Image Credit: Chat GPT o3I’ve been using AI coding assistants like Claude Code for a while now, and I’m here to say (with all due respect to people who have substance abuse issues), I may be an addict. And boy is this is an expensive habit.Its “almost there” quality — the feeling we’re just one prompt away from the perfect solution — is what makes it so addicting. Vibe coding operates on the principle of variable-ratio reinforcement, a powerful form of operant conditioning where rewards come unpredictably. Unlike fixed rewards, this intermittent success pattern (“the code works! it’s brilliant! it just broke! wtf!”), triggers stronger dopamine responses in our brain’s reward pathways, similar to gambling behaviors.What makes this especially effective with AI is the minimal effort required for potentially significant rewards — creating what neuroscientists call an “effort discounting” advantage. Combined with our innate completion bias — the drive to finish tasks we’ve started — this creates a compelling psychological loop that keeps us prompting.I don’t smoke, but don’t these bar graphs look like ciagrettes?Since Claude Code has been released, I have probably spent over $1,000 vibe coding various projects into reality (some of which I hope to announce soon, don’t worry).But lets talk about the expense too, because I think there’s something bad there as well: coding agents, and especially Claude 3.7 (the backend of Claude Code), tend to write too much code, a phenomenon that ends up costing users more than it should.Where an experienced developer might solve a problem with a few elegant lines with a thoughtful functional method, these AI systems often produce verbose, over-engineered solutions that tackle problems incrementally rather than addressing them at their core.My initial reaction was to attribute this to the relative immaturity of LLMs and their limitations when reasoning about abstract logic problems. Since these models are primarily trained to predict and generate text based on patterns they’ve seen before, it makes sense that they might struggle with the deeper architectural thinking that leads to elegant, minimal solutions.My human code on the left, Claude Code on the right implementing the same algorithmAnd indeed, the highly complex tasks I’ve handed to them have largely resulted in failure: implementing a minimax algorithm in a novel card game, crafting thoughtful animations in CSS, completely refactoring a codebase. The LLMs routinely get lost in the sauce when it comes to thinking through the high level principles required to solve difficult problems with computer science.In the example above, my human implemented version of minimax from 2018 totals 400 lines of code, whereas Claude Code’s version comes in at 627 lines. The LLM version also requires almost a dozen other library files. Granted, this version is in TypeScript and has a ton of extra bells and whistles, some of which I explicitly asked for, but the real problem is: it doesn’t actually work. Furthermore, using the LLM to debug it requires sending the bloated code back and forth to the API every time I want to holistically debug it.In an effort to impress the user and over-deliver, LLMs end up creating a rat’s nest of ultra-defensive code littered with debugging statements, neurotic comments and barely-useful helper funcitions. If you’ve ever worked in a highly functional production codebase, this is enough to drive you insane.I think everyone who spends any time vibe coding eventually discovers something like this and realizes that it’s much more worthwhile to work with a plan composed of discrete tasks that could be explained to a junior level developer vs. a feature-level project handed off to a staff engineer.There’s also the likelihood that the vast majority of code that LLMs have been trained on tends to be inelegant and overly verbose. Lord knows there’s a lot of AbstractJavaFinalSerializedFactory code out there.But I’m beginning to think the problem runs deeper, and it has to do with the economics of AI assistance.The economic incentive problemMany AI coding assistants, including Claude Code, charge based on token count — essentially the amount of text processed and generated. This creates what economists would call a “perverse incentive” — an incentive that produces behavior contrary to what’s actually desired.Let’s break down how this works:The AI generates verbose, procedural code for a given taskThis code becomes part of the context when you ask for further changes or additions (this is key)The AI now has to read (and you pay for) this verbose code in every subsequent interactionMore tokens processed = more revenue for the company behind the AIThe LLM developers have no incentive to “fix” the verbose code problem because doing so will meaningfully impact their bottom lineAs Upton Sinclair famously noted: “It is difficult to get a man to understand something when his salary depends on his not understanding it.” Similarly, it might be difficult for AI companies to prioritize code conciseness when their revenue depends on token count.The broader implicationsThis pattern points to a more general concern in AI development: the alignment between how systems are monetized and how well they serve user needs. When charging by token count, there’s naturally less incentive to optimize for elegant, minimal solutions.Even “all you can eat” subscription plans (e.g. Claude’s “Max” subscription) don’t fully resolve this tension, as they typically come with usage caps or other limitations that maintain the underlying incentive structure.System instructions and verbosity trade-offsThe perverse incentives in AI code generation point to a more fundamental issue that extends beyond coding assistants. When she was reading a draft of this, Louise pointed out some recent research from Giskard AI’s Phare benchmark that reveals a troubling pattern that mirrors our coding dilemma: demanding shorter responses jeopardizes the accuracy of the answers.According to their findings, instructions emphasizing conciseness (like “answer this question briefly”) significantly degraded factual reliability across most models tested — in some cases causing a 20% drop in hallucination resistance. When forced to be concise, models face an impossible choice between fabricating short but inaccurate answers or appearing unhelpful by rejecting the question entirely. The data shows models consistently prioritize brevity over accuracy when given these constraints.There’s clearly something going on where the more verbose the LLM is, the better it does. This actually makes sense given the discovery that chain-of-thought reasoning improves accuracy, but this issue has begun to feel like a real tradeoff when it comes to these almost-magical systems.We see this exact tension in code generation every day. When we optimize for conciseness and ask for the problems to be solved in fewer setps, we often sacrifice quality. The difference is that in coding, the sacrifice manifests as over-engineered verbosity — the model produces more tokens to cover all possible edge cases rather than thinking deeply about the elegant core solution or a root cause problem. In both cases, economic incentives (token optimization) work against quality outcomes (factual accuracy or elegant code).Just as Phare’s research suggests that seemingly innocent prompts like “be concise” can sabotage a model’s ability to debunk misinformation, our experience shows that standard prompting approaches can yield bloated, inefficient code. In both domains, the fundamental misalignment between token economics and quality outputs creates a persistent tension that users must actively manage.Some tricks to manage these perverse incentivesWhile we wait for AI companies to better align their incentives with our need for elegant code, I’ve developed several strategies to counteract verbose code generation:1. Force planning before implementationI harass the LLM to write a detailed plan before generating any code. This forces the model to think through the architecture and approach, rather than diving straight into implementation details. Often, I find that a well-articulated plan leads to more concise code, as the model has already resolved the logical structure of the solution before writing a single line.2. Explicit permission protocolI’ve implemented a strict “ask before generating” protocol in my workflow. My personal CLAUDE.md file explicitly instructs Claude to request permission before writing any code. Infuriatingly, Claude Code regularly ignores this, likely due to its massive system prompt that talks so much about writing code it overrides my preferences. Enforcing this boundary and repeatedly belaboring it (“remember, don’t write any code”) helps prevent the automatic generation of unwanted, verbose solutions.3. Git-based experimentation with ruthless pruningVersion control becomes essential when working with AI-generated code. I frequently benchmark code in git when I arrive at an “ok it works as intended” moment. Creating experimental branches is also very helpful. Most importantly, I’m ready to throw out branches entirely when fixing them would require more work than starting from scratch. This willingness to abandon sunk costs is surprisingly important — it helps me work through problems and figure out the AI’s hangups while preventing the accumulation of bandaid solutions on top of fundamentally flawed approaches.4. Use a cheaper modelSometimes the simplest solution works best: using a smaller, cheaper model often results in more direct solutions. These models tend to generate less verbose code simply because they have limited context windows and processing capacity. While they might not handle extremely complex problems as well, for many day-to-day coding tasks, their constraints can actually produce more elegant solutions. For example, Claude 3.5 Haiku is currently 26% the price of Claude 3.7 ($0.80 per token vs. $3). Also, Claude 3.7 seems to overengineer more frequently than Claude 3.5.Moving toward better alignmentWhat might a better approach look like?LLM coding agents could evaluated and incentivized based on code quality metrics rather than just token counts. The challenge here is that this kind of metric is quite subjective.Companies could offer pricing models that reward efficiency rather than verbosity (I have no idea how this would work, this was Claude’s dumb idea)LLMs training should incorporate feedback mechanisms that specifically promote concise, elegant solutions via RLHF (e.g. showing developers multiple versions of the same code and having them pick the opitmal one, perhaps this is already happening)Companies realize that overly verbose code generation is not good for their bottom line (e.g. Sam Altman admitted that users saying “please” and “thank you” to ChatGPT is costing them millions of dollars)This isn’t just about getting better AI — it’s about making sure that the economic incentives driving AI development align with what we actually value as developers: clean, maintainable, elegant code that solves problems at their root.Until then, don’t forget: brevity is the soul of wit, and machines have no soul.Thanks to Louise Macfadyen, Justin Kazmark and Bethany Crystal for reading and suggesting edits to a draft of this.— -PS: Yes, I used Claude to help write this post critiquing AI verbosity. There’s a delicious irony here: these systems will happily help you articulate why they might be ripping you off. Their willingness to steelman arguments against their own economic interests shows that the perverse incentives aren’t embedded in the models themselves, but in the business decisions surrounding them. In other words, don’t blame the AI — blame the humans optimizing the revenue models. The machines are just doing what they’re told, even when that includes explaining how they’re being told to do too much.The perverse incentives of Vibe Coding was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.

In this blog, we will cover five key workflows in the new Memory Profiler package that you can use to diagnose and examine memory-related issues in your game. These are:Monitoring your application’s memory pressureSeeing the distribution of Unity ObjectsDetecting poorly configured assetsLocating unintentional duplicate objectsComparing memory captures to validate optimizationsFor an introduction to the Memory Profiler, please see the recent blog, Everything you need to know about Memory Profiler 1.0.0.This first workflow monitors how demanding your application is on a device’s memory resources. This process is critical to determining whether or not your application is at risk of performance problems, or even being evicted and terminated by the operating system, due to consuming too much memory.To begin, we have a build of an example game running on the target device. Naturally, it is essential that we take a memory capture of the game, running on the actual hardware, to see how it uses the devices' available memory resources. Furthermore, memory does not behave in the same way in the Unity Editor as it does in the Unity runtime, so taking a memory capture of the Editor in Play Mode is not a good representation of how a game’s memory will look on a device. (Taking a memory capture of the Editor is appropriate when developing tools for the Editor, such as custom Editor windows.)After navigating to the stage in our game where we want to analyze the memory usage, we attach the Memory Profiler to our device using the dropdown in the Memory Profiler. We can then take a memory capture, as shown below.After opening this capture, the Memory Profiler displays our application’s memory footprint at the top of the Summary page as “Memory Usage On Device”.Here we can see that our application’s memory footprint is 492.5 MB, out of an available 3.50 GB. We need to use our best judgment next as to whether we believe that is a sensible proportion of the device’s physical memory (RAM) to be using at the time of capture. Remember that a device’s physical memory is shared by all running processes.You’ll notice that this visual indicator is showing you total resident memory. Total resident memory refers to how much of your application’s memory resides in the device’s physical memory hardware (RAM). This is the clearest indicator of how demanding your application’s current memory usage is on the target device for two reasons. First, this is because as your application’s total resident memory usage increases, so does the likelihood of incurring frequent page faults, where the operating system has to page virtual memory in and out of the device’s physical memory. Frequent page faults will cause significant performance degradation of your application. And second, this is because many operating systems use your application’s resident memory usage to determine its current memory footprint. If your application’s memory footprint gets too high, the operating system will evict your application and terminate it, causing a crash for your players.Therefore, you can use the Memory Usage On Device visual indicator in the Memory Profiler to infer if an application might be at risk of performance issues or being terminated by the operating system, due to an overuse of memory at the time of capture.This contrasts with Allocated Memory, sometimes referred to as Committed Memory, which you might notice is displayed in various graphics below this indicator, and is currently the default option shown by all other views, such as Unity Objects. Allocated Memory refers to all memory that your application currently has allocated, regardless of whether it has been made resident in physical memory or not, and therefore it matches your application’s view of memory more closely. As such, this can be useful for exploring all of your application’s currently allocated memory, whilst resident memory usage is key to understanding the memory pressure your application is placing on the hardware at any moment in time.The Memory Profiler’s Unity Objects tab provides you with an overview of your application’s memory from the perspective of Unity Objects; that’s your application’s textures, shaders, meshes, materials, and so on. This is a great place to begin exploration in the Memory Profiler because Unity Objects will be inherently familiar to so many Unity users, as it is what the majority of us work with directly in the Unity Editor. Not only does this provide a familiar entry point to understanding our application’s memory, but it can also help to diagnose and fix a range of potential issues by providing this Unity-specific context.To see the Unity Objects view, simply select the Unity Objects tab at the top of the Memory Profiler after opening a memory capture, as shown above.You can see how the Unity Objects view quickly gives us an understanding of the distribution of Unity Object types in our application. This allows us to both gain a high-level understanding of what types were consuming the most memory at the time of capture, as well as to reason about this, such as whether it is expected that a particular scene is heavy on AudioClip objects, for example. Expanding each type also enables us to view every Unity Object that is currently allocated, individually, as shown below.It’s important to remember that Unity Objects make up a proportion of our application’s total allocated memory. You can see exactly how much in the indicator above the table, highlighted below.Here, you can see that our total allocated memory size, “Total Memory In Snapshot”, is 4.64GB and that our Unity Objects account for 2.37GB of that. Furthermore, if we filter the table – for example, by using the search feature – you’ll notice that this bar updates to reflect our search results. In other words, it displays the size of all the memory currently shown in the table. This helps you to maintain perspective of exactly how much memory you are inspecting as a proportion of the whole capture and can help to inform where to invest optimization efforts.In version 1.0 of Memory Profiler, the Unity Objects table shows you Allocated Memory, or, put another way, it shows you all Unity Objects that are alive in your application. We are exploring adding Resident Memory visibility to these views in an upcoming release, which would enable you to see exactly which of your Unity Objects are currently resident in physical memory, and therefore see exactly which are directly contributing to your application’s current memory footprint. You can use the All Of Memory tab to inspect the remainder of your application’s memory at the time of capture, which will include memory outside of Unity Objects, such as various Unity subsystems, managed-only (C#) memory, and DLLs and executables.The Unity Objects view can help us to diagnose a range of potential issues. One such issue is detecting assets that have been badly configured, causing them to consume more memory than is necessary.In the capture below, you can see that a substantial portion of our Unity Objects are textures. The capture is from a project with high graphical fidelity that uses the High Definition Render Pipeline and makes heavy use of visual effects. So, with this context in mind, we expect to see heavy use of textures, which we do.However, upon expanding our second large category, Texture2D, we can notice that two textures appear much bigger than the others. Using our understanding of our project, we are surprised that these textures are bigger than comparable textures, like HoloTable_Normal or HoloTable_Mask, as we expected them to be similar in size.So, we select one of these textures in the table to learn more details about it, and to begin investigating what might be the cause for this. Here, in the Details view we find our explanation – our texture is writable, or “Read/Write Enabled.”This is a common problem that we see across many user projects: accidentally making a texture writable when it’s not needed by checking the “Read/Write” setting on the texture’s import settings. When a texture has this flag enabled, it will double its size in memory. This is because a second copy of the texture data is required so that it can be accessed on the CPU. A tell-tale sign of this is that the Total Size of a texture is twice the size of what you expected, or twice the size of similar textures.After disabling the “Read/Write” flag on both of these textures and taking a second capture, we can see both of these textures have halved in size.We are exploring adding a column for graphics (GPU) memory to the Unity Objects table in a future release to make locating cases where a Unity Object has allocated graphics memory, such as in this example, easier.A common mistake that we see in Unity projects is unintentionally creating duplicate Unity Objects. For example, it is very easy to accidentally create a duplicate Material by accessing a MeshRenderer’s material property. Not only does this add up quickly in this case – if, for example, it is done on every instance of a particular MeshRenderer – but, furthermore, these dynamically created materials must be explicitly destroyed.To help with locating this type of issue, the Unity Objects table provides a quick filter to show you potential duplicate Unity Objects only. This view will filter the table to show only Unity Objects that have multiple instances with both an identical name and size. It is important to note that many potential duplicates will be expected and not a cause for concern at all. For example, multiple instances of a prefab might have identically named and sized Transform components, and these would be expected duplicates. It is only discovering unintentional duplicates that we are interested in, which we will illustrate in this workflow following example.The capture below was taken in a simple scene with two instances of a Door prefab, and we have enabled the Show Potential Duplicates Only filter located underneath the Unity Objects table. This has filtered the table to show us only Unity Objects that have multiple instances with the same name and size.Because we have two instances of a Door prefab in our scene, we also have, as expected, two instances of all the relevant objects: MeshRenderer, Transform, GameObject, and so on. However, we also have two instances of the “Door” Material in our capture above. These Door instances look the same in our scene, so it is expected that they would share a Material. This is, therefore, an unintentional duplicate, and in this particular example was caused by accessing the MeshRenderer’s material property in the prefab. Removing this property access and taking a second capture shows the duplicate material is no longer present in the Unity Objects table.It’s important to remember that this filter is simply showing you all Unity Objects that have multiple instances with the same name and size. It requires your knowledge of your project to interpret whether the potential duplicates you see are expected, or are, in fact, unintentional and cause for investigation. We recommend paying attention to the Total Memory In Table bar at the top, which gives you a visual indication of what proportion of your application’s allocated memory you are seeing in the table. This can help you to maintain perspective of where to invest your optimization efforts.The Memory Profiler also provides functionality to compare two memory captures. This allows us to make changes to our project, for example to address an issue we might have found, and subsequently test if our changes have indeed had the desired outcome. It is important to always test that your hypothesis is correct and your changes have had the desired outcome on the actual hardware. Here, let’s explore an example of this comparison workflow.Below is a capture of our mobile game taken during the first level. We can see that the biggest category of Unity Objects is Texture2D. After opening this category to check what our biggest textures are, we can see there are a few UI textures that are quite large in relation to the rest of our game – megabytes each. This raises a suspicion for us: Why are these textures so much larger than the others and do they need to be? To discover why, we can first locate the source texture asset in our project by selecting the texture in the Memory Profiler and using the “Select In Editor” button, which will highlight the source texture asset in our Project window.Using the Inspector window, we can see that all of our offending large UI textures are not being compressed due to their dimensions not being a power-of-two, as shown by the “NPOT” (non-power-of-two) text.This explains these large texture sizes. We can now use our knowledge of our project to reduce this memory usage. We know that three of these textures (the help controls) are always displayed together in the UI, as well as the other three textures (the creatures). Therefore, we can hypothesize with high confidence that creating two Sprite Atlases for each set of three textures will reduce our allocated memory usage, because it will enable them to be compressed without increasing the number of textures in memory.To compare two snapshots, begin by opening the first snapshot. This is the “base” against which we want to compare. Now above the open snapshot, select the “Compare Snapshots” tab and choose the second snapshot. The Memory Profiler will now present a summary comparing the two snapshots, as shown below.To see the effect of our change and verify that it did, in fact, reduce the size of our application’s allocated memory for the Texture2D category, we can select the Unity Objects tab. Here, we are presented with a comparison table that shows the Unity Object types that have changed, as well as how they have changed between the captures (shown below).We can see our Texture2D type as a whole has reduced in size by 3.6MB and has four less textures than before. Expanding this category, we can see the removal of our individual, uncompressed Sprite textures, and the addition of our two Sprite Atlas textures, resulting in a net reduction of 3.6MB and 4 Texture2D objects.So this was a success – we have confirmed that our hypothesis was correct using the comparison functionality, and we have reduced the size of these textures in allocated memory.From reading this blog, you should now have a better understanding of five key workflows in the new Memory Profiler package. These workflows are designed for diagnosing and examining memory-related issues in ayour game. We hope the Memory Profiler package released in Unity 2022.2 helps you to better monitor, examine, and understand your game’s memory footprint. Please feel free to reach out to the team to share your feedback on how we can improve performance profiling tools via our forum page –, or share your suggestions through our roadmap page, where you can also see some of the features that are being worked on.If you’re interested in more details on this topic, we will be publishing another blog in the coming weeks that will dive deeper into how an application’s memory footprint is calculated, covering topics such as resident and allocated memory in more detail.

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Google’s new AlphaEvolve shows what happens when an AI agent graduates from lab demo to production work, and you’ve got one of the most talented technology companies driving it. Built by Google’s DeepMind, the system autonomously rewrites critical code and already pays for itself inside Google. It shattered a 56-year-old record in matrix multiplication (the core of many machine learning workloads) and clawed back 0.7% of compute capacity across the company’s global data centers. Those headline feats matter, but the deeper lesson for enterprise tech leaders is how AlphaEvolve pulls them off. Its architecture – controller, fast-draft models, deep-thinking models, automated evaluators and versioned memory – illustrates the kind of production-grade plumbing that makes autonomous agents safe to deploy at scale. Google’s AI technology is arguably second to none. So the trick is figuring out how to learn from it, or even using it directly. Google says an Early Access Program is coming for academic partners and that “broader availability” is being explored, but details are thin. Until then, AlphaEvolve is a best-practice template: If you want agents that touch high-value workloads, you’ll need comparable orchestration, testing and guardrails. Consider just the data center win. Google won’t put a price tag on the reclaimed 0.7%, but its annual capex runs tens of billions of dollars. Even a rough estimate puts the savings in the hundreds of millions annually—enough, as independent developer Sam Witteveen noted on our recent podcast, to pay for training one of the flagship Gemini models, estimated to cost upwards of $191 million for a version like Gemini Ultra. VentureBeat was the first to report about the AlphaEvolve news earlier this week. Now we’ll go deeper: how the system works, where the engineering bar really sits and the concrete steps enterprises can take to build (or buy) something comparable. 1. Beyond simple scripts: The rise of the “agent operating system” AlphaEvolve runs on what is best described as an agent operating system – a distributed, asynchronous pipeline built for continuous improvement at scale. Its core pieces are a controller, a pair of large language models (Gemini Flash for breadth; Gemini Pro for depth), a versioned program-memory database and a fleet of evaluator workers, all tuned for high throughput rather than just low latency. A high-level overview of the AlphaEvolve agent structure. Source: AlphaEvolve paper. This architecture isn’t conceptually new, but the execution is. “It’s just an unbelievably good execution,” Witteveen says. The AlphaEvolve paper describes the orchestrator as an “evolutionary algorithm that gradually develops programs that improve the score on the automated evaluation metrics” (p. 3); in short, an “autonomous pipeline of LLMs whose task is to improve an algorithm by making direct changes to the code” (p. 1). Takeaway for enterprises: If your agent plans include unsupervised runs on high-value tasks, plan for similar infrastructure: job queues, a versioned memory store, service-mesh tracing and secure sandboxing for any code the agent produces.  2. The evaluator engine: driving progress with automated, objective feedback A key element of AlphaEvolve is its rigorous evaluation framework. Every iteration proposed by the pair of LLMs is accepted or rejected based on a user-supplied “evaluate” function that returns machine-gradable metrics. This evaluation system begins with ultrafast unit-test checks on each proposed code change – simple, automatic tests (similar to the unit tests developers already write) that verify the snippet still compiles and produces the right answers on a handful of micro-inputs – before passing the survivors on to heavier benchmarks and LLM-generated reviews. This runs in parallel, so the search stays fast and safe. In short: Let the models suggest fixes, then verify each one against tests you trust. AlphaEvolve also supports multi-objective optimization (optimizing latency and accuracy simultaneously), evolving programs that hit several metrics at once. Counter-intuitively, balancing multiple goals can improve a single target metric by encouraging more diverse solutions. Takeaway for enterprises: Production agents need deterministic scorekeepers. Whether that’s unit tests, full simulators, or canary traffic analysis. Automated evaluators are both your safety net and your growth engine. Before you launch an agentic project, ask: “Do we have a metric the agent can score itself against?” 3. Smart model use, iterative code refinement AlphaEvolve tackles every coding problem with a two-model rhythm. First, Gemini Flash fires off quick drafts, giving the system a broad set of ideas to explore. Then Gemini Pro studies those drafts in more depth and returns a smaller set of stronger candidates. Feeding both models is a lightweight “prompt builder,” a helper script that assembles the question each model sees. It blends three kinds of context: earlier code attempts saved in a project database, any guardrails or rules the engineering team has written and relevant external material such as research papers or developer notes. With that richer backdrop, Gemini Flash can roam widely while Gemini Pro zeroes in on quality. Unlike many agent demos that tweak one function at a time, AlphaEvolve edits entire repositories. It describes each change as a standard diff block – the same patch format engineers push to GitHub – so it can touch dozens of files without losing track. Afterward, automated tests decide whether the patch sticks. Over repeated cycles, the agent’s memory of success and failure grows, so it proposes better patches and wastes less compute on dead ends. Takeaway for enterprises: Let cheaper, faster models handle brainstorming, then call on a more capable model to refine the best ideas. Preserve every trial in a searchable history, because that memory speeds up later work and can be reused across teams. Accordingly, vendors are rushing to provide developers with new tooling around things like memory. Products such as OpenMemory MCP, which provides a portable memory store, and the new long- and short-term memory APIs in LlamaIndex are making this kind of persistent context almost as easy to plug in as logging. OpenAI’s Codex-1 software-engineering agent, also released today, underscores the same pattern. It fires off parallel tasks inside a secure sandbox, runs unit tests and returns pull-request drafts—effectively a code-specific echo of AlphaEvolve’s broader search-and-evaluate loop. 4. Measure to manage: targeting agentic AI for demonstrable ROI AlphaEvolve’s tangible wins – reclaiming 0.7% of data center capacity, cutting Gemini training kernel runtime 23%, speeding FlashAttention 32%, and simplifying TPU design – share one trait: they target domains with airtight metrics. For data center scheduling, AlphaEvolve evolved a heuristic that was evaluated using a simulator of Google’s data centers based on historical workloads. For kernel optimization, the objective was to minimize actual runtime on TPU accelerators across a dataset of realistic kernel input shapes. Takeaway for enterprises: This clarity allows the agent to self-improve and demonstrate unambiguous value. 5. Laying the groundwork: essential prerequisites for enterprise agentic success While AlphaEvolve’s achievements are inspiring, Google’s paper is also clear about its scope and requirements. The primary limitation is the need for an automated evaluator; problems requiring manual experimentation or “wet-lab” feedback are currently out of scope for this specific approach. The system can consume significant compute – “on the order of 100 compute-hours to evaluate any new solution” (AlphaEvolve paper, page 8), necessitating parallelization and careful capacity planning. Before allocating significant budget to complex agentic systems, technical leaders must ask critical questions: Machine-gradable problem? Do we have a clear, automatable metric against which the agent can score its own performance? Compute capacity? Can we afford the potentially compute-heavy inner loop of generation, evaluation, and refinement, especially during the development and training phase? Codebase & memory readiness? Is your codebase structured for iterative, possibly diff-based, modifications? And can you implement the instrumented memory systems vital for an agent to learn from its evolutionary history? Takeaway for enterprises: The increasing focus on robust agent identity and access management, as seen with platforms like Frontegg, Auth0 and others, also points to the maturing infrastructure required to deploy agents that interact securely with multiple enterprise systems. The agentic future is engineered, not just summoned AlphaEvolve’s message for enterprise teams is manifold. First, your operating system around agents is now far more important than model intelligence. Google’s blueprint shows three pillars that can’t be skipped: Deterministic evaluators that give the agent an unambiguous score every time it makes a change. Long-running orchestration that can juggle fast “draft” models like Gemini Flash with slower, more rigorous models – whether that’s Google’s stack or a framework such as LangChain’s LangGraph. Persistent memory so each iteration builds on the last instead of relearning from scratch. Enterprises that already have logging, test harnesses and versioned code repositories are closer than they think. The next step is to wire those assets into a self-serve evaluation loop so multiple agent-generated solutions can compete, and only the highest-scoring patch ships.  As Cisco’s Anurag Dhingra, VP and GM of Enterprise Connectivity and Collaboration, told VentureBeat in an interview this week: “It’s happening, it is very, very real,” he said of enterprises using AI agents in manufacturing, warehouses, customer contact centers. “It is not something in the future. It is happening there today.” He warned that as these agents become more pervasive, doing “human-like work,” the strain on existing systems will be immense: “The network traffic is going to go through the roof,” Dhingra said. Your network, budget and competitive edge will likely feel that strain before the hype cycle settles. Start proving out a contained, metric-driven use case this quarter – then scale what works. Watch the video podcast I did with developer Sam Witteveen, where we go deep on production-grade agents, and how AlphaEvolve is showing the way: Daily insights on business use cases with VB Daily If you want to impress your boss, VB Daily has you covered. We give you the inside scoop on what companies are doing with generative AI, from regulatory shifts to practical deployments, so you can share insights for maximum ROI. Read our Privacy Policy Thanks for subscribing. Check out more VB newsletters here. An error occured.

Villa 144 / Lin3rd ArchitectsSave this picture!© Parham TaghioffHouses•Bushehr, Iran Architects: Lin3rd Architects Area Area of this architecture project Area:  313 m² Year Completion year of this architecture project Year:  2025 Photographs Photographs:Parham TaghioffMore SpecsLess Specs Save this picture! Text description provided by the architects. Dwelling emerges through a presentful, dynamic, and constructive dialogue between the subject and the place. Villa 144 is situated near Bushehr, a city in southwestern Iran along the Persian Gulf, characterized by a hot and humid climate. The residential architecture of old Bushehr, shaped mainly during the 18th and 19th centuries, features unique traits such as narrow, labyrinthine alleyways, making it one of the most expressive and unconventional narrations of Iran's traditional residential architecture.Save this picture!The villa's structure comprises three white cubes that interlock at varying levels, defining both interior and exterior spaces. At the intersection of these cubes lies the main entrance, which acts as a pivotal boundary between the primary living area, bedrooms, and the western and eastern courtyards. The arrangement of sliding and overlapping cubes evokes a sense of lightness and suspension, subtly echoing a defining spatial quality of Bushehr's historic homes. Along the path to the entrance, the main living room cube appears to float above a reflecting pool clad in turquoise glazed Iranian tiles. In the region's intense heat, the presence and sound of water offer a soft, sensory counterpoint and create a moment of gentle release.Save this picture!Save this picture!Save this picture!Save this picture!The main living room is the most transparent and outward-facing space, visually and functionally linked to the villa's four courtyards. In the living room, the focal point is not a television but an antique Heriz rug —a masterpiece of Persian weaving—displayed as a framed artwork on the wall. Similarly, the kitchen's window seat, adorned with handwoven Zanjan rugs, reflects a deliberate practice to reveal Persian carpets in settings that amplify their symbolic, aesthetic, and spatial resonance. A 3.3-meter-long corridor connects the main entrance to the living room and kitchen. The walls of this corridor are built from coral stones sourced from the shores of Bushehr—once the principal construction material in the region's historic buildings—while its ceiling and floor are clad in raw iron sheets.Save this picture!Save this picture!The villa's landscape has been designed to offer multiple pathways, allowing visitors to explore the garden in various intended and unintended ways. These routes provide deeper engagement and a more profound understanding of both the garden and the villa. The garden is a collection of native and nostalgic trees of Bushehr, including different varieties of bougainvillea (the most beloved tree in Bushehr), several types of the Ficus family, six rescued mature eucalyptus trees, and the majestic silk tree—the most striking tree of Bushehr. Throughout the design process, we excavated prominent archetypes of Persian—and more specifically, Bushehri—architecture, distilled their functional essence, and reconfigured them to align with contemporary lifestyle.Save this picture!In traditional Bushehri architecture, there are no cul-de-sacs. Instead, users move through a network of narrow alleys—reminiscent of Borges' labyrinths—allowing for layered, narrative-based spatial experiences. In the design of Villa 144, inspired by this ethos and with the aim of deepening subject-place interaction, the structure has been intentionally freed from spatial dead ends. The result is a layout that invites a more intuitive, intertwined, and multi-narrative engagement with place.Save this picture! Project gallerySee allShow less About this officeLin3rd ArchitectsOffice••• Published on May 16, 2025Cite: "Villa 144 / Lin3rd Architects " 16 May 2025. ArchDaily. Accessed . <https://www.archdaily.com/1030148/villa-144-lin3rd-architects&gt ISSN 0719-8884Save世界上最受欢迎的建筑网站现已推出你的母语版本!想浏览ArchDaily中国吗?是否 You've started following your first account!Did you know?You'll now receive updates based on what you follow! Personalize your stream and start following your favorite authors, offices and users.Go to my stream

America’s quest to protect its semiconductor technology from China has taken increasingly dramatic turns over the past few years—from export bans to global restrictions—but the latest proposal from Congress ventures into unprecedented territory. Lawmakers are now pushing for mandatory GPS-style tracking embedded in every AI chip exported from the United States, essentially turning advanced semiconductors into devices that report their location back to Washington.On May 15, 2025, a bipartisan group of eight House representatives introduced the Chip Security Act, which would require companies like Nvidia to embed location verification mechanisms in their processors before export. This represents perhaps the most invasive approach yet in America’s technological competition with China, moving far beyond restricting where chips can go to actively monitoring where they end up.The mechanics of AI chip surveillanceUnder the proposed Chip Security Act, AI chip surveillance would become mandatory for all “covered integrated circuit products”—including those classified under Export Control Classification Numbers 3A090, 3A001.z, 4A090, and 4A003.z. Companies like Nvidia would be required to embed location verification mechanisms in their AI chips before export, reexport, or in-country transfer to foreign nations.Representative Bill Huizenga, the Michigan Republican who introduced the House bill, stated that “we must employ safeguards to help ensure export controls are not being circumvented, allowing these advanced AI chips to fall into the hands of nefarious actors.” His co-lead, Representative Bill Foster—an Illinois Democrat and former physicist who designed chips during his scientific career—added, “I know that we have the technical tools to prevent powerful AI technology from getting into the wrong hands.”The legislation goes far beyond simple location tracking. Companies would face ongoing surveillance obligations, required to report any credible information about chip diversion, including location changes, unauthorized users, or tampering attempts. This creates a continuous monitoring system that extends indefinitely beyond the point of sale, fundamentally altering the relationship between manufacturers and their products.Cross-party support for technology controlPerhaps most striking about this AI chip surveillance initiative is its bipartisan nature. The bill enjoys broad support across party lines, co-led by House Select Committee on China Chairman John Moolenaar and Ranking Member Raja Krishnamoorthi. Other cosponsors include Representatives Ted Lieu, Rick Crawford, Josh Gottheimer, and Darin LaHood.Moolenaar said that “the Chinese Communist Party has exploited weaknesses in our export control enforcement system—using shell companies and smuggling networks to divert sensitive US technology.” The bipartisan consensus on AI chip surveillance reflects how deeply the China challenge has penetrated American political thinking, transcending traditional partisan divisions.The Senate has already introduced similar legislation through Senator Tom Cotton, suggesting that semiconductor surveillance has broad congressional support. Coordination between chambers indicates that some form of AI chip surveillance may become law regardless of which party controls Congress.Technical challenges and implementation questionsThe technical requirements for implementing AI chip surveillance raise significant questions about feasibility, security, and performance. The bill mandates that chips implement “location verification using techniques that are feasible and appropriate” within 180 days of enactment, but provides little detail on how such mechanisms would work without compromising chip performance or introducing new vulnerabilities.For industry leaders like Nvidia, implementing mandatory surveillance technology could fundamentally alter product design and manufacturing processes. Each chip would need embedded capabilities to verify its location, potentially requiring additional components, increased power consumption, and processing overhead that could impact performance—precisely what customers in AI applications cannot afford.The bill also grants the Secretary of Commerce broad enforcement authority to “verify, in a manner the Secretary determines appropriate, the ownership and location” of exported chips. This creates a real-time surveillance system where the US government could potentially track every advanced semiconductor worldwide, raising questions about data sovereignty and privacy.Commercial surveillance meets national securityAI chip surveillance proposal represents an unprecedented fusion of national security imperatives with commercial technology products. Unlike traditional export controls that simply restrict destinations, the approach creates ongoing monitoring obligations that blur the lines between private commerce and state surveillance.Representative Foster’s background as a physicist lends technical credibility to the initiative, but it also highlights how scientific expertise can be enlisted in geopolitical competition. The legislation reflects a belief that technical solutions can solve political problems—that embedding surveillance capabilities in semiconductors can prevent their misuse.Yet the proposed law raises fundamental questions about the nature of technology export in a globalized world. Should every advanced semiconductor become a potential surveillance device? How will mandatory AI chip surveillance affect innovation in countries that rely on US technology? What precedent does this set for other nations seeking to monitor their technology exports?Accelerating technological decouplingThe mandatory AI chip surveillance requirement could inadvertently accelerate the development of alternative semiconductor ecosystems. If US chips come with built-in tracking mechanisms, countries may intensify efforts to develop domestic alternatives or source from suppliers without such requirements.China, already investing heavily in semiconductor self-sufficiency following years of US restrictions, may view these surveillance requirements as further justification for technological decoupling. The irony is striking: efforts to track Chinese use of US chips may ultimately reduce their appeal and market share in global markets.Meanwhile, allied nations may question whether they want their critical infrastructure dependent on chips that can be monitored by the US government. The legislation’s broad language suggests that AI chip surveillance would apply to all foreign countries, not just adversaries, potentially straining relationships with partners who value technological sovereignty.The future of semiconductor governanceAs the Trump administration continues to formulate its replacement for Biden’s AI Diffusion Rule, Congress appears unwilling to wait. The Chip Security Act represents a more aggressive approach than traditional export controls, moving from restriction to active surveillance in ways that could reshape the global semiconductor industry.This evolution reflects deeper changes in how nations view technology exports in an era of great power competition. The semiconductor industry, once governed primarily by market forces and technical standards, increasingly operates under geopolitical imperatives that prioritize control over commerce.Whether AI chip surveillance becomes law depends on congressional action and industry response. But the bipartisan support suggests that some form of semiconductor monitoring may be inevitable, marking a new chapter in the relationship between technology, commerce, and national security.Conclusion: The end of anonymous semiconductors from America?The question facing the industry is no longer whether the US will control technology exports, but how extensively it will monitor them after they leave American shores. In this emerging paradigm, every chip becomes a potential intelligence asset, and every export a data point in a global surveillance network.The semiconductor industry now faces a critical choice: adapt to a future where products carry their own tracking systems, or risk being excluded from the US market entirely. As Congress pushes for mandatory AI chip surveillance, we may be witnessing the end of anonymous semiconductors and the beginning of an era where every processor knows exactly where it belongs—and reports back accordingly.See also: US-China tech war escalates with new AI chips export controlsWant to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.Explore other upcoming enterprise technology events and webinars powered by TechForge here.

10 incredible new tabletop games for you to play in summer 2025 Lucy Orr Published May 16, 2025 2:00pm Pokémon TCG is in for a hot summer (The Pokémon Company) GameCentral looks at the most exciting new summer tabletop releases, including adaptations of Final Fantasy, Assassin’s Creed, and Citizen Sleeper The tabletop games industry has become an unlikely victim of Trump’s tariff trade war. Just after it was recovering from Covid supply chain issues it now sees itself hit with manufacturing issues and an uncertain future. It’s so bad that board game developer CMON has already shut up shop and Stonemaier, famous for the hugely successful Wingspan, is suing the Trump administration. Meanwhile Cephalofair, developer of fan favourite Gloomhaven, can’t even get their product on to the shelves, as it’s stuck in China. While I don’t expect any empty shelves at the UK Games Expo this month, there’s definitely panic in the air. Although one company that doesn’t seem to be too concerned is Games Workshop, who have always manufactured most of their products in the UK – although accessories and terrain for your favourite Warhammer army might become harder to find in the future. After everyone got into it during lockdown, the tabletop industry was riding a huge boom, with recent industry projections of the market doubling to around £20 billion by 2030. But US tariffs have left the industry reeling and could see the price for tabletop games around the world rise significantly. Despite the doom and gloom there’s plenty of exciting new products already out this year and many more on the way from massive brands such as Pokémon and Disney, as well as new Kickstarters that you can print yourself, thereby completely bypassing any manufacturing issues. There’s also never been a better time to support your local board game shop or Dungeons & Dragons club, as their overheads rise, so here’s a selection of games you might find on their shelves right now and those coming soon, that I’m excited to play during the summer break. Magic: The Gathering – Final Fantasy Square Enix’s iconic Final Fantasy franchise is stepping onto the cardboard battlefield with Magic: The Gathering, and the result is as gloriously nostalgic as it is mechanically exciting. The Universes Beyond initiative brings beloved characters, summons, and settings into Magic’s gameplay, with Cloud Strife leading the charge. The Final Fantasy Starter Kit offers two pre-constructed 60 card decks, packed with flavour and function and perfect for newcomers attracted by the sight of a Chocobo or Moogle. Each deck includes five rares, a foil mythic legendary, deck boxes, and digital codes for Magic: The Gathering Arena. But the real draw? The cards themselves. They are stunning. Cloud channels Final Fantasy 7’s environmental and emotional themes, with equipment-focused synergies that feel spot-on. Stiltzkin the Moogle is a flavourful support piece for donation strategies and the terrifying Tonberry arrives with Deathtouch and First Strike, which is fitting for a creature that’s haunted players for decades. With gorgeous full art treatments by amazing artists such as Takahashi Kazuya and Yoshitaka Amano, underpinned by clever mechanical call backs, this crossover is more than fan service, it’s a lovingly crafted bridge between two gaming giants that’s bound to fly of the shelves. £15.99 on Amazon – releases June 13 Warhammer 40,000 Kill Team: Typhon There’s a possible future where Games Workshop is the final tabletop company left standing, as they dodge tariffs and take down licence infringers like a particularly vicious swarm of tyrranids. You too can act out this future in Kill Team: Typhon, which delivers the chaos of Warhammer 40,000 in a claustrophobic, subterranean brawl between flesh-rending Tyranid Raveners and a desperate Adeptus Mechanicus Battleclade. The latest Kill Team box looks stunning and turns up the tension with asymmetric forces: a lean, elite brood of Raveners – deadly melee predators that can tunnel through terrain – versus a jury-rigged Mechanicus strike team, built from repurposed servitors and guided by a technoarchaeologist scouring ancient relics. But this isn’t Helldivers 2. The Raveners can be customised into deadly variants like the Tremorscythe and Felltalon, each armed with bio-engineered weapons designed for close-quarters carnage. On the other side, the Mechanicus bring massed, lobotomised firepower: breachers, gunners, medics, and overseers to allow for some tactical coordination. Also included are Hormagaunts, the swarming Tyrranids shock troops and new Tyranid-infested terrain – always the standout feature of these kill team boxes, in my opinion, and perfect for narrative or larger 40K battles. Typhon embraces Kill Team’s strength: cinematic asymmetry and high stakes. Price TBA – releases June Scalpers are going to love this (The Pokémon Company) Pokémon Trading Card Game Scarlet & Violet – Destined Rivals Pokémon Center Elite Trainer Box This red and black box is going for gangbusters on eBay, before it’s even supposed to be out. The Scarlet & Violet Destined Rivals set reintroduces the beloved Trainer’s Pokémon mechanic, but now with a twist where players can align with iconic duos like Arven’s Mabosstiff and Ho-Oh ex or Cynthia and Garchomp ex. Or fall in with Team Rocket under Giovanni’s command, fielding heavy hitters like Mewtwo ex. It’s a rich throwback to the Gym Heroes era, with cards that spotlight specific trainer and pokémon bonds, each emblazoned with the trainer’s name. The expansion includes 83 cards branded under Team Rocket, 17 new Pokémon ex cards (10 of which are Trainer’s Pokémon ex), and a trove of high rarity collectibles: 23 illustration rares, 11 special illustration rares, and six hyper rare gold-etched cards. But the pre-launch hasn’t been all Sunflora and Jigglypuffs. Since its full reveal on March 24, pre-orders have sparked a frenzy amongst scalpers, with sellouts and early store hiccups are already marring the rollout. Still, between the nostalgia bait and villainous charm, Destined Rivals is shaping up to be one of 2025’s hottest trading card releases. RRP £54.99 – releases May 30 Finspan One game I just can’t put back on the shelf at the moment is Finspan; who’d of thought fish could be so much fun? Since Wingspan took flight in 2019, it’s become a modern classic: part art piece, part engine builder, and a benchmark for gateway games. Finspan, the third entry in the series, swaps feathers for fins, inviting players to explore marine ecosystems across oceanic zones in a beautifully illustrated, medium-lightweight game that last about 45 minutes. Mechanically, Finspan is more accessible than Wingspan, thanks to forgiving resource generation and a gentler deck structure. Strategic depth is still there, whether you chase high value fish, go wide with schools, or balance both. It’s more of a solo puzzle, and less about blocking opponents, which might suit more casual groups. Replayability is strong, and with one to five player support it scales well. The art is stunning, and the fish facts make you feel like a would-be marine biologist. But I missed the funny components (no birdhouse dice tower), so this is missing some of that Wingspan magic. Finspan is a fantastic entry point to the series and an accessible and fun addition to the franchise. It’s not as perfect as Wingspan, but it swims confidently in its own current. Could we see whale and crustacean expansions? I hope so. RRP £41.99 – available now Star Wars Unlimited – Jump To Lightspeed While you might have missed the Star Wars Celebration in Japan last month, and be bereft over the end of Andor Season 2, don’t worry – there are plenty of alternatives for Star Wars fandom. Fantasy Flight Games is revving its hyperdrive with Jump To Lightspeed, the fourth set for trading card game Star Wars Unlimited. A dramatic shift from previous ground-focused releases, this set propels players into orbit, with an emphasis on space combat and a host of gameplay refinements. Headlining the release are two new Spotlight Decks, each featuring a classic rivalry, such as Han Solo vs. Boba Fett. These 50-card preconstructed decks introduce Pilots, a brand-new card type that changes how space units operate. Pilots can be deployed to enhance ships with improved health and damage dealing abilities, offering fresh tactical depth. The set also debuts the Piloting keyword, a hyperspace mechanic, and five special rarity cards per deck, including one new Leader per Spotlight release. It’s a sleek continuation of Unlimited’s mission, with deep strategy wrapped in Star Wars flair. Fantasy Flight isn’t just releasing a new set; they’re effectively entering year two of the game with a soft reboot, that smartly rebalances and refreshes. For new and returning players, the standalone Spotlight Decks offer a refined on-ramp into the meta, while the stellar art and fan favourite make this one of the best sci-fi trading card games around. RRP: £34.99 – available now An indie tabletop game adapting an indie video game (Jump Over the Age) Citizen Sleeper: Spindlejack I was gutted I didn’t manage to nab some physical Cycles of the Eye Data-Cloud dice from Lost in Cult, before they sold out, so I was ecstatic to see the shadow drop of Citizen Sleeper: Spindlejack, especially as it’s completely free. It’s a lean, solo tabletop role-player set in the neon-drenched corridors of the Far Spindle, part of the Citizen Sleeper universe. Released on May 5th (aka Citizen Sleeper Day), it’s a print-and-play experience that trades dense narrative for kinetic delivery runs and tactical movement through a crumbling space station. Inspired by Kadet, the courier from Citizen Sleeper 2, Spindlejack casts you as one of the eponymous daredevils: airbike mounted messengers who dodge cargo haulers and urban decay to deliver sensitive payloads in a haunted, half-dead network. The draw? Not just the cryo or reputation, but the thrill, the competition, and the culture. Using your 10 six-sided dice, a pencil, and some printed sheets you’ll chart courses across randomly generated intersections, upgrade your bike, and edge toward Spindlejack legend status. Designed by Gareth Damian Martin, with stylish, gritty art from Guillaume Singelin, this is a tight, systems-focused dive into a beloved sci-fi setting. No campaign scheduling. No group required. Just you, your dice, and the Spindle’s rusted arteries. For fans of Citizen Sleeper or those craving a focused, atmospheric solo experience, Spindlejack is a no-brainer. DIY or DIE. Available now Disney Lorcana has become a certified hit (Ravensburger) Disney Lorcana – Reign Of Jafar Set 8 and Illumineer’s Quest: Palace Heist The internet has been on fire with the announcement that forthcoming Lorcana sets are to include Darkwing Duck and The Goofy Movie cast, emphasising that Disney Lorcana has become something of a juggernaut since its 2023 debut, captivating collectors and competitive players with a blend of nostalgic charm and evolving mechanics – judging scandals aside. During the Next Chapter of Lorcana livestream earlier this month, Ravensburger dropped major news. The autumn 2025 set, Fabled, will introduce Lorcana’s first ever set rotation, a sign the game is maturing into a competitive force. To support this shift, Fabled will include reprints from earlier sets, while also debuting two new rarity levels: epic and the ultra-rare Iconic. Reign Of Jafar, the game’s eighth set, sees Jafar rise as the new central villain, corrupting Archazia’s Island and bringing a darker twist to the narrative. Familiar faces like Mulan, Stitch, Rapunzel, and Bruno return, alongside new cards and accessories, including updated sleeves and deck boxes featuring classic Enchanted artwork. The new Illumineer’s Quest: Palace Heist PvE box expands on the beloved Deep Trouble, letting players face Jafar co-op style. Expect pre-built decks (Amethyst Amber and Ruby Steel), booster boxes, and enough lore-packed cardboard to fuel your summer break. £16.99 starter pack – releases May 30 The Lord Of The Rings: Fate Of The Fellowship While Finspan might be missing a dice tower, Fate Of The Fellowship more than makes up for that with a dice tower Barad-dûr. This is a one to five player co-op strategy game that builds on the Pandemic System but adds enough fresh features to feel distinct, deeper, and more precious than ever. Players take on the roles of Fellowship members and allies, racing to protect havens from surging shadow troops and helping Frodo sneak past the Nazgûl en route to Mount Doom. Unlike previous Pandemic adaptations, Fate Of The Fellowship leans hard into narrative mechanics. You’ll juggle four resources – stealth, valour, resistance, and friendship – across a sprawling map as you battle despair and shifting objectives. Each player commands two characters, with asymmetric abilities and layered decisions every turn. With 24 rotating objectives, a constant threat from the Eye of Sauron, and a cleverly tuned solo mode, designer Matt Leacock has crafted his richest Pandemic variant yet. I’ve seen plenty of tabletop gamers saying this will be their must-play at UK Games Expo. RRP £69.99 – releases June 27 Assassin’s Creed Animus Animus brings the Assassin’s Creed universe to the tabletop in a wholly fresh, narrative-driven experience. Up to four players select historical eras, each tied to a legendary assassin like Ezio or Eivor, resulting in distinct, asymmetric playstyles, unique objectives, and specialised mechanics. Rather than a miniatures skirmish, this is a competitive, timeline-jumping adventure where players dive into ancestral memories via the titular Animus. Strategic stealth and precision matter: while one player might rush to the end, victory favours those who stay synchronised with their ancestor’s memory by completing tasks efficiently and, of course, stealthily. While there’s still not much information about this game at the moment, Animus looks to employ modular and evolving dynamics driven by interactive card play. Players can impact each other’s timelines, which will hopefully keep the experience reactive and organic. With deep lore integration, and Ubisoft’s full support, this could the most ambitious Assassin’s Creed tabletop title yet. Crowdfunding starts summer 2025 Some like it Hoth (Days of Wonder) Star Wars: Battle Of Hoth Days of Wonder, the studio behind tabletop classic Ticket To Ride, has unveiled its next major release, with Star Wars: Battle Of Hoth. Designed for two to four players, aged 8 and up, this fast-paced board game runs around 30 minutes per session and leans on the accessible, card-driven Commands & Colors system. More Trending Players will face off as Imperial or Rebel forces across 17 scenario-driven missions, with options to escalate into campaign mode. Leader cards introduce familiar names like Vader, Luke, Leia, and Han to influence the tide of battle. Although it should be easy to learn, concerns linger about the scope of the battlefield. A cramped board could reduce tactical play to simple dice duels, something fans of strategic depth may find frustrating. Questions also remain about unit range and movement dynamics. Still, Battle Of Hoth promises cinematic nostalgia and the potential for layered tactics, and all for a very reasonable price. RRP: £49.99 – crowdfunding starts summer 2025 Email gamecentral@metro.co.uk, leave a comment below, follow us on Twitter, and sign-up to our newsletter. To submit Inbox letters and Reader’s Features more easily, without the need to send an email, just use our Submit Stuff page here. For more stories like this, check our Gaming page. GameCentral Sign up for exclusive analysis, latest releases, and bonus community content. This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply. Your information will be used in line with our Privacy Policy

Author(s): Adam BEN KHALIFA Originally published on Towards AI. Why are agentic workflows essential for achieving AGI Let me ask you this, what if the path to truly smart and effective AI , the kind we call AGI, isn’t just about building one colossal, all-knowing brain? What if the real breakthrough lies not in making our models only smarter, but in making them also capable of acting, adapting, and evolving? Well, LLMs continue to amaze us day after day, but the road to AGI demands more than raw intellect. It requires agency. Getting Our Terms Straight: AGI, Agency, and Agentic Workflows Before we dive in, let’s define the main concepts here: AGI — Artificial General Intelligence: You can see it as an AI model that can perform any intellectual task a human can. This means not just understanding language or generating images, but adapting, learning, reasoning, and acting across entirely new domains. Agency: The capacity of an entity to act purposefully in its environment to achieve goals. A rock has no agency; a human planning their day has plenty. For an AI, agency means it’s not just passively responding to prompts but actively pursuing objectives. Simply put, it’s the capacity to pursue goals autonomously through planning, acting, and adapting. Agentic Workflows: If agency is the “what”, agentic workflows are the “how”. These are the dynamic processes and systems an AI uses to exercise its agency. Think beyond a simple input-output model. Agentic workflows involve: Autonomous Goal-Setting & Planning: the AI doesn’t just execute a pre-defined plan, it can formulate goals and strategize how to achieve them. Tool Use & Orchestration: like a skilled craftsperson, it can select, combine, and utilize various “tools” (other AI models, databases, APIs, code execution environments) to get the job done. Memory & Learning: it remembers past actions, learns from successes and failures, and adapts its strategies over time. Adaptation in Dynamic Environments: the real world is messy, an agentic AI can adjust its plan when encountering unexpected obstacles or new information. It’s vital to understand the difference here: An LLM calling a weather API is just tool use. An agentic workflow is when an LLM, tasked with “analyzing market trends for a new product,” autonomously decides to:1) search recent financial news, 2) query a sales database, 3) use a data analysis tool to spot correlations, 4) ask a specialized forecasting model for projections, and then 5) compile a summary report, re-evaluating its approach at each step.It’s like the difference between a single musician playing one note, and a conductor leading an entire orchestra. The Limitation of Isolated Intelligence Let’s consider a human analogy. Imagine a brilliant engineer, a genius in their field. Now, strip away their tools: no computer, no internet for research, no pen and paper for sketching ideas or taking notes, no lab for prototyping, no colleagues to bounce ideas off. Confine them to only their thoughts. How much could they truly achieve? Their raw intellect, however vast, becomes severely handicapped when uncoupled from the ability to interact, experiment, and leverage external resources. This “intelligence in isolation” scenario illustrates a fundamental truth: intelligence doesn’t operate in a vacuum. It thrives on interaction, tool use, and the ability to execute plans in the world. If we want AGI, we can’t just build a disembodied digital brain, we need to build something that can act. Agentic Workflows: AI can act like us, and perhaps even better Humans are masters of adapting their “workflows.” A painter uses different tools and processes than an engineer, who uses different methods than a chef. We intuitively understand context, choose the right approach, and even invent new methods when old ones fail. Agentic workflows aim to achieve similar capabilities: Contextual Flexibility: An Agentic AI could switch between “investigative journalist mode” (querying databases, cross-referencing sources, interviewing) and “creative writer mode” (generating narratives, exploring styles) as needed for a complex task. Learning by Doing (and Re-doing): Human learning is an iterative workflow: observe, hypothesize, experiment, analyze, conclude, refine. Agentic systems can embody this, trying approaches, evaluating outcomes, and improving their strategies. Beyond Monolithic Thought: We don’t store everything in our heads. We use notes, computers, books, and critically, we delegate tasks to others. Agentic AI can similarly leverage external knowledge bases, specialized sub-agents, and computational tools, creating a distributed, more powerful form of intelligence. Thinking About Thinking: Humans possess meta-cognition — the ability to reflect on our own thought processes and adjust them. Agentic workflows, with their capacity for self-monitoring and re-planning, are a foundational step towards AI developing its own form of meta-cognition. Inventing New Ways: Perhaps soon enough, an advanced agentic AI won’t just use existing tools and workflows, but identify the need for entirely new ones and even contribute to their creation. A hallmark of true general intelligence. Not Just Helpful, But Mandatory: Why AGI Needs Agentic Workflows These capabilities aren’t just fancy add-ons. They are arguably essential for anything we’d recognize as AGI: Tackling Complexity: Real-world problems are messy, multifaceted, and rarely solved by a single, linear process. Agentic workflows will allow AI to break down these complex challenges into manageable sub-tasks, orchestrating diverse capabilities. Achieving Scale: Imagine trying to manage global logistics, conduct large-scale scientific research, or personalize education for millions with a single, rigid program. Agentic systems offer the modularity and dynamic coordination needed for such scale. Adaptability and Robustness: What happens when the data changes, a tool fails, or an assumption proves wrong? A static AI might grind to a halt. An agentic AI can adapt, re-plan, find alternative solutions, and continue pursuing its goal. It can handle the unexpected. Resourcefulness: Like our engineer, an AGI needs to be able to identify and use the right “tool” (be it a specific algorithm, dataset, or external service) for the job at hand, rather than trying to be a jack-of-all-trades with a single block massive model. Surpassing Human Adaptability The first step is for AI to achieve a human-like ability to set goals, plan, use tools, and adapt through agentic workflows. But the true promise of AGI lies in surpassing these capabilities: Speed: Learn and adapt at speeds incomprehensible to us, iterating through problem-solving cycles in milliseconds. Scale: Manage and orchestrate operations of immense complexity, juggling thousands of variables and “tools” simultaneously. Novelty: Devise entirely new, perhaps counter-intuitive, workflows and solutions to problems that humans haven’t even conceived of. Self-Improvement of Workflows: An AGI that doesn’t just use workflows but actively refines, optimizes, and even discovers fundamentally new and more efficient ways to achieve its goals. Deeper Meta-Learning: Learning how to learn, plan, and strategize more effectively over time, becoming increasingly more intelligent and capable. Long-Horizon Reasoning: Successfully breaking down and navigating extremely complex, multi-stage goals that unfold over extended periods, adapting robustly along the way. Obviously, this is easier said than done. Building true AGI presents formidable challenges: How do we design systems that can reliably plan in open-ended environments? How can they discover and integrate new tools seamlessly? How does the system learn which part of a long, complex workflow was responsible for success or failure? These are active areas of research, pushing the boundaries of what AI can do. Thankfully we are witnessing more and more breakthroughs everyday, and RL — Reinforcement Learning based approaches are showing great promise (which make sense actually, but that’s for another article). Conclusion: Agency as the Cornerstone of AGI The quest for AGI is more than a race for larger models or faster processing. It’s a quest for intelligence that is versatile, adaptive, and purposeful. Agentic workflows provide the framework for such intelligence, enabling AI to move beyond mere pattern recognition to become an active participant in the problem-solving process. Just as human collective general intelligence emerged not merely from neurons, but from networks of thought, culture, and action — we must build AGI not as a single-block model, but as an AI capable of learning, adapting, and acting. Agency, in this light, isn’t just a feature; it’s the fundamental engine that will drive us towards true Artificial General Intelligence. If you liked this article, make sure to follow for more.And you can find me on: Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor. Published via Towards AI