chatbot app until early 2025. In March, Alibaba rebranded its cloud storage and search app Quark into an all-in-one AI search tool. By June, Quark had introduced DeepResearch—a new mode that marks its most agent-like effort to date.  ByteDance and Alibaba did not reply to MIT Technology Review’s request for comments. “Historically, Chinese tech products tend to pursue the all-in-one, super-app approach, and the latest Chinese AI agents reflect just that,” says Li of Simular, who previously worked at Google DeepMind on AI-enabled work automation. “In contrast, AI agents in the US are more focused on serving specific verticals.” Pei, the researcher at Stanford, says that existing tech giants could have a huge advantage in bringing the vision of general AI agents to life—especially those with built-in integration across services. “The customer-facing AI agent market is still very early, with tons of problems like authentication and liability,” he says. “But companies that already operate across a wide range of services have a natural advantage in deploying agents at scale.” #manus #has #kickstarted #agent #boom

It’s easy to look at President Donald Trump’s second term and conclude that the less power and reach the federal government has, the better. After all, a smaller government might provide Trump or someone like him with fewer opportunities to disrupt people’s lives, leaving America less vulnerable to the whims of an aspiring autocrat. Weaker law-enforcement agencies could lack the capacity to enforce draconian policies. The president would have less say in how universities like Columbia conduct their business if they weren’t so dependent on federal funding. And he would have fewer resources to fundamentally change the American way of life.Trump’s presidency has the potential to reshape an age-old debate between the left and the right: Is it better to have a big government or a small one? The left, which has long advocated for bigger government as a solution to society’s problems, might be inclined to think that in the age of Trump, a strong government may be too risky. Say the United States had a single-payer universal health care system, for example. As my colleague Kelsey Piper pointed out, the government would have a lot of power to decide what sorts of medical treatments should and shouldn’t be covered, and certain forms of care that the right doesn’t support — like abortion or transgender health — would likely get cut when they’re in power. That’s certainly a valid concern. But the dangers Trump poses do not ultimately make the case for a small or weak government because the principal problem with the Trump presidency is not that he or the federal government has too much power. It’s that there’s not enough oversight.Reducing the power of the government wouldn’t necessarily protect us. In fact, “making government smaller” is one of the ways that Trump might be consolidating power.First things first: What is “big government”?When Americans are polled about how they feel about “big government” programs — policies like universal health care, Social Security, welfare for the poor — the majority of people tend to support them. Nearly two-thirds of Americans believe the government should be responsible for ensuring everyone has health coverage. But when you ask Americans whether they support “big government” in the abstract, a solid majority say they view it as a threat.That might sound like a story of contradictions. But it also makes sense because “big government” can have many different meanings. It can be a police state that surveils its citizens, an expansive regulatory state that establishes and enforces rules for the private sector, a social welfare state that directly provides a decent standard of living for everyone, or some combination of the three. In the United States, the debate over “big government” can also include arguments about federalism, or how much power the federal government should have over states. All these distinctions complicate the debate over the size of government: Because while someone might support a robust welfare system, they might simultaneously be opposed to being governed by a surveillance state or having the federal government involved in state and local affairs.As much as Americans like to fantasize about small government, the reality is that the wealthiest economies in the world have all been a product of big government, and the United States is no exception. That form of government includes providing a baseline social safety net, funding basic services, and regulating commerce. It also includes a government that has the capacity to enforce its rules and regulations.A robust state that caters to the needs of its people, that is able to respond quickly in times of crisis, is essential. Take the Covid-19 pandemic. The US government, under both the Trump and Biden administrations, was able to inject trillions of dollars into the economy to avert a sustained economic downturn. As a result, people were able to withstand the economic shocks, and poverty actually declined. Stripping the state of the basic powers it needs to improve the lives of its citizens will only make it less effective and erode people’s faith in it as a central institution, making people less likely to participate in the democratic process, comply with government policies, or even accept election outcomes.A constrained government does not mean a small governmentBut what happens when the people in power have no respect for democracy? The argument for a weaker and smaller government often suggests that a smaller government would be more constrained in the harm it can cause, while big government is more unrestrained. In this case, the argument is that if the US had a smaller government, then Trump could not effectively use the power of the state — by, say, deploying federal law enforcement agencies or withholding federal funds — to deport thousands of immigrants, bully universities, and assault fundamental rights like the freedom of speech. But advocating for bigger government does not mean you believe in handing the state unlimited power to do as it pleases. Ultimately, the most important way to constrain government has less to do with its size and scope and more to do with its checks and balances. In fact, one of the biggest checks on Trump’s power so far has been the structure of the US government, not its size. Trump’s most dangerous examples of overreach — his attempts to conduct mass deportations, eliminate birthright citizenship, and revoke student visas and green cards based on political views — have been an example of how proper oversight has the potential to limit government overreach. To be sure, Trump’s policies have already upended people’s lives, chilled speech, and undermined the principle of due process. But while Trump has pushed through some of his agenda, he hasn’t been able to deliver at the scale he promised. But that’s not because the federal government lacks the capacity to do those things. It’s because we have three equal branches of government, and the judicial branch, for all of its shortcomings in the Trump era, is still doing its most basic job to keep the executive branch in check. Reforms should include more oversight, not shrinking governmentThe biggest lesson from Trump’s first term was that America’s system of checks and balances — rules and regulations, norms, and the separate branches of government — wasn’t strong enough. As it turned out, a lot of potential oversight mechanisms did not have enough teeth to meaningfully restrain the president from abusing his power. Trump incited an assault on the US Capitol in an effort to overturn the 2020 election, and Congress ultimately failed in its duty to convict him for his actions. Twice, impeachment was shown to be a useless tool to keep a president in check.But again that’s a problem of oversight, not of the size and power of government. Still, oversight mechanisms need to be baked into big government programs to insulate them from petty politics or volatile changes from one administration to the next. Take the example of the hypothetical single-payer universal health care system. Laws dictating which treatments should be covered should be designed to ensure that changes to them aren’t dictated by the president alone, but through some degree of consensus that involves regulatory boards, Congress, and the courts. Ultimately, social programs should have mechanisms that allow for change so that laws don’t become outdated, as they do now. And while it’s impossible to guarantee that those changes will always be good, the current system of employer-sponsored health insurance is hardly a stable alternative.By contrast, shrinking government in the way that Republicans often talk about only makes people more vulnerable. Bigger governments — and more bureaucracy — can also insulate public institutions from the whims of an erratic president. For instance, Trump has tried to shutter the Consumer Financial Protection Bureau, a regulatory agency that gets in the way of his and his allies’ business. This assault allows Trump to serve his own interests by pleasing his donors.In other words, Trump is currently trying to make government smaller — by shrinking or eliminating agencies that get in his way — to consolidate power. “Despite Donald Trump’s rhetoric about the size or inefficiency of government, what he has done is eradicate agencies that directly served people,” said Julie Margetta Morgan, president of the Century Foundation who served as an associate director at the CFPB. “He may use the language of ‘government inefficiency’ to accomplish his goals, but I think what we’re seeing is that the goals are in fact to open up more lanes for big businesses to run roughshod over the American people.” The problem for small-government advocates is that the alternative to big government is not just small government. It’s also big business because fewer services, rules, and regulations open up the door to privatization and monopolization. And while the government, however big, has to answer to the public, businesses are far less accountable. One example of how business can replace government programs is the Republicans’ effort to overhaul student loan programs in the latest reconciliation bill the House passed, which includes eliminating subsidized loans and limiting the amount of aid students receive. The idea is that if students can’t get enough federal loans to cover the cost of school, they’ll turn to private lenders instead. “It’s not only cutting Pell Grants and the affordability of student loan programs in order to fund tax cuts to the wealthy, but it’s also creating a gap whereare all too happy to come in,” Margetta Morgan said. “This is the small government alternative: It’s cutting back on programs that provided direct services for people — that made their lives better and more affordable — and replacing it with companies that will use that gap as an opportunity for extraction and, in some cases, for predatory services.”Even with flawed oversight, a bigger and more powerful government is still preferable because it can address people’s most basic needs, whereas small government and the privatization of public services often lead to worse outcomes.So while small government might sound like a nice alternative when would-be tyrants rise to power, the alternative to big government would only be more corrosive to democracy, consolidating power in the hands of even fewer people. And ultimately, there’s one big way for Trump to succeed at destroying democracy, and that’s not by expanding government but by eliminating the parts of government that get in his way.See More: #big #government #still #good #even

More enterprises have moved to pay-for-performance salary and promotion models that measure progress toward goals -- but how do you measure goals for a maintenance programmer who barrels through a request backlog but delivers marginal value for the business, or for a business analyst whose success is predicated on forging intangibles like trust and cooperation with users so things can get done? It’s an age-old question facing companies, now that 77% of them use some type of pay-for-performance model. What are some popular pay-for-performance use cases? A factory doing piece work that pays employees based upon the number of items they assemble. A call center that pays agents based on how many calls they complete per day. A bank teller who gets rewarded for how many customers they sign up for credit cards. An IT project team that gets a bonus for completing a major project ahead of schedule. The IT example differs from the others, because it depends on team and not individual execution, but there nevertheless is something tangible to measure. The other use cases are more clearcut -- although they don’t account for pieces in the plant that were poorly assembled in haste to make quota and had to be reworked, or a call center agent who pushes calls off to someone else so they can end their calls in six minutes or less, or the teller who signs up X number of customers for credit cards, although two-thirds of them never use the credit card they signed up for. Related:In short, there are flaws in pay-for-performance models just as there are in other types of compensation models that organizations use. So, what’s the best path for IT for CIOs who want to implement pay for performance? One approach is to measure pay for performance based upon four key elements: hard results, effort, skill, and communications. The mix of these elements will vary, depending on the type of position each IT staff member performs. Here are two examples of pay per performance by position: 1. Computer maintenance programmers and help desk specialists Historically, IT departments have used hard numbers like how many open requests a computer maintenance programmer has closed, or how many calls a help desk employee has solved. There is merit in using hard results, and hard results should be factored into performance reviews for these individuals -- but hard numbers don’t tell the whole story.  For example, how many times has a help desk agent gone the extra mile with a difficult user or software bug, taking the time to see the entire process through until it is thoroughly solved? lf the issue was of a global nature, did the Help Desk agent follow up by letting others who use the application know that a bug was fixed? For the maintenance programmer who has completed the most open requests, which of these requests really solved a major business pain point? For both help desk and maintenance programming employees, were the changes and fixes properly documented and communicated to everyone with a need to know? And did these employees demonstrate the skills needed to solve their issues? Related:It’s difficult to capture hard results on elements like effort, communication and skills, but one way to go about it is to survey user departments on individual levels of service and effectiveness. From there, it’s up to IT managers to determinate the “mix” of hard results, effort, communication and skills on which the employee will be evaluated, and to communicate upfront to the employee what the pay for performance assessment will be based on. 2. Business analysts and trainers Business analysts and trainers are difficult to quantify in pay for performance models because so much of their success depends upon other people. A business analyst can know everything there is to know about a particular business area and its systems, but if the analyst is working with unresponsive users, or lacks the soft skills needed to communicate with users, the pay for performance can’t be based upon the technology skillset alone.  Related:IT trainers face a somewhat different dilemma when it  comes to performance evaluation: they can produce the training that new staff members need before staff is deployed on key projects,  but if a project gets delayed and this causes trainees to lose the knowledge that they learned, there is little the trainer can do aside from offering a refresher course. Can pay for performance be used for positions like these? It’s a mixed answer. Yes, pay per performance can be used for trainers, based upon how many individuals the trainer trains and how many new courses the trainer obtains or develops. These are the hard results. However, since so much of training’s execution depends upon other people downstream, like project managers who must start projects on time so new skills aren’t lost,  managers of training should also consider pay for performance elements such as effort, skills and communication.  In sum, for both business analysts and trainers, there are hard results that can be factored into a pay for performance formula, but there is also a need to survey each position’s “customers” -- those individualswho utilized the business analyst’s or trainer’s skills and products to accomplish their respective objectives in projects and training. Were these user-customers satisfied?  Summary Remarks The value that IT employees contribute to overall IT and to the business at large is a combination of tangible and intangible results. Pay for performance models are well suited to gauge tangible outcomes, but they fall short when it comes to the intangibles that could be just as important. Many years ago, when Pat Riley was coaching the Los Angeles Lakers, an interviewer asked what type of metrics he used when he measured the effectiveness of individual players on the basketball court. Was it the number of points, rebounds, or assists? Riley said he used an “effort" index. For example, how many times did a player go up to get a rebound, even if he didn’t end up with the ball? Riley said the effort individual players exhibited mattered, because even if they didn’t get the rebound, they were creating situations so someone else on the team could. IT is similar. It’s why OKR International, a performance consultancy, stated “Intangibles often create or destroy value quietly -- until their impact is too big to ignore. In the long run, they are the unseen levers that determine whether strategy thrives or withers.”  What CIOs and IT leadership can do when they use pay for performance is to assure that hard results, effort, communications and skills are appropriately blended for each IT staff position, and its responsibilities and realities -- because you can’t attach a numerical measurement to everything -- but you can observe visible changes that begin to manifest when a business analyst turns around what has been a hostile relationship with a user department and you begin to get things done.  #pay #performance #how #you #measure

PERFORMANCE ART With technical improvements, Pokemon Scarlet & Violet are utterly transformed on Nintendo Switch 2 What a difference a frame rate makes. On more powerful hardware, Pokemon Scarlet & Violet are practically different games Article by Alex Donaldson Assistant Editor Published on June 4, 2025 I can scarcely think of a game as hampered by its performance as the original Nintendo Switch release of Pokemon Scarlet & Violet. I can think of loads of games that perform worse, of course - broken, shattered releases - but I struggle to think of a great game so thoroughly compromised just from how it runs. That was the log behind my 2022 review of the games, where I called the games a “super-effective new vision” for the series but bemoaned how it ran. The game became famous for stop-motion windmills and distant cliffsides that looked like they’d fallen out of a Nintendo 64 game. The truth is, Scarlet & Violet’s brilliant design and peppy attitude deserved better. Now, three years on from release, Scarlet & Violet is about to get better with a Switch 2 update that I’m going to go ahead and call a total I was invited by The Pokemon Company to take an early look at the patched Switch 2 version of the game - which is always a sign of confidence, given I was quite a noisy detractor of the original game’s performance. I’m honestly not sure what to say other than: wow, what a difference. It is absolute night and day stuff. To the sort of people who say that frame rate doesn’t really matter, I challenge them to play Scarlet & Violet on Switch 2 and then go back to the original. I dare you. As friend of the site Joe says in Serebii's video preview embedded below, it ain’t doable. Watch on YouTube With a crisp presentation at a higher resolution and with a frame rate that as far as I can tell sits at a rock solid 60fps for the vast majority of the time, this is a world apart from the stomach-turning rollercoaster highs and lows of the original release. Distant Pokemon and world elements are no longer slideshows - yes, those infamous windmills are fixed! This isn’t just about technical bragging rights. The difference in how this game now runs is profound enough that it changes the flow and feel of the game. Scarlet & Violet were by design the most footloose and expansive Pokemon games of all time, channeling the open world chops of everything from Skyrim to Breath of the Wild into a Pokemon setting and setup. The performance was a drag on that - if the frame rate tanks every time you whip the camera around to see a nearby approaching threat or take in a distant vista, you’re ripped right out of the game. By stabilising everything, the performance is utterly transformed. Aside from the nebulous concept of ‘game feel’ being improved by the technical advancements, there are also real boons in terms of gameplay. Wild Pokemon spawn in and swarm across the rolling fields and the like in greater numbers. The subtle delay that you’d perceive, that hitch when encountering a wild Pokemon, is eliminated. Menus that were sluggish are now snappier and more responsive. Pokemon Box sprite icons now spring to life instantly. These are small quality of life changes that add up to something greater. Perhaps most importantly battles are now less plodding in their pace, which was frequently obliterated by certain move animations could send the frame rate crashing. It has to be said, it’s not all perfect. The level of detail settings remain pretty aggressive - which means as you’re galloping along at a glorious 60 frames atop your trusty Poke-steed, flowers and other micro detail pop in around you. It’s not ideal. Also, to be honest, the game now being technically accomplished does help to expose the art style for what it is - which is in need of a bit of tightening, I feel. The addition of HDR does really help the colorful exuberance of Paldea to shine, though. In all it’s a triumph, anyway. This is the game Scarlet & Violet should’ve been. Moreover, it feels like the most technically accomplished main-line Pokemon game… possibly ever? Certainly of the 3D era. As with 120fps mouselook Metroid, playing 3D Pokemon at 60fps feels like you’re doing something illegal, frankly. But this is now the way to play these excellent games - and with good-performing Pokemon games now on the table, my excitement for this year’s Pokemon Legends Z-A has skyrocketed. #with #technical #improvements #pokemon #scarlet

For the past two years, it’s been a ghost in the machine of American tech. Between 2022 and today, a little-noticed tweak to the U.S. tax code has quietly rewired the financial logic of how American companies invest in research and development. Outside of CFO and accounting circles, almost no one knew it existed. “I work on these tax write-offs and still hadn’t heard about this,” a chief operating officer at a private-equity-backed tech company told Quartz. “It’s just been so weirdly silent.”AdvertisementStill, the delayed change to a decades-old tax provision — buried deep in the 2017 tax law — has contributed to the loss of hundreds of thousands of high-paying, white-collar jobs. That’s the picture that emerges from a review of corporate filings, public financial data, analysis of timelines, and interviews with industry insiders. One accountant, working in-house at a tech company, described it as a “niche issue with broad impact,” echoing sentiments from venture capital investors also interviewed for this article. Some spoke on condition of anonymity to discuss sensitive political matters.Since the start of 2023, more than half-a-million tech workers have been laid off, according to industry tallies. Headlines have blamed over-hiring during the pandemic and, more recently, AI. But beneath the surface was a hidden accelerant: a change to what’s known as Section 174 that helped gut in-house software and product development teams everywhere from tech giants such as Microsoftand Metato much smaller, private, direct-to-consumer and other internet-first companies.Now, as a bipartisan effort to repeal the Section 174 change moves through Congress, bigger questions are surfacing: How did a single line in the tax code help trigger a tsunami of mass layoffs? And why did no one see it coming? For almost 70 years, American companies could deduct 100% of qualified research and development spending in the year they incurred the costs. Salaries, software, contractor payments — if it contributed to creating or improving a product, it came off the top of a firm’s taxable income.AdvertisementThe deduction was guaranteed by Section 174 of the IRS Code of 1954, and under the provision, R&D flourished in the U.S.Microsoft was founded in 1975. Applelaunched its first computer in 1976. Googleincorporated in 1998. Facebook opened to the general public in 2006. All these companies, now among the most valuable in the world, developed their earliest products — programming tools, hardware, search engines — under a tax system that rewarded building now, not later.The subsequent rise of smartphones, cloud computing, and mobile apps also happened in an America where companies could immediately write off their investments in engineering, infrastructure, and experimentation. It was a baseline assumption — innovation and risk-taking subsidized by the tax code — that shaped how founders operated and how investors made decisions.In turn, tech companies largely built their products in the U.S. AdvertisementMicrosoft’s operating systems were coded in Washington state. Apple’s early hardware and software teams were in California. Google’s search engine was born at Stanford and scaled from Mountain View. Facebook’s entire social architecture was developed in Menlo Park. The deduction directly incentivized keeping R&D close to home, rewarding companies for investing in American workers, engineers, and infrastructure.That’s what makes the politics of Section 174 so revealing. For all the rhetoric about bringing jobs back and making things in America, the first Trump administration’s major tax bill arguably helped accomplish the opposite.When Congress passed the Tax Cuts and Jobs Act, the signature legislative achievement of President Donald Trump’s first term, it slashed the corporate tax rate from 35% to 21% — a massive revenue loss on paper for the federal government.To make the 2017 bill comply with Senate budget rules, lawmakers needed to offset the cost. So they added future tax hikes that wouldn’t kick in right away, wouldn’t provoke immediate backlash from businesses, and could, in theory, be quietly repealed later.AdvertisementThe delayed change to Section 174 — from immediate expensing of R&D to mandatory amortization, meaning that companies must spread the deduction out in smaller chunks over five or even 15-year periods — was that kind of provision. It didn’t start affecting the budget until 2022, but it helped the TCJA appear “deficit neutral” over the 10-year window used for legislative scoring.The delay wasn’t a technical necessity. It was a political tactic. Such moves are common in tax legislation. Phase-ins and delayed provisions let lawmakers game how the Congressional Budget Office— Congress’ nonpartisan analyst of how bills impact budgets and deficits — scores legislation, pushing costs or revenue losses outside official forecasting windows.And so, on schedule in 2022, the change to Section 174 went into effect. Companies filed their 2022 tax returns under the new rules in early 2023. And suddenly, R&D wasn’t a full, immediate write-off anymore. The tax benefits of salaries for engineers, product and project managers, data scientists, and even some user experience and marketing staff — all of which had previously reduced taxable income in year one — now had to be spread out over five- or 15-year periods. To understand the impact, imagine a personal tax code change that allowed you to deduct 100% of your biggest source of expenses, and that becoming a 20% deduction. For cash-strapped companies, especially those not yet profitable, the result was a painful tax bill just as venture funding dried up and interest rates soared.AdvertisementSalesforce office buildings in San Francisco.Photo: Jason Henry/BloombergIt’s no coincidence that Meta announced its “Year of Efficiency” immediately after the Section 174 change took effect. Ditto Microsoft laying off 10,000 employees in January 2023 despite strong earnings, or Google parent Alphabet cutting 12,000 jobs around the same time.Amazonalso laid off almost 30,000 people, with cuts focused not just on logistics but on Alexa and internal cloud tools — precisely the kinds of projects that would have once qualified as immediately deductible R&D. Salesforceeliminated 10% of its staff, or 8,000 people, including entire product teams.In public, companies blamed bloat and AI. But inside boardrooms, spreadsheets were telling a quieter story. And MD&A notes — management’s notes on the numbers — buried deep in 10-K filings recorded the change, too. R&D had become more expensive to carry. Headcount, the leading R&D expense across the tech industry, was the easiest thing to cut.AdvertisementIn its 2023 annual report, Meta described salaries as its single biggest R&D expense. Between the first and second years that the Section 174 change began affecting tax returns, Meta cut its total workforce by almost 25%. Over the same period, Microsoft reduced its global headcount by about 7%, with cuts concentrated in product-facing, engineering-heavy roles.Smaller companies without the fortress-like balance sheets of Big Tech have arguably been hit even harder. Twilioslashed 22% of its workforce in 2023 alone. Shopifycut almost 30% of staff in 2022 and 2023. Coinbasereduced headcount by 36% across a pair of brutal restructuring waves.Since going into effect, the provision has hit at the very heart of America’s economic growth engine: the tech sector.By market cap, tech giants dominate the S&P 500, with the “Magnificent 7” alone accounting for more than a third of the index’s total value. Workforce numbers tell a similar story, with tech employing millions of Americans directly and supporting the employment of tens of millions more. As measured by GDP, capital-T tech contributes about 10% of national output.AdvertisementIt’s not just that tech layoffs were large, it’s that they were massively disproportionate. Across the broader U.S. economy, job cuts hovered around in low single digits across most sectors. But in tech, entire divisions vanished, with a whopping 60% jump in layoffs between 2022 and 2023. Some cuts reflected real inefficiencies — a response to over-hiring during the zero-interest rate boom. At the same time, many of the roles eliminated were in R&D, product, and engineering, precisely the kind of functions that had once benefitted from generous tax treatment under Section 174.Throughout the 2010s, a broad swath of startups, direct-to-consumer brands, and internet-first firms — basically every company you recognize from Instagram or Facebook ads — built their growth models around a kind of engineered break-even.The tax code allowed them to spend aggressively on product and engineering, then write it all off as R&D, keeping their taxable income close to zero by design. It worked because taxable income and actual cash flow were often notGAAP accounting practices. Basically, as long as spending counted as R&D, companies could report losses to investors while owing almost nothing to the IRS.But the Section 174 change broke that model. Once those same expenses had to be spread out, or amortized, over multiple years, the tax shield vanished. Companies that were still burning cash suddenly looked profitable on paper, triggering real tax bills on imaginary gains.AdvertisementThe logic that once fueled a generation of digital-first growth collapsed overnight.So it wasn’t just tech experiencing effects. From 1954 until 2022, the U.S. tax code had encouraged businesses of all stripes to behave like tech companies. From retail to logistics, healthcare to media, if firms built internal tools, customized a software stack, or invested in business intelligence and data-driven product development, they could expense those costs. The write-off incentivized in-house builds and fast growth well outside the capital-T tech sector. This lines up with OECD research showing that immediate deductions foster innovation more than spread-out ones.And American companies ran with that logic. According to government data, U.S. businesses reported about billion in R&D expenditures in 2019 alone, and almost half of that came from industries outside traditional tech. The Bureau of Economic Analysis estimates that this sector, the broader digital economy, accounts for another 10% of GDP.Add that to core tech’s contribution, and the Section 174 shift has likely touched at least 20% of the U.S. economy.AdvertisementThe result? A tax policy aimed at raising short-term revenue effectively hid a time bomb inside the growth engines of thousands of companies. And when it detonated, it kneecapped the incentive for hiring American engineers or investing in American-made tech and digital products.It made building tech companies in America look irrational on a spreadsheet.A bipartisan group of lawmakers is pushing to repeal the Section 174 change, with business groups, CFOs, crypto executives, and venture capitalists lobbying hard for retroactive relief. But the politics are messy. Fixing 174 would mean handing a tax break to the same companies many voters in both parties see as symbols of corporate excess. Any repeal would also come too late for the hundreds of thousands of workers already laid off.And of course, the losses don’t stop at Meta’s or Google’s campus gates. They ripple out. When high-paid tech workers disappear, so do the lunch orders. The house tours. The contract gigs. The spending habits that sustain entire urban economies and thousands of other jobs. Sandwich artists. Rideshare drivers. Realtors. Personal trainers. House cleaners. In tech-heavy cities, the fallout runs deep — and it’s still unfolding.AdvertisementWashington is now poised to pass a second Trump tax bill — one packed with more obscure provisions, more delayed impacts, more quiet redistribution. And it comes as analysts are only just beginning to understand the real-world effects of the last round.The Section 174 change “significantly increased the tax burden on companies investing in innovation, potentially stifling economic growth and reducing the United States’ competitiveness on the global stage,” according to the tax consulting firm KBKG. Whether the U.S. will reverse course — or simply adapt to a new normal — remains to be seen. #hidden #time #bomb #tax #code

More enterprises have moved to pay-for-performance salary and promotion models that measure progress toward goals -- but how do you measure goals for a maintenance programmer who barrels through a request backlog but delivers marginal value for the business, or for a business analyst whose success is predicated on forging intangibles like trust and cooperation with users so things can get done? It’s an age-old question facing companies, now that 77% of them use some type of pay-for-performance model. What are some popular pay-for-performance use cases? A factory doing piece work that pays employees based upon the number of items they assemble. A call center that pays agents based on how many calls they complete per day. A bank teller who gets rewarded for how many customers they sign up for credit cards. An IT project team that gets a bonus for completing a major project ahead of schedule. The IT example differs from the others, because it depends on team and not individual execution, but there nevertheless is something tangible to measure. The other use cases are more clearcut -- although they don’t account for pieces in the plant that were poorly assembled in haste to make quota and had to be reworked, or a call center agent who pushes calls off to someone else so they can end their calls in six minutes or less, or the teller who signs up X number of customers for credit cards, although two-thirds of them never use the credit card they signed up for. Related:In short, there are flaws in pay-for-performance models just as there are in other types of compensation models that organizations use. So, what’s the best path for IT for CIOs who want to implement pay for performance? One approach is to measure pay for performance based upon four key elements: hard results, effort, skill, and communications. The mix of these elements will vary, depending on the type of position each IT staff member performs. Here are two examples of pay per performance by position: 1. Computer maintenance programmers and help desk specialists Historically, IT departments have used hard numbers like how many open requests a computer maintenance programmer has closed, or how many calls a help desk employee has solved. There is merit in using hard results, and hard results should be factored into performance reviews for these individuals -- but hard numbers don’t tell the whole story.  For example, how many times has a help desk agent gone the extra mile with a difficult user or software bug, taking the time to see the entire process through until it is thoroughly solved? lf the issue was of a global nature, did the Help Desk agent follow up by letting others who use the application know that a bug was fixed? For the maintenance programmer who has completed the most open requests, which of these requests really solved a major business pain point? For both help desk and maintenance programming employees, were the changes and fixes properly documented and communicated to everyone with a need to know? And did these employees demonstrate the skills needed to solve their issues? Related:It’s difficult to capture hard results on elements like effort, communication and skills, but one way to go about it is to survey user departments on individual levels of service and effectiveness. From there, it’s up to IT managers to determinate the “mix” of hard results, effort, communication and skills on which the employee will be evaluated, and to communicate upfront to the employee what the pay for performance assessment will be based on. 2. Business analysts and trainers Business analysts and trainers are difficult to quantify in pay for performance models because so much of their success depends upon other people. A business analyst can know everything there is to know about a particular business area and its systems, but if the analyst is working with unresponsive users, or lacks the soft skills needed to communicate with users, the pay for performance can’t be based upon the technology skillset alone.  Related:IT trainers face a somewhat different dilemma when it  comes to performance evaluation: they can produce the training that new staff members need before staff is deployed on key projects,  but if a project gets delayed and this causes trainees to lose the knowledge that they learned, there is little the trainer can do aside from offering a refresher course. Can pay for performance be used for positions like these? It’s a mixed answer. Yes, pay per performance can be used for trainers, based upon how many individuals the trainer trains and how many new courses the trainer obtains or develops. These are the hard results. However, since so much of training’s execution depends upon other people downstream, like project managers who must start projects on time so new skills aren’t lost,  managers of training should also consider pay for performance elements such as effort, skills and communication.  In sum, for both business analysts and trainers, there are hard results that can be factored into a pay for performance formula, but there is also a need to survey each position’s “customers” -- those individualswho utilized the business analyst’s or trainer’s skills and products to accomplish their respective objectives in projects and training. Were these user-customers satisfied?  Summary Remarks The value that IT employees contribute to overall IT and to the business at large is a combination of tangible and intangible results. Pay for performance models are well suited to gauge tangible outcomes, but they fall short when it comes to the intangibles that could be just as important. Many years ago, when Pat Riley was coaching the Los Angeles Lakers, an interviewer asked what type of metrics he used when he measured the effectiveness of individual players on the basketball court. Was it the number of points, rebounds, or assists? Riley said he used an “effort" index. For example, how many times did a player go up to get a rebound, even if he didn’t end up with the ball? Riley said the effort individual players exhibited mattered, because even if they didn’t get the rebound, they were creating situations so someone else on the team could. IT is similar. It’s why OKR International, a performance consultancy, stated “Intangibles often create or destroy value quietly -- until their impact is too big to ignore. In the long run, they are the unseen levers that determine whether strategy thrives or withers.”  What CIOs and IT leadership can do when they use pay for performance is to assure that hard results, effort, communications and skills are appropriately blended for each IT staff position, and its responsibilities and realities -- because you can’t attach a numerical measurement to everything -- but you can observe visible changes that begin to manifest when a business analyst turns around what has been a hostile relationship with a user department and you begin to get things done.  #pay #performance #how #you #measure

After detailing the VFX work on The Last of Us Season 1 in 2023, Alex Wang returns to reflect on how the scope and complexity have evolved in Season 2. With close to 30 years of experience in the visual effects industry, Fiona Campbell Westgate has contributed to major productions such as Ghost in the Shell, Avatar: The Way of Water, Ant-Man and the Wasp: Quantumania, and Nyad. Her work on Nyad earned her a VES Award for Outstanding Supporting Visual Effects in a Photoreal Feature. Collaboration with Craig Mazin and Neil Druckmann is key to shaping the visual universe of The Last of Us. Can you share with us how you work with them and how they influence the visual direction of the series? Alex Wang // Craig visualizes the shot or scene before putting words on the page. His writing is always exceptionally detailed and descriptive, ultimately helping us to imagine the shot. Of course, no one understands The Last of Us better than Neil, who knows all aspects of the lore very well. He’s done much research and design work with the Naughty Dog team, so he gives us good guidance regarding creature and environment designs. I always try to begin with concept art to get the ball rolling with Craig and Neil’s ideas. This season, we collaborated with Chromatic Studios for concept art. They also contributed to the games, so I felt that continuity was beneficial for our show. Fiona Campbell Westgate // From the outset, it was clear that collaborating with Craig would be an exceptional experience. Early meetings revealed just how personable and invested Craig is. He works closely with every department to ensure that each episode is done to the highest level. Craig places unwavering trust in our VFX Supervisor, Alex Wang. They have an understanding between them that lends to an exceptional partnership. As the VFX Producer, I know how vital the dynamic between the Showrunner and VFX Supervisor is; working with these two has made for one of the best professional experiences of my career.  Photograph by Liane Hentscher/HBO How has your collaboration with Craig evolved between the first and second seasons? Were there any adjustments in the visual approach or narrative techniques you made this season? Alex Wang // Since everything was new in Season 1, we dedicated a lot of time and effort to exploring the show’s visual language, and we all learned a great deal about what worked and what didn’t for the show. In my initial conversations with Craig about Season 2, it was clear that he wanted to expand the show’s scope by utilizing what we established and learned in Season 1. He felt significantly more at ease fully committing to using VFX to help tell the story this season. The first season involved multiple VFX studios to handle the complexity of the effects. How did you divide the work among different studios for the second season? Alex Wang // Most of the vendors this season were also in Season 1, so we already had a shorthand. The VFX Producer, Fiona Campbell Westgate, and I work closely together to decide how to divide the work among our vendors. The type of work needs to be well-suited for the vendor and fit into our budget and schedule. We were extremely fortunate to have the vendors we did this season. I want to take this opportunity to thank Weta FX, DNEG, RISE, Distillery VFX, Storm Studios, Important Looking Pirates, Blackbird, Wylie Co., RVX, and VDK. We also had ILM for concept art and Digital Domain for previs. Fiona Campbell Westgate // Alex Wang and I were very aware of the tight delivery schedule, which added to the challenge of distributing the workload. We planned the work based on the individual studio’s capabilities, and tried not to burden them with back to back episodes wherever possible. Fortunately, there was shorthand with vendors from Season One, who were well-acquainted with the process and the quality of work the show required. The town of Jackson is a key location in The Last of Us. Could you explain how you approached creating and expanding this environment for the second season? Alex Wang // Since Season 1, this show has created incredible sets. However, the Jackson town set build is by far the most impressive in terms of scope. They constructed an 822 ft x 400 ft set in Minaty Bay that resembled a real town! I had early discussions with Production Designer Don MacAulay and his team about where they should concentrate their efforts and where VFX would make the most sense to take over. They focused on developing the town’s main street, where we believed most scenes would occur. There is a big reveal of Jackson in the first episode after Ellie comes out of the barn. Distillery VFX was responsible for the town’s extension, which appears seamless because the team took great pride in researching and ensuring the architecture aligned with the set while staying true to the tone of Jackson, Wyoming. Fiona Campbell Westgate // An impressive set was constructed in Minaty Bay, which served as the foundation for VFX to build upon. There is a beautiful establishing shot of Jackson in Episode 1 that was completed by Distillery, showing a safe and almost normal setting as Season Two starts. Across the episodes, Jackson set extensions were completed by our partners at RISE and Weta. Each had a different phase of Jackson to create, from almost idyllic to a town immersed in Battle.  What challenges did you face filming Jackson on both real and virtual sets? Was there a particular fusion between visual effects and live-action shots to make it feel realistic? Alex Wang // I always advocate for building exterior sets outdoors to take advantage of natural light. However, the drawback is that we cannot control the weather and lighting when filming over several days across two units. In Episode 2, there’s supposed to be a winter storm in Jackson, so maintaining consistency within the episode was essential. On sunny and rainy days, we used cranes to lift large 30x60ft screens to block the sun or rain. It was impossible to shield the entire set from the rain or sun, so we prioritized protecting the actors from sunlight or rain. Thus, you can imagine there was extensive weather cleanup for the episode to ensure consistency within the sequences. Fiona Campbell Westgate // We were fortunate that production built a large scale Jackson set. It provided a base for the full CG Jackson aerial shots and CG Set Extensions. The weather conditions at Minaty Bay presented a challenge during the filming of the end of the Battle sequence in Episode 2. While there were periods of bright sunshine, rainfall occurred during the filming of the end of the Battle sequence in Episode 2. In addition to the obvious visual effects work, it became necessary to replace the ground cover. Photograph by Liane Hentscher/HBO The attack on Jackson by the horde of infected in season 2 is a very intense moment. How did you approach the visual effects for this sequence? What techniques did you use to make the scale of the attack feel as impressive as it did? Alex Wang // We knew this would be a very complex sequence to shoot, and for it to be successful, we needed to start planning with the HODs from the very beginning. We began previs during prep with Weta FX and the episode’s director, Mark Mylod. The previs helped us understand Mark and the showrunner’s vision. This then served as a blueprint for all departments to follow, and in many instances, we filmed the previs. Fiona Campbell Westgate // The sheer size of the CG Infected Horde sets the tone for the scale of the Battle. It’s an intimidating moment when they are revealed through the blowing snow. The addition of CG explosions and atmospheric effects contributed in adding scale to the sequence.  Can you give us an insight into the technical challenges of capturing the infected horde? How much of the effect was done using CGI, and how much was achieved with practical effects? Alex Wang // Starting with a detailed previs that Mark and Craig approved was essential for planning the horde. We understood that we would never have enough stunt performers to fill a horde, nor could they carry out some stunts that would be too dangerous. I reviewed the previs with Stunt Coordinator Marny Eng numerous times to decide the best placements for her team’s stunt performers. We also collaborated with Barrie Gower from the Prosthetics team to determine the most effective allocation of his team’s efforts. Stunt performers positioned closest to the camera would receive the full prosthetic treatment, which can take hours. Weta FX was responsible for the incredible CG Infected horde work in the Jackson Battle. They have been a creative partner with HBO’s The Last of Us since Season 1, so they were brought on early for Season 2. I began discussions with Weta’s VFX supervisor, Nick Epstein, about how we could tackle these complex horde shots very early during the shoot. Typically, repetition in CG crowd scenes can be acceptable, such as armies with soldiers dressed in the same uniform or armour. However, for our Infected horde, Craig wanted to convey that the Infected didn’t come off an assembly line or all shop at the same clothing department store. Any repetition would feel artificial. These Infected were once civilians with families, or they were groups of raiders. We needed complex variations in height, body size, age, clothing, and hair. We built our base library of Infected, and then Nick and the Weta FX team developed a “mix and match” system, allowing the Infected to wear any costume and hair groom. A procedural texturing system was also developed for costumes, providing even greater variation. The most crucial aspect of the Infected horde was their motion. We had numerous shots cutting back-to-back with practical Infected, as well as shots where our CG Infected ran right alongside a stunt horde. It was incredibly unforgiving! Weta FX’s animation supervisor from Season 1, Dennis Yoo, returned for Season 2 to meet the challenge. Having been part of the first season, Dennis understood the expectations of Craig and Neil. Similar to issues of model repetition within a horde, it was relatively easy to perceive repetition, especially if they were running toward the same target. It was essential to enhance the details of their performances with nuances such as tripping and falling, getting back up, and trampling over each other. There also needed to be a difference in the Infected’s running speed. To ensure we had enough complexity within the horde, Dennis motion-captured almost 600 unique motion cycles. We had over a hundred shots in episode 2 that required CG Infected horde. Fiona Campbell Westgate // Nick Epstein, Weta VFX Supervisor, and Dennis Yoo, Weta Animation Supervisor, were faced with having to add hero, close-up Horde that had to integrate with practical Stunt performers. They achieved this through over 60 motion capture sessions and running it through a deformation system they developed. Every detail was applied to allow for a seamless blend with our practical Stunt performances. The Weta team created a custom costume and hair system that provided individual looks to the CG Infected Horde. We were able to avoid the repetitive look of a CG crowd due to these efforts. The movement of the infected horde is crucial for the intensity of the scene. How did you manage the animation and simulation of the infected to ensure smooth and realistic interaction with the environment? Fiona Campbell Westgate // We worked closely with the Stunt department to plan out positioning and where VFX would be adding the CG Horde. Craig Mazin wanted the Infected Horde to move in a way that humans cannot. The deformation system kept the body shape anatomically correct and allowed us to push the limits from how a human physically moves.  The Bloater makes a terrifying return this season. What were the key challenges in designing and animating this creature? How did you work on the Bloater’s interaction with the environment and other characters? Alex Wang // In Season 1, the Kansas City cul-de-sac sequence featured only a handful of Bloater shots. This season, however, nearly forty shots showcase the Bloater in broad daylight during the Battle of Jackson. We needed to redesign the Bloater asset to ensure it looked good in close-up shots from head to toe. Weta FX designed the Bloater for Season 1 and revamped the design for this season. Starting with the Bloater’s silhouette, it had to appear large, intimidating, and menacing. We explored enlarging the cordyceps head shape to make it feel almost like a crown, enhancing the Bloater’s impressive and strong presence. During filming, a stunt double stood in for the Bloater. This was mainly for scale reference and composition. It also helped the Infected stunt performers understand the Bloater’s spatial position, allowing them to avoid running through his space. Once we had an edit, Dennis mocapped the Bloater’s performances with his team. It is always challenging to get the motion right for a creature that weighs 600 pounds. We don’t want the mocap to be overly exaggerated, but it does break the character if the Bloater feels too “light.” The brilliant animation team at Weta FX brought the Bloater character to life and nailed it! When Tommy goes head-to-head with the Bloater, Craig was quite specific during the prep days about how the Bloater would bubble, melt, and burn as Tommy torches him with the flamethrower. Important Looking Pirates took on the “Burning Bloater” sequence, led by VFX Supervisor Philip Engstrom. They began with extensive R&D to ensure the Bloater’s skin would start to bubble and burn. ILP took the final Bloater asset from Weta FX and had to resculpt and texture the asset for the Bloater’s final burn state. Craig felt it was important for the Bloater to appear maimed at the end. The layers of FX were so complex that the R&D continued almost to the end of the delivery schedule. Fiona Campbell Westgate // This season the Bloater had to be bigger, more intimidating. The CG Asset was recreated to withstand the scrutiny of close ups and in daylight. Both Craig Mazin and Neil Druckmann worked closely with us during the process of the build. We referenced the game and applied elements of that version with ours. You’ll notice that his head is in the shape of crown, this is to convey he’s a powerful force.  During the Burning Bloater sequence in Episode 2, we brainstormed with Philip Engström, ILP VFX Supervisor, on how this creature would react to the flamethrower and how it would affect the ground as it burns. When the Bloater finally falls to the ground and dies, the extraordinary detail of the embers burning, fluid draining and melting the surrounding snow really sells that the CG creature was in the terrain.  Given the Bloater’s imposing size, how did you approach its integration into scenes with the actors? What techniques did you use to create such a realistic and menacing appearance? Fiona Campbell Westgate // For the Bloater, a stunt performer wearing a motion capture suit was filmed on set. This provided interaction with the actors and the environment. VFX enhanced the intensity of his movements, incorporating simulations to the CG Bloater’s skin and muscles that would reflect the weight and force as this terrifying creature moves.  Seattle in The Last of Us is a completely devastated city. Can you talk about how you recreated this destruction? What were the most difficult visual aspects to realize for this post-apocalyptic city? Fiona Campbell Westgate // We were meticulous in blending the CG destruction with the practical environment. The flora’s ability to overtake the environment had to be believable, and we adhered to the principle of form follows function. Due to the vastness of the CG devastation it was crucial to avoid repetitive effects. Consequently, our vendors were tasked with creating bespoke designs that evoked a sense of awe and beauty. Was Seattle’s architecture a key element in how you designed the visual effects? How did you adapt the city’s real-life urban landscape to meet the needs of the story while maintaining a coherent aesthetic? Alex Wang // It’s always important to Craig and Neil that we remain true to the cities our characters are in. DNEG was one of our primary vendors for Boston in Season 1, so it was natural for them to return for Season 2, this time focusing on Seattle. DNEG’s VFX Supervisor, Stephen James, who played a crucial role in developing the visual language of Boston for Season 1, also returns for this season. Stephen and Melaina Maceled a team to Seattle to shoot plates and perform lidar scans of parts of the city. We identified the buildings unique to Seattle that would have existed in 2003, so we ensured these buildings were always included in our establishing shots. Overgrowth and destruction have significantly influenced the environments in The Last of Us. The environment functions almost as a character in both Season 1 and Season 2. In the last season, the building destruction in Boston was primarily caused by military bombings. During this season, destruction mainly arises from dilapidation. Living in the Pacific Northwest, I understand how damp it can get for most of the year. I imagined that, over 20 years, the integrity of the buildings would be compromised by natural forces. This abundant moisture creates an exceptionally lush and vibrant landscape for much of the year. Therefore, when designing Seattle, we ensured that the destruction and overgrowth appeared intentional and aesthetically distinct from those of Boston. Fiona Campbell Westgate // Led by Stephen James, DNEG VFX Supervisor, and Melaina Mace, DNEG DFX Supervisor, the team captured photography, drone footage and the Clear Angle team captured LiDAR data over a three-day period in Seattle. It was crucial to include recognizable Seattle landmarks that would resonate with people familiar with the game.  The devastated city almost becomes a character in itself this season. What aspects of the visual effects did you have to enhance to increase the immersion of the viewer into this hostile and deteriorated environment? Fiona Campbell Westgate // It is indeed a character. Craig wanted it to be deteriorated but to have moments where it’s also beautiful in its devastation. For instance, in the Music Store in Episode 4 where Ellie is playing guitar for Dina, the deteriorated interior provides a beautiful backdrop to this intimate moment. The Set Decorating team dressed a specific section of the set, while VFX extended the destruction and overgrowth to encompass the entire environment, immersing the viewer in strange yet familiar surroundings. Photograph by Liane Hentscher/HBO The sequence where Ellie navigates a boat through a violent storm is stunning. What were the key challenges in creating this scene, especially with water simulation and the storm’s effects? Alex Wang // In the concluding episode of Season 2, Ellie is deep in Seattle, searching for Abby. The episode draws us closer to the Aquarium, where this area of Seattle is heavily flooded. Naturally, this brings challenges with CG water. In the scene where Ellie encounters Isaac and the W.L.F soldiers by the dock, we had a complex shoot involving multiple locations, including a water tank and a boat gimbal. There were also several full CG shots. For Isaac’s riverine boat, which was in a stormy ocean, I felt it was essential that the boat and the actors were given the appropriate motion. Weta FX assisted with tech-vis for all the boat gimbal work. We began with different ocean wave sizes caused by the storm, and once the filmmakers selected one, the boat’s motion in the tech-vis fed the special FX gimbal. When Ellie gets into the Jon boat, I didn’t want it on the same gimbal because I felt it would be too mechanical. Ellie’s weight needed to affect the boat as she got in, and that wouldn’t have happened with a mechanical gimbal. So, we opted to have her boat in a water tank for this scene. Special FX had wave makers that provided the boat with the appropriate movement. Instead of guessing what the ocean sim for the riverine boat should be, the tech- vis data enabled DNEG to get a head start on the water simulations in post-production. Craig wanted this sequence to appear convincingly dark, much like it looks out on the ocean at night. This allowed us to create dramatic visuals, using lightning strikes at moments to reveal depth. Were there any memorable moments or scenes from the series that you found particularly rewarding or challenging to work on from a visual effects standpoint? Alex Wang // The Last of Us tells the story of our characters’ journey. If you look at how season 2 begins in Jackson, it differs significantly from how we conclude the season in Seattle. We seldom return to the exact location in each episode, meaning every episode presents a unique challenge. The scope of work this season has been incredibly rewarding. We burned a Bloater, and we also introduced spores this season! Photograph by Liane Hentscher/HBO Looking back on the project, what aspects of the visual effects are you most proud of? Alex Wang // The Jackson Battle was incredibly complex, involving a grueling and lengthy shoot in quite challenging conditions, along with over 600 VFX shots in episode 2. It was truly inspiring to witness the determination of every department and vendor to give their all and create something remarkable. Fiona Campbell Westgate // I am immensely proud of the exceptional work accomplished by all of our vendors. During the VFX reviews, I found myself clapping with delight when the final shots were displayed; it was exciting to see remarkable results of the artists’ efforts come to light.  How long have you worked on this show? Alex Wang // I’ve been on this season for nearly two years. Fiona Campbell Westgate // A little over one year; I joined the show in April 2024. What’s the VFX shots count? Alex Wang // We had just over 2,500 shots this Season. Fiona Campbell Westgate // In Season 2, there were a total of 2656 visual effects shots. What is your next project? Fiona Campbell Westgate // Stay tuned… A big thanks for your time. WANT TO KNOW MORE?Blackbird: Dedicated page about The Last of Us – Season 2 website.DNEG: Dedicated page about The Last of Us – Season 2 on DNEG website.Important Looking Pirates: Dedicated page about The Last of Us – Season 2 website.RISE: Dedicated page about The Last of Us – Season 2 website.Weta FX: Dedicated page about The Last of Us – Season 2 website. © Vincent Frei – The Art of VFX – 2025 #last #season #alex #wang #production