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 $75 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 $36 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 software (think Notion) than 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.”

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 intelligence (AGI) is 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 because (presumably) those 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 (see the link here). 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 intelligence (AGI) or maybe even the outstretched possibility of achieving artificial superintelligence (ASI). 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 (all in), 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 (see my coverage at the link here). 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.

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 $75 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 $36 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 software (think Notion) than 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.”

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 $75 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 $36 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 software (think Notion) than 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.”

Every few weeks, it seems like there’s a new headline about a lawyer getting in trouble for submitting filings containing, in the words of one judge, “bogus AI-generated research.” The details vary, but the throughline is the same: an attorney turns to a large language model (LLM) like ChatGPT to help them with legal research (or worse, writing), the LLM hallucinates cases that don’t exist, and the lawyer is none the wiser until the judge or opposing counsel points out their mistake. In some cases, including an aviation lawsuit from 2023, attorneys have had to pay fines for submitting filings with AI-generated hallucinations. So why haven’t they stopped?The answer mostly comes down to time crunches, and the way AI has crept into nearly every profession. Legal research databases like LexisNexis and Westlaw have AI integrations now. For lawyers juggling big caseloads, AI can seem like an incredibly efficient assistant. Most lawyers aren’t necessarily using ChatGPT to write their filings, but they are increasingly using it and other LLMs for research. Yet many of these lawyers, like much of the public, don’t understand exactly what LLMs are or how they work. One attorney who was sanctioned in 2023 said he thought ChatGPT was a “super search engine.” It took submitting a filing with fake citations to reveal that it’s more like a random-phrase generator — one that could give you either correct information or convincingly phrased nonsense.Andrew Perlman, the dean of Suffolk University Law School, argues many lawyers are using AI tools without incident, and the ones who get caught with fake citations are outliers. “I think that what we’re seeing now — although these problems of hallucination are real, and lawyers have to take it very seriously and be careful about it — doesn’t mean that these tools don’t have enormous possible benefits and use cases for the delivery of legal services,” Perlman said. Legal databases and research systems like Westlaw are incorporating AI services.In fact, 63 percent of lawyers surveyed by Thomson Reuters in 2024 said they’ve used AI in the past, and 12 percent said they use it regularly. Respondents said they use AI to write summaries of case law and to research “case law, statutes, forms or sample language for orders.” The attorneys surveyed by Thomson Reuters see it as a time-saving tool, and half of those surveyed said “exploring the potential for implementing AI” at work is their highest priority. “The role of a good lawyer is as a ‘trusted advisor’ not as a producer of documents,” one respondent said. But as plenty of recent examples have shown, the documents produced by AI aren’t always accurate, and in some cases aren’t real at all.RelatedIn one recent high-profile case, lawyers for journalist Tim Burke, who was arrested for publishing unaired Fox News footage in 2024, submitted a motion to dismiss the case against him on First Amendment grounds. After discovering that the filing included “significant misrepresentations and misquotations of supposedly pertinent case law and history,” Judge Kathryn Kimball Mizelle, of Florida’s middle district, ordered the motion to be stricken from the case record. Mizelle found nine hallucinations in the document, according to the Tampa Bay Times.Mizelle ultimately let Burke’s lawyers, Mark Rasch and Michael Maddux, submit a new motion. In a separate filing explaining the mistakes, Rasch wrote that he “assumes sole and exclusive responsibility for these errors.” Rasch said he used the “deep research” feature on ChatGPT pro, which The Verge has previously tested with mixed results, as well as Westlaw’s AI feature.Rasch isn’t alone. Lawyers representing Anthropic recently admitted to using the company’s Claude AI to help write an expert witness declaration submitted as part of the copyright infringement lawsuit brought against Anthropic by music publishers. That filing included a citation with an “inaccurate title and inaccurate authors.” Last December, misinformation expert Jeff Hancock admitted he used ChatGPT to help organize citations in a declaration he submitted in support of a Minnesota law regulating deepfake use. Hancock’s filing included “two citation errors, popularly referred to as ‘hallucinations,’” and incorrectly listed authors for another citation. These documents do, in fact, matter — at least in the eyes of judges. In a recent case, a California judge presiding over a case against State Farm was initially swayed by arguments in a brief, only to find that the case law cited was completely made up. “I read their brief, was persuaded (or at least intrigued) by the authorities that they cited, and looked up the decisions to learn more about them – only to find that they didn’t exist,” Judge Michael Wilner wrote.Perlman said there are several less risky ways lawyers use generative AI in their work, including finding information in large tranches of discovery documents, reviewing briefs or filings, and brainstorming possible arguments or possible opposing views. “I think in almost every task, there are ways in which generative AI can be useful — not a substitute for lawyers’ judgment, not a substitute for the expertise that lawyers bring to the table, but in order to supplement what lawyers do and enable them to do their work better, faster, and cheaper,” Perlman said.But like anyone using AI tools, lawyers who rely on them to help with legal research and writing need to be careful to check the work they produce, Perlman said. Part of the problem is that attorneys often find themselves short on time — an issue he says existed before LLMs came into the picture. “Even before the emergence of generative AI, lawyers would file documents with citations that didn’t really address the issue that they claimed to be addressing,” Perlman said. “It was just a different kind of problem. Sometimes when lawyers are rushed, they insert citations, they don’t properly check them; they don’t really see if the case has been overturned or overruled.” (That said, the cases do at least typically exist.)Another, more insidious problem is the fact that attorneys — like others who use LLMs to help with research and writing — are too trusting of what AI produces. “I think many people are lulled into a sense of comfort with the output, because it appears at first glance to be so well crafted,” Perlman said.Alexander Kolodin, an election lawyer and Republican state representative in Arizona, said he treats ChatGPT as a junior-level associate. He’s also used ChatGPT to help write legislation. In 2024, he included AI text in part of a bill on deepfakes, having the LLM provide the “baseline definition” of what deepfakes are and then “I, the human, added in the protections for human rights, things like that it excludes comedy, satire, criticism, artistic expression, that kind of stuff,” Kolodin told The Guardian at the time. Kolodin said he “may have” discussed his use of ChatGPT with the bill’s main Democratic cosponsor but otherwise wanted it to be “an Easter egg” in the bill. The bill passed into law. Kolodin — who was sanctioned by the Arizona State Bar in 2020 for his involvement in lawsuits challenging the result of the 2020 election — has also used ChatGPT to write first drafts of amendments, and told The Verge he uses it for legal research as well. To avoid the hallucination problem, he said, he just checks the citations to make sure they’re real.“You don’t just typically send out a junior associate’s work product without checking the citations,” said Kolodin. “It’s not just machines that hallucinate; a junior associate could read the case wrong, it doesn’t really stand for the proposition cited anyway, whatever. You still have to cite-check it, but you have to do that with an associate anyway, unless they were pretty experienced.”Kolodin said he uses both ChatGPT’s pro “deep research” tool and the LexisNexis AI tool. Like Westlaw, LexisNexis is a legal research tool primarily used by attorneys. Kolodin said that in his experience, it has a higher hallucination rate than ChatGPT, which he says has “gone down substantially over the past year.” AI use among lawyers has become so prevalent that in 2024, the American Bar Association issued its first guidance on attorneys’ use of LLMs and other AI tools. Lawyers who use AI tools “have a duty of competence, including maintaining relevant technological competence, which requires an understanding of the evolving nature” of generative AI, the opinion reads. The guidance advises lawyers to “acquire a general understanding of the benefits and risks of the GAI tools” they use — or, in other words, to not assume that an LLM is a “super search engine.” Attorneys should also weigh the confidentiality risks of inputting information relating to their cases into LLMs and consider whether to tell their clients about their use of LLMs and other AI tools, it states.Perlman is bullish on lawyers’ use of AI. “I do think that generative AI is going to be the most impactful technology the legal profession has ever seen and that lawyers will be expected to use these tools in the future,” he said. “I think that at some point, we will stop worrying about the competence of lawyers who use these tools and start worrying about the competence of lawyers who don’t.”Others, including one of the judges who sanctioned lawyers for submitting a filing full of AI-generated hallucinations, are more skeptical. “Even with recent advances,” Wilner wrote, “no reasonably competent attorney should out-source research and writing to this technology — particularly without any attempt to verify the accuracy of that material.”See More:

Key Takeaways New EU rules take effect June 20, 2025, requiring mobile phones and tablets to be easier and cheaper to repair, including mandatory access to spare parts for up to seven years. Devices must meet durability standards, such as resistance to drops, water, and dust, and batteries must retain 80% capacity after 800 cycles. Manufacturers must offer 5 years of software updates and grant repair professionals access to firmware and repair tools to ensure long-term device usability. Energy consumption is expected to drop, with households projected to save €98 over five years due to longer-lasting, more repairable devices. In a win for the right to repair, the European Union (EU) has released its new energy labeling and ecodesign requirements for mobile phones, tablets, and even cordless phones.  Once implemented, this would help extend the product lifetime of these devices by making them cheaper and easier to repair. Up Ahead: Long-Lasting and Easy-to-Repair Devices The new ecodesign and energy labeling regulations will apply to devices sold in the EU effective June 20, 2025. One key area is the availability of spare parts. Once the regulations take effect, critical spare parts for a device must remain available for seven years from the last day the producer stops selling it. These parts must also be available to repairers within 5–10 working days. Mobile phones, cordless phones, and tablets under the ecodesign regulations must also be resistant to scratches and accidental drops, and be protected from dust and water.  The EU also specified that batteries must withstand at least 800 charge and discharge cycles while holding 80% of their initial capacity. These measures could help ensure devices last longer, thus reducing the need to replace them and lowering their cost of ownership. The requirements go beyond just the hardware. On the software side, manufacturers must also provide at least 5 years of operating system upgrades and give professional repairers access to software or firmware.  These would allow devices to continue to run smoothly and protect them from security vulnerabilities. On top of the ecodesign requirements, the EU also specified energy labeling requirements on the applicable devices. The labels would provide information about the device’s repairability score, energy efficiency, battery longevity, dust and water protection, and resistance to accidental drops. Having clear and easy-to-understand labels can help consumers in the EU make better choices when buying phones and tablets. They’ll know at a glance whether a device can withstand drops or if the batteries will remain reliable for as long as possible. The Projected Impact of the Ecodesign and Energy Labeling Requirements With the new requirements, the EU expects a significant reduction in the energy consumption of the applicable devices. Regarding charging time, this should go down from 2.5 to 1.75 hours per day.  Also, the annual electrical consumption of high-end smartphones is projected to decrease by 25%, while it’s 23% for tablets by 2030. Making devices easier to repair is also seen to reduce the acquisition costs of devices by € 98 per household in five years. A Win for the Right to Repair, But Obstacles Lie Ahead While the EU has already made significant progress in the area of the right to repair, many countries remain well behind. In the US, for instance, only a handful of states like California and Colorado have already enacted right-to-repair laws, while most of the country has yet to catch up. Without laws like these, most manufacturers won’t have the incentive to make their devices easier to repair.  A quick look at iFixit’s smartphone repairability scores shows that only Fairphone, which scored a perfect 10, HMD (9), and Nokia (8) had high scores among current phones. At the same time, Apple’s iPhone 16 line received a respectable 7. The Fairphone 5, launched in 2023, is especially notable for its repairability. According to iFixit, it’s easy to disassemble with common tools, the battery can be replaced quickly, and manuals and replacement parts are available to DIYers. As for the others, Ifixit gave Google’s Pixel 9 series a 5 for making it difficult to remove the battery and the USB-C port and requiring a replacement adhesive to secure the screen or back glass instead of using clips or fasteners. Meanwhile, the Samsung Galaxy S23 Ultra received a 4 because the battery is attached with a strong adhesive, and repairing the screen needs a lot of disassembly. All in all, we remain hopeful that with the EU’s lead, manufacturers can finally be encouraged to make repairing their devices easier than ever before. As technology continues to evolve—from the return of 'dumbphones' to faster and sleeker computers—seasoned tech journalist, Cedric Solidon, continues to dedicate himself to writing stories that inform, empower, and connect with readers across all levels of digital literacy. With 20 years of professional writing experience, this University of the Philippines Journalism graduate has carved out a niche as a trusted voice in tech media. Whether he's breaking down the latest advancements in cybersecurity or explaining how silicon-carbon batteries can extend your phone’s battery life, his writing remains rooted in clarity, curiosity, and utility. Long before he was writing for Techreport, HP, Citrix, SAP, Globe Telecom, CyberGhost VPN, and ExpressVPN, Cedric's love for technology began at home courtesy of a Nintendo Family Computer and a stack of tech magazines. Growing up, his days were often filled with sessions of Contra, Bomberman, Red Alert 2, and the criminally underrated Crusader: No Regret. But gaming wasn't his only gateway to tech.  He devoured every T3, PCMag, and PC Gamer issue he could get his hands on, often reading them cover to cover. It wasn’t long before he explored the early web in IRC chatrooms, online forums, and fledgling tech blogs, soaking in every byte of knowledge from the late '90s and early 2000s internet boom. That fascination with tech didn’t just stick. It evolved into a full-blown calling. After graduating with a degree in Journalism, he began his writing career at the dawn of Web 2.0. What started with small editorial roles and freelance gigs soon grew into a full-fledged career. He has since collaborated with global tech leaders, lending his voice to content that bridges technical expertise with everyday usability. He’s also written annual reports for Globe Telecom and consumer-friendly guides for VPN companies like CyberGhost and ExpressVPN, empowering readers to understand the importance of digital privacy. His versatility spans not just tech journalism but also technical writing. He once worked with a local tech company developing web and mobile apps for logistics firms, crafting documentation and communication materials that brought together user-friendliness with deep technical understanding. That experience sharpened his ability to break down dense, often jargon-heavy material into content that speaks clearly to both developers and decision-makers. At the heart of his work lies a simple belief: technology should feel empowering, not intimidating. Even if the likes of smartphones and AI are now commonplace, he understands that there's still a knowledge gap, especially when it comes to hardware or the real-world benefits of new tools. His writing hopes to help close that gap. Cedric’s writing style reflects that mission. It’s friendly without being fluffy and informative without being overwhelming. Whether writing for seasoned IT professionals or casual readers curious about the latest gadgets, he focuses on how a piece of technology can improve our lives, boost our productivity, or make our work more efficient. That human-first approach makes his content feel more like a conversation than a technical manual. As his writing career progresses, his passion for tech journalism remains as strong as ever. With the growing need for accessible, responsible tech communication, he sees his role not just as a journalist but as a guide who helps readers navigate a digital world that’s often as confusing as it is exciting. From reviewing the latest devices to unpacking global tech trends, Cedric isn’t just reporting on the future; he’s helping to write it. View all articles by Cedric Solidon Our editorial process The Tech Report editorial policy is centered on providing helpful, accurate content that offers real value to our readers. We only work with experienced writers who have specific knowledge in the topics they cover, including latest developments in technology, online privacy, cryptocurrencies, software, and more. Our editorial policy ensures that each topic is researched and curated by our in-house editors. We maintain rigorous journalistic standards, and every article is 100% written by real authors.

IT Pros ‘Extremely Worried’ About Shadow AI: Report By John P. Mello Jr. June 4, 2025 5:00 AM PT ADVERTISEMENT Enterprise IT Lead Generation Services Fuel Your Pipeline. Close More Deals. Our full-service marketing programs deliver sales-ready leads. 100% Satisfaction Guarantee! Learn more. Shadow AI — the use of AI tools under the radar of IT departments — has information technology directors and executives worried, according to a report released Tuesday. The report, based on a survey of 200 IT directors and executives at U.S. enterprise organizations of 1,000 employees or more, found nearly half the IT pros (46%) were “extremely worried” about shadow AI, and almost all of them (90%) were concerned about it from a privacy and security viewpoint. “As our survey found, shadow AI is resulting in palpable, concerning outcomes, with nearly 80% of IT leaders saying it has resulted in negative incidents such as sensitive data leakage to Gen AI tools, false or inaccurate results, and legal risks of using copyrighted information,” said Krishna Subramanian, co-founder of Campbell, Calif.-based Komprise, the unstructured data management company that produced the report. “Alarmingly, 13% say that shadow AI has caused financial or reputational harm to their organizations,” she told TechNewsWorld. Subramanian added that shadow AI poses a much greater problem than shadow IT, which primarily focuses on departmental power users purchasing cloud instances or SaaS tools without obtaining IT approval. “Now we’ve got an unlimited number of employees using tools like ChatGPT or Claude AI to get work done, but not understanding the potential risk they are putting their organizations at by inadvertently submitting company secrets or customer data into the chat prompt,” she explained. “The data risk is large and growing in still unforeseen ways because of the pace of AI development and adoption and the fact that there is a lot we don’t know about how AI works,” she continued. “It is becoming more humanistic all the time and capable of making decisions independently.” Shadow AI Introduces Security Blind Spots Shadow AI is the next step after shadow IT and is a growing risk, noted James McQuiggan, security awareness advocate at KnowBe4, a security awareness training provider in Clearwater, Fla. “Users use AI tools for content, images, or applications and to process sensitive data or company information without proper security checks,” he told TechNewsWorld. “Most organizations will have privacy, compliance, and data protection policies, and shadow AI introduces blind spots in the organization’s data loss prevention.” “The biggest risk with shadow AI is that the AI application has not passed through a security analysis as approved AI tools may have been,” explained Melissa Ruzzi, director of AI at AppOmni, a SaaS security management software company, in San Mateo, Calif. “Some AI applications may be training models using your data, may not adhere to relevant regulations that your company is required to follow, and may not even have the data storage security level you deem necessary to keep your data from being exposed,” she told TechNewsWorld. “Those risks are blind spots of potential security vulnerabilities in shadow AI.” Krishna Vishnubhotla, vice president of product strategy at Zimperium, a mobile security company based in Dallas, noted that shadow AI extends beyond unapproved applications and involves embedded AI components that can process and disseminate sensitive data in unpredictable ways. “Unlike traditional shadow IT, which may be limited to unauthorized software or hardware, shadow AI can run on employee mobile devices outside the organization’s perimeter and control,” he told TechNewsWorld. “This creates new security and compliance risks that are harder to track and mitigate.” Vishnubhotla added that the financial impact of shadow AI varies, but unauthorized AI tools can lead to significant regulatory fines, data breaches, and loss of intellectual property. “Depending on the scale of the agency and the sensitivity of the data exposed, the costs could range from millions to potentially billions in damages due to compliance violations, remediation efforts, and reputational harm,” he said. “Federal agencies handling vast amounts of sensitive or classified information, financial institutions, and health care organizations are particularly vulnerable,” he said. “These sectors collect and analyze vast amounts of high-value data, making AI tools attractive. But without proper vetting, these tools could be easily exploited.” Shadow AI Everywhere and Easy To Use Nicole Carignan, SVP for security and AI strategy at Darktrace, a global cybersecurity AI company, predicts an explosion of tools that utilize AI and generative AI within enterprises and on devices used by employees. “In addition to managing AI tools that are built in-house, security teams will see a surge in the volume of existing tools that have new AI features and capabilities embedded, as well as a rise in shadow AI,” she told TechNewsWorld. “If the surge remains unchecked, this raises serious questions and concerns about data loss prevention, as well as compliance concerns as new regulations start to take effect.” “That will drive an increasing need for AI asset discovery — the ability for companies to identify and track the use of AI systems throughout the enterprise,” she said. “It is imperative that CIOs and CISOs dig deep into new AI security solutions, asking comprehensive questions about data access and visibility.” Shadow AI has become so rampant because it is everywhere and easy to access through free tools, maintained Komprise’s Subramanian. “All you need is a web browser,” she said. “Enterprise users can inadvertently share company code snippets or corporate data when using these Gen AI tools, which could create data leakage.” “These tools are growing and changing exponentially,” she continued. “It’s really hard to keep up. As the IT leader, how do you track this and determine the risk? Managers might be looking the other way because their teams are getting more done. You may need fewer contractors and full-time employees. But I think the risk of the tools is not well understood.” “The low, or in some cases non-existent, learning curve associated with using Gen AI services has led to rapid adoption, regardless of prior experience with these services,” added Satyam Sinha, CEO and co-founder of Acuvity, a provider of runtime Gen AI security and governance solutions, in Sunnyvale, Calif. “Whereas shadow IT focused on addressing a specific challenge for particular employees or departments, shadow AI addresses multiple challenges for multiple employees and departments. Hence, the greater appeal,” he said. “The abundance and rapid development of Gen AI services also means employees can find the right solution [instantly]. Of course, all these traits have direct security implications.” Banning AI Tools Backfires To support innovation while minimizing the threat of shadow AI, enterprises must take a three-pronged approach, asserted Kris Bondi, CEO and co-founder of Mimoto, a threat detection and response company in San Francisco. They must educate employees on the dangers of unsupported, unmonitored AI tools, create company protocols for what is not acceptable use of unauthorized AI tools, and, most importantly, provide AI tools that are sanctioned. “Explaining why one tool is sanctioned and another isn’t greatly increases compliance,” she told TechNewsWorld. “It does not work for a company to have a zero-use mandate. In fact, this results in an increase in stealth use of shadow AI.” In the very near future, more and more applications will be leveraging AI in different forms, so the reality of shadow AI will be present more than ever, added AppOmni’s Ruzzi. “The best strategy here is employee training and AI usage monitoring,” she said. “It will become crucial to have in place a powerful SaaS security tool that can go beyond detecting direct AI usage of chatbots to detect AI usage connected to other applications,” she continued, “allowing for early discovery, proper risk assessment, and containment to minimize possible negative consequences.” “Shadow AI is just the beginning,” KnowBe4’s McQuiggan added. “As more teams use AI, the risks grow.” He recommended that companies start small, identify what’s being used, and build from there. They should also get legal, HR, and compliance involved. “Make AI governance part of your broader security program,” he said. “The sooner you start, the better you can manage what comes next.” John P. Mello Jr. has been an ECT News Network reporter since 2003. His areas of focus include cybersecurity, IT issues, privacy, e-commerce, social media, artificial intelligence, big data and consumer electronics. He has written and edited for numerous publications, including the Boston Business Journal, the Boston Phoenix, Megapixel.Net and Government Security News. Email John. Leave a Comment Click here to cancel reply. Please sign in to post or reply to a comment. New users create a free account. Related Stories More by John P. Mello Jr. view all More in IT Leadership

Modeling, Texturing & Rendering Techniques with Joshua HaunLearn how to use Blender and Unreal Engine 5 to craft a Victorian-style architectural game environment design using a professional workflow. Environment Artist Joshua Haun shares how to take design elements from Victorian history and bring them into the digital realm, allowing game players to immerse themselves in the game world. As graphics in games have become more realistic over the years, players are also more likely to notice when any aspects of a world feel “off.” This workshop aims to discuss the fundamentals of architectural design theory while explaining how to create art suitable for a real gaming production environment, implementing key techniques commonly used in architectural design. This workshop is geared toward Intermediate artists seeking to create environments with deeper levels of meaning. Beginners will also enjoy this workshop, provided they have a foundational level of modeling and UV-ing in Blender and basic knowledge of the Unreal Engine 5 interface.Modeling, texturing with trim sheets, and UV-ing architecture are covered in Blender, while a focus on shader creation, garden design with foliage, lighting, camera work, and shading are all topics showcased in UE5. The final result of the environment will be rendered in Unreal 5; therefore, the material shown in this workshop is primarily geared towards a real-time game engine. However, much of the explanation on design theory, history, and architecture will be universally applicable to all levels of art. Other software used in this workshop includes Rhinoceros 3D, Affinity Photo, Substance Painter, and Davinci Resolve.WATCH NOW

The drama between US president Donald Trump and his former buddy-in-chief Elon Musk is far from over.As ABC reported today, now that he's been summarily retired from the White House, the billionaire SpaceX boss has been privately venting his frustrations at Trump. One particularly stinging betrayal, per the network's reporting: Trump's sudden withdrawal of Musk's buddy and financial benefactor, Jared Isaacman, from consideration to be the next NASA administrator.As the day progressed, Musk's tension with Trump exploded into public view as history's richest man tweeted or amplified no less than 25 posts blasting Trump's "big, beautiful" tax and spending bill, which takes the form of yet another piece of legislation meant to gut assistance for the poorest Americans while siphoning money to the ultra-wealthy.However, that isn't Musk's issue with the package. Instead, his commentary is centered on the bill's impact on the US national deficit — something he tried and failed to curb in any meaningful way during his time as a pay-to-play government operative.On X-formerly-Twitter, Musk's frenzied posts range from Rand Paul interview clips to hysterical conspiracy peddling. ("America is in the fast lane to debt slavery," he fomented at one point.)"Call your Senator, call your Congressman, bankrupting America is NOT ok!" Musk urged his 220 million followers on X-formerly-Twitter. "KILL the BILL."The tech titan also went out of his way to amplify some low-res footage of Warren Buffett explaining his theoretical plan to reduce the deficit. "Anytime there's a deficit of more than 3 percent of GDP, all sitting members of Congress are ineligible for reelection," the investor suggested, to which Musk replied that "this is the way."Needless to say, a month ago — or even a week — this type of assault on Trump by Musk would have been unthinkable. The bill is also a baffling hill for the tech mogul to die on, especially considering that government spending is what made his tech dynasty possible in the first place. It's more plausible, as Axios notes, that national debt is a smokescreen for other issues nearer to Musk's heart. Most notably, the big beautiful bill is set to cut the electric vehicle tax credits that made Tesla the automotive giant it is today. (Confusingly, as recently as last year, Musk was publicly calling for an end to the tax credit — but that was before his activities in the White House eviscerated Tesla's brand image and sent it deeply into the red.)Of course, that raises another intriguing possibility: that at least some portion of Musk's rage at Trump is essentially kayfabe, with Musk betting that a break from the president could resuscitate at least some enthusiasm for the Tesla brand among the left-leaning customers that he's successfully turned off over the past year.If so, it's not hard to imagine Musk instead accidentally alienating more or less everybody — failing to get the environmental left back on board, but also creating a powerful enemy with Trump, who holds immense power over the government contracts and policy that keep Musk's business empire afloat.More on politics: Elon Musk’s Dad Slams His Son's Whimpering Failure at PoliticsShare This Article

Many don’t like it, but (almost) everybody agrees it’s the future.“Vibe Coding” is everywhere. Tools and game engines are implementing AI-assisted coding, vibe coding interest skyrocketed on Google search, on social media, everybody claims to build apps and games in minutes, while the comment section gets flooded with angry developers calling out the pile of garbage code that will never be shipped.A screenshot from Andrej Karpathy with the original “definition” of Vibe CodingBUT, how do professionals feel about it?This is what I will cover in this article. We will look at:How people react to the term vibe coding,How their attitude differs based on who they are and their professional experienceThe reason for their stance towards “vibe coding” (with direct quotes)How they feel about the impact “vibe coding” will have in the next 5 yearsIt all started with this survey on LinkedIn. I have always been curious about how technology can support creatives and I believe that the only way to get a deeper understanding is to go beyond buzzwords and ask the hard questions. That’s why for over a year, I’ve been conducting weekly interviews with both the founders developing these tools and the creatives utilising them. If you want to learn their journeys, I’ve gathered their insights and experiences on my blog called XR AI Spotlight.Driven by the same motives and curious about people’s feelings about “vibe coding”, I asked a simple question: How does the term “Vibe Coding” make you feel?Original LinkedIn poll by Gabriele RomagnoliIn just three days, the poll collected 139 votes and it was clear that most responders didn’t have a good “vibe” about it. The remaining half was equally split between excitement and no specific feeling.But who are these people? What is their professional background? Why did they respond the way they did?Curious, I created a more comprehensive survey and sent it to everyone who voted on the LinkedIn poll.The survey had four questions:Select what describes you best: developers, creative, non-creative professionalHow many years of experience do you have? 1–5, 6–10, 11–15 or 16+Explain why the term “vibe coding” makes you feel excited/neutral/dismissive?Do you think “vibe coding” will become more relevant in the next 5 years?: It’s the future, only in niche use cases, unlikely, no idea)In a few days, I collected 62 replies and started digging into the findings, and that’s when I finally started understanding who took part in the initial poll.The audienceWhen characterising the audience, I refrained from adding too many options because I just wanted to understand:If the people responding were the ones making stuffWhat percentage of makers were creatives and what developersI was happy to see that only 8% of respondents were non-creative professionals and the remaining 92% were actual makers who have more “skin in the game“ with almost a 50/50 split between creatives and developers. There was also a good spread in the degree of professional experience of the respondents, but that’s where things started to get surprising.Respondents are mostly “makers” and show a good variety in professional experienceWhen creating 2 groups with people who have more or less than 10 years of experience, it is clear that less experienced professionals skew more towards a neutral or negative stance than the more experienced group.Experienced professionals are more positive and open to vibe codingThis might be because senior professionals see AI as a tool to accelerate their workflows, while more juniors perceive it as a competitor or threat.I then took out the non-professional creatives and looked at the attitude of these 2 groups. Not surprisingly, fewer creatives than developers have a negative attitude towards “vibe coding” (47% for developers Vs 37% for creatives), but the percentage of creatives and developers who have a positive attitude stays almost constant. This means that creatives have a more indecisive or neutral stance than developers.Creatives have a more positive attitude to vibe coding than developersWhat are people saying about “vibe coding”?As part of the survey, everybody had the chance to add a few sentences explaining their stance. This was not a compulsory field, but to my surprise, only 3 of the 62 left it empty (thanks everybody). Before getting into the sentiment analysis, I noticed something quite interesting while filtering the data. People with a negative attitude had much more to say, and their responses were significantly longer than the other group. They wrote an average of 59 words while the others barely 37 and I think is a good indication of the emotional investment of people who want to articulate and explain their point. Let’s now look at what the different groups of people replied.😍 Patterns in Positive Responses to “Vibe Coding”Positive responders often embraced vibe coding as a way to break free from rigid programming structures and instead explore, improvise, and experiment creatively.“It puts no pressure on it being perfect or thorough.”“Pursuing the vibe, trying what works and then adapt.”“Coding can be geeky and laborious… ‘vibing’ is quite nice.”This perspective repositions code not as rigid infrastructure, but something that favors creativity and playfulness over precision.Several answers point to vibe coding as a democratizing force opening up coding to a broader audience, who want to build without going through the traditional gatekeeping of engineering culture.“For every person complaining… there are ten who are dabbling in code and programming, building stuff without permission.”“Bridges creative with technical perfectly, thus creating potential for independence.”This group often used words like “freedom,” “reframing,” and “revolution.”.😑 Patterns in Neutral Responses to “Vibe Coding”As shown in the initial LinkedIn poll, 27% of respondents expressed mixed feelings. When going through their responses, they recognised potential and were open to experimentation but they also had lingering doubts about the name, seriousness, and future usefulness.“It’s still a hype or buzzword.”“I have mixed feelings of fascination and scepticism.”“Unsure about further developments.”They were on the fence and were often enthusiastic about the capability, but wary of the framing.Neutral responders also acknowledged that complex, polished, or production-level work still requires traditional approaches and framed vibe coding as an early-stage assistant, not a full solution.“Nice tool, but not more than autocomplete on steroids.”“Helps get setup quickly… but critical thinking is still a human job.”“Great for prototyping, not enough to finalize product.”Some respondents were indifferent to the term itself, viewing it more as a label or meme than a paradigm shift. For them, it doesn’t change the substance of what’s happening.“At the end of the day they are just words. Are you able to accomplish what’s needed?”“I think it’s been around forever, just now with a new name.”These voices grounded the discussion in the terminology and I think they bring up a very important point that leads to the polarisation of a lot of the conversations around “vibe coding”.🤮 Patterns in Negative Responses to “Vibe Coding”Many respondents expressed concern that vibe coding implies a casual, unstructured approach to coding. This was often linked to fears about poor code quality, bugs, and security issues.“Feels like building a house without knowing how electricity and water systems work.”“Without fundamental knowledge… you quickly lose control over the output.”The term was also seen as dismissive or diminishing the value of skilled developers. It really rubbed people the wrong way, especially those with professional experience.“It downplays the skill and intention behind writing a functional, efficient program.”“Vibe coding implies not understanding what the AI does but still micromanaging it.”Like for “neutral” respondents, there’s a strong mistrust around how the term is used (especially on social media) where it’s seen as fueling unrealistic expectations or being pushed by non-experts.“Used to promote coding without knowledge.”“Just another overhyped term like NFTs or memecoins.”“It feels like a joke that went too far.”Ultimately, I decided to compare attitudes that are excited (positive) and accepting (neutral) of vibe coding vs. those that reject or criticise it. After all, even among people who were neutral, there was a general acceptance that vibe coding has its place. Many saw it as a useful tool for things like prototyping, creative exploration, or simply making it easier to get started. What really stood out, though, was the absence of fear that was very prominent in the “negative” group and saw vibe coding as a threat to software quality or professional identity.People in the neutral and positive groups generally see potential. They view it as useful for prototyping, creative exploration, or making coding more accessible, but they still recognise the need for structure in complex systems. In contrast, the negative group rejects the concept outright, and not just the name, but what it stands for: a more casual, less rigorous approach to coding. Their opinion is often rooted in defending software engineering as a disciplined craft… and probably their job.😍 “As long as you understand the result and the process, AI can write and fix scripts much faster than humans can.”🤮 “It’s a joke. It started as a joke… but to me doesn’t encapsulate actual AI co-engineering.”On the topic of skill and control, the neutral and positive group sees AI as a helpful assistant, assuming that a human is still guiding the process. They mention refining and reviewing as normal parts of the workflow. The negative group sees more danger, fearing that vibe coding gives a false sense of competence. They describe it as producing buggy or shallow results, often in the hands of inexperienced users.😑 “Critical thinking is still a human job… but vibe coding helps with fast results.”🤮“Vibe-Coding takes away the very features of a good developer… logical thinking and orchestration are crucial.”Culturally, the divide is clear. The positive and neutral voices often embrace vibe coding as part of a broader shift, welcoming new types of creators and perspectives. They tend to come from design or interdisciplinary backgrounds and are more comfortable with playful language. On the other hand, the negative group associates the term with hype and cringe, criticising it as disrespectful to those who’ve spent years honing their technical skills.😍“It’s about playful, relaxed creation — for the love of making something.”🤮Creating a lot of unsafe bloatware with no proper planning.”What’s the future of “Vibe Coding”?The responses to the last question were probably the most surprising to me. I was expecting that the big scepticism towards vibe coding would align with the scepticism on its future, but that was not the case. 90% of people still see “vibe coding” becoming more relevant overall or in niche use cases.Vibe coding is here to stayOut of curiosity, I also went back to see if there was any difference based on professional experience, and that’s where we see the more experienced audience being more conservative. Only 30% of more senior Vs 50% of less experienced professionals see vibe coding playing a role in niche use cases and 13 % Vs only 3% of more experienced users don’t see vibe coding becoming more relevant at all.More experienced professionals are less likely to think Vibe Coding is the futureThere are still many open questions. What is “vibe coding” really? For whom is it? What can you do with it?To answer these questions, I decided to start a new survey you can find here. If you would like to further contribute to this research, I encourage you to participate and in case you are interested, I will share the results with you as well.The more I read or learn about this, I feel “Vibe Coding” is like the “Metaverse”:Some people hate it, some people love it.Everybody means something differentIn one form or another, it is here to stay.What professionals really think about “Vibe Coding” was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.