Committee

In this article, you will learn how to match an LLM inference optimization technique to the actual bottleneck in your workload by working through a…

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. How “reverse game theory” could keep AI agents in check A new approach known as “reverse game theory” aims to control powerful AI models by designing systems that incentivize them to stick to the rules. While game theory predicts how rational actors behave under particular rules, reverse game theory designs rules to elicit the behavior we want. Think of splitting a cake: if one person cuts it and the other gets first choice, the cutter has an incentive to make the portions equal. Researchers believe the approach could align AI more closely with our rules and instructions. But big questions remain about whether these mechanisms will work in practice. Here’s how reverse game theory could help make AI agents behave. —Amit Katwala Here’s who is falling in line for Trump’s “Super Intelligence” loyalty test At the end of September, President Trump signed an executive order with just one directive: government agencies must rebrand artificial intelligence as “Super Intelligence.” One month on, the rebrand is slowly appearing across government websites and documents, but it hasn’t caught on with everyone. Even some of the biggest names in tech still use “AI.” But the bigger concern is the new term becoming a political loyalty test: Those who use it signal allegiance to the administration. Those who don’t become targets for attacks. So one month in, who’s on board? —James O’Donnell This story is from The Algorithm, our weekly AI newsletter. Sign up to receive it in your inbox every Monday. The must-reads I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology. 1 Today’s midterms will test AI’s influence on US politicsCandidates are betting that AI regulation will sway voters to their side. (Axios)+ Angry constituents could derail Elon Musk’s Republican bets. (BI)+ Tech investors are on edge about the results. (Reuters $)+ America is divided over AI regulation. (MIT Technology Review) 2 Nvidia is ordering data centre clients to limit their business with China It told providers that the majority of their sales must be non-Chinese. (FT $)+ The move follows pressure from Washington. (Bloomberg $)+ Chinese AI has ruptured the White House. (MIT Technology Review) 3 Hackers have turned supermarket fridges into a food security riskAttacks in Europe have put US supermarket systems on alert. (CNBC)+ They’ve also sparked concerns over other domestic appliances.(BBC)+ It’s a good time for doing crime. (MIT Technology Review) 4 Scientists have detected signals at the smallest scale ever exploredThe discovery offers clues to a world of exotic new particles. (BBC)+ It came via CERN’s Large Hadron Collider. (The Times $) 5 Zapping part of the brain makes you think harder—at a costIt can both improve performance and reduce motivation. (New Scientist $)+ Neuropixels probes are transforming how we study the brain. (Nature) 6 Big tech is splurging on mega gas plants to power data centersThe projectsthreaten to reverse years of climate pledges. (WP $) 7 Google says an Android partner infected 1.3 million devicesThe malware infected TV boxes and other Android-based devices. (Verge) 8 Apple is trying to create its first robotIt’s proving a lot harder than expected, though. (Information $)+ AI robots are still a long way from useful. (MIT Technology Review) 9 China’s biggest shopping festival has a new product to sell: AIAI tokens have become a major new Singles’ Day category. (Reuters $) 10 A meteorite in the Sahara has revealed new clues about MarsIts chemistry may help explain how the planet lost its water. (Wired $) Quote of the day “Don’t spend money unduly by asking a human expert to enter into a debate with a robot expert.” —Steven Benmor, a family lawyer, tells the Wall Street Journal that clients increasingly need to pay human experts to correct advice they’ve received from chatbots. One more thing The noise we make is hurting animals. Can we learn to shut up? As human society has expanded, animals have started struggling to hear one another. For many birds, the noise has grown so loud that they’ve begun to sing with faster trills. Now, their mating calls aren’t as effective.  The growing hubbub can also increase bird-on-bird conflict, and entire species that can’t handle urban clamor simply leave town for good. But there are technological solutions to the noises hurting animals—and they could help humans, too. Read the full story. —Clive Thompson We can still have nice things A place for comfort, fun, and distraction to brighten up your day. (Got any ideas? Drop me a line.) + Watch five newborn cheetah cubs grow up on the Smithsonian’s livestream.+ Fifty years after Songs in the Key of Life, Stevie Wonder has released four lost tracks from the album.+ 50 air guitarists shredded absolutely nothing at the 30th Air Guitar World Championships in Finland.+ The Ocean Photographer of the Year winners show both the beauty of marine life and the perils it faces.

In 2001, a decade after he invented the World Wide Web, Tim Berners-Lee articulated a new vision. Together with two colleagues, he published an article in the magazine Scientific American in which he outlined what he called the “Semantic Web”: an internet designed to be navigated by autonomous software agents that could address our every need, without requiring us to muck about with browsers and search engines ourselves. To illustrate what this might look like, Berners-Lee and his colleagues told a story about a pair of siblings who use their agents to automatically research and book physical therapy appointments for their mother, making sure they would be covered by insurance and fit their busy calendars.  At the turn of the millennium, this vision seemed far-fetched: Google had only been online for a couple of years, and most people were still on dial-up connections. But a quarter-century on, the dream of autonomous software agents that use the internet and carry out tasks for us has finally been realized. New tools such as Meta’s blockbuster Muse agent, OpenAI’s Dots, and Anthropic’s Claude Duet are being given access to people’s email inboxes and bank accounts,  Meanwhile, in a series of worrying incidents, agents created by OpenAI broke free of their sandboxes and caused havoc online.   Just as the rise of smartphones sparked the development of a more atomized and attention-grabbing internet, the advent of powerful AI systems heralds a new era: We’re facing not the semantic web but the agentic web. This new web is developing out of sight and in ways that might not be entirely within our control. As it grows, the agentic web will inevitably transform the human web—and whether the human web survives this transition is anyone’s guess. For Berners-Lee’s original vision to work, internet users would have had to annotate the pages they built with special tags so that a software agent could understand each element’s meaning—say, “street address” for the part of a physical therapy office’s web page that indicates its location and “appointment types” for the list of services. And so the idea never really caught on. “It’s a lot of labor,” says Eni Mustafaraj, a professor of computer science at Wellesley College. “And the benefits of that were not obvious.” For the next two decades, then, the internet remained a place that put humans first, with the odd Twitter reply bot or web scraper to round things out. Large language models changed that. Unlike the agents that Berners-Lee was imagining, today’s LLM-based agents don’t need any tags to understand which part of the text on a clinic’s site is its address and which details the services it offers. They can search for and process text far faster than any human, and agents with multimodal capabilities can even interpret screenshots of websites and click on specific elements such as hyperlinks and drop-down menus.  They’re also just a way more intuitive means of finding information and accomplishing tasks online. Rather than poring over pages of Google results to find a budget-friendly product dupe or a half-remembered meme, why not have Claude or ChatGPT peruse dozens of sites in just a few minutes, guided only by your vague query? Or go one step further and use a personal assistant agent like Muse to hunt down a restaurant reservation or book a flight.  As agents flood online spaces, this new user base is changing the web for good. Time magazine, for instance, is publishing versions of its articles that are easier for agents to parse and has started selling ads targeted at them. Companies from Etsy to Expedia have partnered with the major AI firms to build tools that allow agents to interface seamlessly with their services. Despite these early efforts, it’s still tough to tell exactly what forms the agentic web will take and how it will change our online experience. Depending on how it’s designed and who controls access to it, we could be in for something that’s as open and customizable as the web of the early 2000s, only far easier to use. Alternatively, we could be reduced to viewing much of the web through tiny portholes controlled by a few dominant platforms, as independent sites and creators struggle to survive the shift. And as this summer’s series of OpenAI security incidents suggests, the agents might even build their own online networks and resources—whether we want them to or not. How this plays out also depends on us: which parts of our online lives we’re happy to delegate to our agents and where we will still demand a human-centric internet. “There’s a lot we don’t know yet about how users want to use [agents],” says Ajit Varma, head of Firefox at Mozilla. But whatever does happen, the changes will be big. “It’s probably on par with getting a mobile phone,” he says. “I think things are going to change very dramatically.” Building a web for agents There are two ways to approach building the agentic web. The first involves training agents to use browsers like humans—scrolling, clicking, typing—and letting them loose on the web that we’re all used to. The second requires building tools that allow agents to interface with the internet via text and code, their preferred media. That possibility offers more separation between the human and the agentic web: designers can create different experiences for those who use text commands and those who use browsers. When Siva Reddy, a professor of computer science and linguistics at McGill University, began studying LLM agents in 2024, he focused on the first option. “At that time, we thought it should be that the best agent in the world is probably an agent that can use a browser,” he says. Such an agent would integrate seamlessly with the preexisting human web, and its owner could easily supervise it by watching as it navigated familiar interfaces. The frontier AI companies were already developing these “computer-use agents.” But they didn’t work terribly well—at least not at first. “There’s a lot

In this article, you will learn how agent memory works, what types of memory exist, and how different frameworks approach storage, retrieval, and memory management…

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. The agentic web is here In 2001, a decade after inventing the World Wide Web, Tim Berners-Lee articulated a new vision: an internet navigated by autonomous software agents that address our every need, without requiring us to muck about with browsers and search engines ourselves. At the time, the idea seemed far-fetched. But a quarter-century on, AI agents are beginning to make it a reality. They can triage your inbox, do your grocery shopping, and even access your bank accounts. This new era is being called the agentic web. It will inevitably transform the internet—and whether the human web survives this transition is anyone’s guess. Find out how AI agents are transforming the internet, and what we stand to lose in the process. —Grace Huckins Are we there yet? Eliminating traffic deaths Every year, traffic crashes kill more than 1 million people worldwide. They’re also the leading cause of death for children and young adults. Cheap, proven ways to make roads safer exist, including tightening intersections with planters and lowering speed limits. But government logjams and red tape can make it difficult to roll out these changes. Still, some cities are making impressive progress by investing in public transit, extending curbs, and using cameras to spot danger zones in need of a redesign. Here’s how close we are to eradicating traffic deaths. —Sara Kiley Watson This story is the final part of “Are we there yet?”, our series that takes stock of six technologies that have long promised to change the world—and scores them on a progress meter. Subscribe to read the other five instalments in our new Impossible issue. The must-reads I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology. 1 A $1 trillion tech selloff has deepened fears of an AI bubbleBig AI bills and weak earnings have spooked investors. (BBC)+ The Nasdaq 100 had its worst day since June on Friday. (NBC News)+ OpenAI has cut its funding target by $10 billion. (CNBC)+ What even is the AI bubble? (MIT Technology Review) 2 Hackers have hit Ukraine and Russia’s emergency alert appsKyiv residents were warned of “nuclear danger.” (NBC News)+ Russia’s app said nuclear warheads had been launched. (CNN)+ Ukraine has blamed Russia, which hasn’t yet accused anyone. (BI)+ AI will transform cyber warfare. (MIT Technology Review) 3 Binance users withdrew $3.6 billion after the largest ever crypto hackThe funds moved when the exchange reopened on Sunday. (Guardian)+ Binance says it’s covered $6 billion of the stolen $6.3 billion. (Reuters $)+ But the industry’s response has undermined its “decentralized” image. (Gizmodo) 4 An AI agent injected drugs into a live mouse’s brainIt chose the gene therapy and controlled the needle itself. (Wired $)+ We need real-time tracking of AI agents. (MIT Technology Review) 5 Data centres have made utilities a flashpoint in the midtermsAmericans are angry over rising electricity bills. (Reuters $)+ Virtual power plants could help. (MIT Technology Review) 6 Apple has overtaken Nvidia as the world’s most valuable companyInvestors are favouring its restraint on AI spending. (Bloomberg $)+ But critics say it’s entering its “slop era.” (404 Media) 7 Supermarket fridges have overheated in a new wave of cyberattacksStores in Belgium, Sweden, and Luxembourg were hit. (Politico) 8 The first chocolate bars made from lab-grown cocoa are hereStartups are cultivating chocolate to tackle a cocoa shortage. (WSJ $) 9 Hackers hijacked digital billboards in multiple countries to show pornThey also criticized security flaws and Egypt’s government.(Gizmodo) 10 A new 3D map reveals instability in interstellar spaceThe discovery could help explain why stars form so inefficiently. (Quanta) Quote of the day “It would be great if all this was just a house of cards.” —Kevin Roose, a technology columnist at the New York Times, tells NPR that an AI investment crash would be preferable to the technology permanently upending work and society. One more thing Useful quantum computing is inevitable—and increasingly imminent Last year, Nvidia CEO Jensen Huang jolted the stock market by saying that practical quantum computing is still 15 to 30 years away. He also suggested that those computers would need Nvidia GPUs to function. But Huang’s predictions miss the mark—both on the timeline and the role his company’s technology will play.  Quantum computing is rapidly converging on utility. And that’s good news, because the hope is that they will be able to perform calculations that no amount of AI or classical computation could ever achieve. Read the full story. —Peter Barrett We can still have nice things A place for comfort, fun, and distraction to brighten up your day. (Got any ideas? Drop me a line.) + A miniaturist has built a delightful diorama of the world’s creepiest living room.+ The year’s best dog photos have been announced, and they’re very good boys and girls.+ At the Green Zone golf club, you tee off in Sweden and putt in Finland, an hour in the future.+ Astronomers say they’ve found the real celestial arrangement that inspired van Gogh’s The Starry Night.

When a developer gives an AI agent a coding task, the agent does more than answer once. It plans, generates code, runs tests, and repeats the process to correct errors. A single task can use 10 to 50 times more tokens than a basic chat prompt. The more useful the agent becomes, the more calls it makes. With per-token pricing, the workflows creating the most value can also become the most expensive. The financial pressure is already visible. Uber reportedly exhausted its entire 2026 AI budget by April after expanding its use of Claude Code. GitHub Copilot moved from flat-rate pricing to metered credits. Gartner predicts that AI coding token costs will surpass average developer salaries by 2028. The natural response is rationing usage caps, seat limits, and approvals before developers launch agents. But blanket limits can restrict the workflows delivering value. They also optimize the wrong number. Cost per token measures consumption, not results. Some spending reflects waste, including retries and runaway loops. Some represents the work of shipping software. A meter cannot distinguish them, and neither can a spending cap. Measure the software that ships A better measure is cost per merged pull request: total AI-assisted development cost divided by the number of changes that pass review and are merged. Track rework and defects alongside it so cheaper changes do not conceal lower quality. If that cost falls as AI usage rises, without compromising quality, developers are converting more consumption into useful software. The question is not how many tokens they use, but how much finished work the investment produces. Improving that equation requires two changes: reserve expensive inference for tasks that need it, and lower the cost of routine work. Match the model to the task Agentic coding does not require the most powerful model at every step. Use a frontier model, or an on-premises flagship when policy requires it, for difficult planning and novel reasoning. Route suitable work, including code generation, test creation, and repeated refactoring, to open-weight models you operate yourself. A gateway can route requests according to task complexity, data sensitivity, cost, and latency. This reserves premium inference for the moments that justify it rather than applying premium pricing to an entire workflow. Meter the exception, not the workload. Open-weight models make this division increasingly practical. Software-engineering benchmarks show that leading open-weight models can approach frontier performance on some tasks. That does not establish equivalence across every repository or workflow. Evaluate candidates against your own code, tests, and acceptance criteria before assigning production work. Change the economics of repeated use Next, decide who owns the capacity. With rented inference, every call adds to the bill. With owned capacity, most of the cost is committed upfront, so each additional call gets cheaper as usage grows. The economics are simple: per-token pricing makes heavy use more expensive, while owned capacity becomes more efficient as teams use it more. Ownership is not free, though. Power, operations, maintenance, and depreciation all must be considered. The number that matters is the fully loaded cost per merged pull request. Owned capacity also delivers a type of value that never appears on an invoice: control over sensitive code. When code generation and testing run on-premises, proprietary code stays inside the enterprise. Only approved planning prompts and governed exceptions go to external models, through a gateway the organization controls. That opens the sensitive repositories where security usually stalls adoption, and where the biggest gains are waiting. Test the operating model At HPE, a hybrid AI platform routes routine workloads from more than 150 internal applications to open-weight models running on premises, including on HPE Private Cloud AI. The result: token costs down 30 to 50%. That is the point of hybrid. With HPE Private Cloud AI, owning your inference does not mean giving up frontier models. Run the volume on capacity you own and let a gateway send the hardest work. Data also shows that usage, not licensed headcount, drives spend. The top 5% of users generated most of the consumption, and the heaviest used roughly 450 times more than the lightest on the same license. If power users grow from a small minority to half the workforce, monthly spend can rise about 4.4 times with flat headcount. Your mix will differ, which is exactly why you need to measure it. HPE tracks two metrics and recommends customers do the same. First, it tracks cost per active developer, because averaging across all licenses hides the power users driving spend. Second, it tracks cost per merged pull request, because it connects AI spending to software that ships. Baseline both costs before changing infrastructure. Then run a complete development cycle on owned capacity. If both metrics do not improve as usage grows, your current model may still be the better choice. Build for adoption, not rationing Owned capacity is not for everyone. Low or unpredictable usage favors APIs, with no capital tied up in idle infrastructure. Ownership pays off when sustained demand keeps it busy. Most enterprises will run hybrid: frontier APIs for demanding or intermittent work, owned capacity for repeatable volume. A governed gateway decides what leaves the building, so you get frontier models without giving up sovereignty over your code. As developers lean harder on coding agents, usage will keep rising. The real choice is whether that growth shows up as a bigger variable-cost bill or as more software shipped on capacity you own. The companies that lead AI-assisted development will not ration adoption to control costs. They will build an operating model where more AI use improves the economics instead of undermining them. To learn more about HPE Private Cloud AI, please visit HPE. This content was produced by HPE. It was not written by MIT Technology Review’s editorial staff.

In just a few years, generative AI has changed the face of enterprise analytics, notably by making analytics accessible to users across the organization. Agentic AI promises to do even more, by automating entire workflows and taking actions based on agent insights. But there is a problem: The quality of insights that agents currently generate leaves much to be desired. Fewer than half of agent-generated insights at the organizations surveyed for this report lead to business improvement. And while most respondents praise the quality of their agents’ descriptive insights, reviews of their prescriptive insights—those on which actions are based—are decidedly mixed. DOWNLOAD THE REPORT Failure to produce insights that improve the business deprives organizations of their desired return on investment in agentic analytics. Worse, flawed insights that trigger incorrect actions can cause reputational or financial damage. This report gauges how companies view the quality of the outputs produced by their analytics agents and explores how organizations are seeking to boost insight quality and get their agents working harder for the business. Some key findings from the report include the following: Agentic analytics is only just starting to deliver. On average, 48% of insights generated by analytics agents at the surveyed companies lead to actions that improve the business. That is a marked advance from the 37% average the respondents say generated such actions one year ago. But it means that more than half of the insights generated by analytics agents today are still failing to help businesses. Better quality of insights leads to clear business advantage. At a small cohort of organizations, which we term “insight leaders,” AI agents perform much better than the average. In this group, more than 70% of agent insights lead to actions that improve the business. The insight leaders also give the quality of their agents’ insights higher marks than do the total respondents overall. The quality gap is particularly wide when it comes to prescriptive insights. Insufficient context reduces insight quality. When analytics agents deliver insights of substandard quality, the main reason is a lack of business context, as cited by 55% of surveyed executives. Fragmented or inaccessible data is the other major factor, according to 45%. Fewer insight leaders say their agents struggle with access to context. Insight overload further inhibits agent effectiveness. At a majority (52%) of surveyed organizations, analytics agents generate too many insights to be properly evaluated and therefore acted upon. Less than one-quarter (24%) are able to evaluate all agent insights; most do so on a case-by-case basis, and that process is currently human-intensive. Governance and semantics are areas of focus for improving insights. The surveyed companies are prioritizing initiatives that enable better agent insights by strengthening AI and data governance (cited by 60%). They are also adding a semantic layer to their data architecture (45%). The experts we interviewed agree that the semantic layer is crucial to ensure high-performing agents. Download the report This content was produced by Insights, MIT Technology Review’s custom content arm, not its editorial staff. It was researched and written by humans, with any AI tools that may have been used limited to production processes under human oversight.