Breaking AI & Tech News Analyzed

The Rise of Recursive Self-Improvement in AIThe word "recursion" is the latest buzzword in AI circles. Two separate startups have taken on the name, and many more have started referencing recursive self-improvement (RSI) in their roadmaps. Like AGI before it, RSI has become a three-letter byword for a cataclysmic AI takeoff – even if there's still a little disagreement about what it exactly means.In basic terms, RSI refers to an AI system that can continuously upgrade itself. Once AI systems can manage the upgrade cycle better than humans, the process can become a closed loop, limited only by the compute power they can access, and humans are no longer necessary or even helpful.Scary or not, that's a vision that a lot of AI labs are eager to chase.Key Players Pursuing Recursive SystemsEarlier this month, well-known AI researcher Richard Socher launched the aptly named Recursive Superintelligence with RSI as an explicit goal. "Our main focus is to build truly recursive, self-improving superintelligence at scale," Socher told TechCrunch at launch, "which means that the entire process of ideation, implementation, and validation of research ideas would be automatic."A number of other prominent researchers are already chasing that same goal, hoping for a breakthrough that will make recursive self-improvement possible.One of the most prominent is Andrej Karpathy, a legendary figure from Tesla and OpenAI, who is using agent swarms to train LLMs on simple tasks for a project he calls Auto-Research. Karpathy has been unusually open about the project, tweeting about milestones regularly and making the building blocks available through a public GitHub repo. So far, the work has mostly been confined to making minor improvements on a GPT-2 scale model — as Karpathy noted in March, "It's not novel, ground-breaking 'research' (yet)" — but it's been enough to convince lots of other researchers to follow the RSI dream. And with Karpathy now working on pre-training at Anthropic, he will have plenty of opportunity to apply the idea at a larger scale.Adaption — founded by Cohere and Google alum Sara Hooker — recently launched a similar tool called AutoScientist in an effort to automate frontier training. Like Karpathy's auto-researchers, the system trains agents to make incremental improvements — but for Adaption, the goal is to make it easier to train a full-scale frontier model. If those same researchers start to push the frontier forward, the system could quickly spiral into something very much like RSI.Disarray founder Doris Xin drew more specific RSI interest when her self-trained machine learning agent took home 28 medals in a recent Kaggle competition, beating out many human-trained agents. As she sees it, the major challenge is reliability."I would argue, given infinite compute and infinite time horizon, we are already there," Xin told me. "I want to make an argument that this is not a creative endeavor, really. It's just a lot of meat-and-potatoes engineering."The Current State of Self-Improving AIThere's also plenty of evidence that the AI industry isn't very close to recursive systems in any meaningful way — and is still grappling with talking to a wary public about its progress. So Google CEO Sundar Pichai basically admitted in a recent podcast interview."It's a continuum, and we are all definitely making progress," Pichai said. "But in the way people describe RSI, that would represent a next level of acceleration and would have a lot of implications, but we aren't quite there yet."But the continuum includes an awful lot of self-improving AI systems.In January, one of Anthropic's lead programmers for Claude Code estimated that "close to 100%" of his team's code was written by the tool — a frank admission that Claude Code was literally writing itself.Just because engineers are using an AI tool doesn't mean the tool can replace them — but Anthropic seems to be getting close to replacing engineers too. In a recent survey tied to the Mythos preview, five out of 18 Anthropic engineers believed that, with harness improvements, this version of Mythos could soon substitute for an L4 engineer — a midlevel programmer who can take on involved projects without supervision.Still, there were some of the same weaknesses you might expect."Some of Claude's major reported weaknesses compared to an L4 include: self-managing week-long ambiguous tasks, understanding org priorities, taste, verification, instruction-following, and epistemics," the report reads.In other words, its weaknesses are everything involved with self-direction, which is the cornerstone for RSI. But sure, for everything else, Claude is ready to step right in.Expert Perspectives on RSI TimelinesJust like the AGI term before it, the AI industry also can't tell us how far away it is from showcasing a meaningful recursive system. When Georgetown's Center for Security and Emerging Technology assembled a group of experts to study RSI last year, the group found a major split in assessments — some expecting an imminent "superintelligence" style explosion while others expected slower progress and an eventual plateau. But all agreed that recursion made the future especially difficult to predict.Helen Toner, director of CSET and a former board member at OpenAI, told TechCrunch that simply using AI tools to do AI research isn't enough to qualify as RSI. "They're just using AI for as much as they can," Toner told TechCrunch. "And I think that is different from the classic definition of RSI, which is really that there are no humans needed."Toner pointed to a recent post by METR's Ajeya Cotra, which distinguishes different milestones on the path to the AI research takeover. One step, which Cotra calls "adequacy," would come when the system can still perform research after all humans are removed — even if the resulting research isn't as valuable or efficient. "Parity" comes when an AI-only system is as good at research as a human-only system. "Supremacy," the final stage, comes when an AI-only system outperforms a collaborative system between humans and AI.Ultimately, Cotra concludes that AI is very close to the adequacy threshold of being able to produce some work on its own — similar to the incremental changes made by Karpathy's Auto-Research system. "I wouldn't be totally shocked if you told me this milestone had already passed, and I expect it to happen in the next couple years," Cotra wrote.She was less clear on when parity will come, but once it does, she thinks it would "massively accelerate the pace of AI progress, leading to AI research supremacy within another year."The Challenges Ahead for Recursive AIWith so much of AI built on scaling laws, there's a strong tendency to think RSI will follow the same curve. Toner thinks that many of those pursuing AI research and development via RSI "think of it as a pretty smooth ladder, where you can just keep scaling up."But even if AI researchers are able to make incremental improvements like Karpathy's auto-researchers, there will be larger challenges in handing off the whole process of research. Toner put it in terms of the history of computing, which has seen human beings handing off more and more of the process while still directing things from the top."We went from machine languages to assembly language and compiled languages; you're getting further and further from the guts of the computer," Toner said. "But the human is still, in some intuitive sense, running the show."Moving beyond that paradigm will take significant challenges, both in engineering and alignment. But even with the massive investments happening, there's no infinite compute available — and the basic trade-off between human labor and machine intelligence will be hard to overcome.The Future of Recursive Self-ImprovementAs for a total recursive AI system of apocalyptic visions? The only thing researchers essentially agree on is that, like AGI, it's not here yet.

The Lead Enterprise organizations are not rejecting AI. They are rejecting operational instability. This shift is becoming a defining reality for enterprise AI companies that scale versus those that stall after early momentum. The Event Details At TechCrunch Disrupt 2026, taking place October 13–15 at Moscone West in San Francisco, Arsalan Tavakoli-Shiraji, co-founder and SVP of field engineering at Databricks, will discuss this shift during his AI Stage session, “The Enterprise Isn’t Broken. Your Assumptions About It Are.” The Data Analysis The enterprise AI market is full of successful pilots that never became real deployments. Not because the technology failed, but because the organization could not absorb the operational consequences of adopting it. Databricks and other AI startups gaining traction inside large organizations increasingly share one thing in common: They reduce uncertainty. The Impact Analysis Enterprise buyers are asking different questions now. Concerns are no longer secondary; in many organizations, they have become core to the buying decision itself. For AI founders selling into the enterprise, understanding how technical systems interact with organizational behavior, infrastructure realities, procurement processes, governance concerns, and operational risk is crucial. The Prediction The startups that succeed in enterprise AI over the next several years may not necessarily be the ones with the most advanced models. They may be the ones that best understand how enterprises actually absorb change. The market is maturing, and enterprise AI success increasingly depends on more than strong engineering alone.

YouTube announced on May 28, 2026 that its Premium service will soon include an AI‑powered podcast recommendation tool, an “Auto speed” playback feature, and an on‑the‑go listening mode, aiming to deepen engagement with its growing podcast audience. AI‑Driven Podcast Recommendation Engine Launches The new recommendation tool leverages the same generative AI behind YouTube’s "Ask Music" to suggest podcasts based on genre, listener mood, or shows already enjoyed. Premium users will see personalized suggestions directly in the Podcasts tab, streamlining discovery without leaving the app. Auto Speed Playback and On‑the‑Go Mode Arrive on Android First Auto speed: Dynamically adjusts playback speed during slower speech or dense segments, preserving comprehension while reducing total listening time. On‑the‑go mode: Adds quick‑skip controls, episode‑jump shortcuts, and background‑play optimization for activities like running or commuting. Both features are live for Premium users on Android and will roll out to iOS in the coming months. Premium Podcast Consumption Metrics Highlight Growth Potential Premium users logged over 800 million hours of podcast playback in April 2026. YouTube Podcasts now boasts more than 1 billion monthly active users. The platform’s "Ask Music" already powers personalized radio stations, indicating a ready AI infrastructure for podcast recommendations. Strategic Play to Capture Audio‑First Audiences By enhancing discovery and hands‑free listening, YouTube is positioning itself against established audio platforms such as Spotify and Apple Podcasts, while also responding to Netflix's recent push into video podcasts. The focus on AI personalization and adaptive playback reflects a broader industry shift toward seamless, user‑centric audio experiences. What This Means for the Future of Podcast Platforms Analysts expect the AI recommendation engine to increase user retention, potentially driving Premium subscription growth by double‑digit percentages over the next year. If the Auto speed feature delivers measurable time‑saving benefits, it could set a new standard for intelligent playback, prompting competitors to develop similar adaptive technologies. The on‑the‑go mode further blurs the line between video and audio consumption, suggesting that YouTube will continue to integrate podcasting deeper into its core ecosystem.

Visa has disclosed an undisclosed investment in AI coding platform Replit, aiming to embed its payment suite directly into the developer environment so that both developers and AI agents can accept payments without leaving the platform. Strategic Investment and Joint Exploration of AI‑Powered Payments The two companies are testing how Visa Intelligent Commerce and the Trusted Agent Protocol can be woven into Replit’s workflow. More than 1,000 Visa employees already use Replit for prototyping, and the collaboration remains in an exploratory stage with no formal product announcements. Valuation Surge and Funding Milestones Highlight Replit’s Growth September 2025: Replit reached a $3 billion valuation. March 2026: Raised $400 million in a Series D led by Georgian Partners, pushing valuation to $9 billion. Enterprise self‑serve contracts now allow deals up to $200,000 without sales interaction. Customer churn is described as "very, very low" with net retention hitting 300 % in some cases. Implications for the Emerging Agentic Payments Ecosystem The move underscores a broader race to build infrastructure for "agentic payments," where AI agents transact on behalf of users. Competitors such as Robinhood (agent‑driven trading) and Google (shopping agents) are pursuing similar capabilities, suggesting the market will soon demand secure, verifiable AI‑mediated transactions. Future Trajectory: From Prototype to Mainstream Agentic Commerce If the exploratory projects mature, Replit could become a one‑stop shop for developers to build, host, and monetize AI agents, accelerating adoption of Visa’s Trusted Agent Protocol. Analysts anticipate that as enterprise adoption grows and churn remains low, the partnership may evolve into a commercial product suite within the next 12‑18 months, positioning Visa and Replit at the forefront of the next wave of AI‑driven commerce.

The Final Days of Early Bird Pricing Time is running out to secure discounted tickets to TechCrunch Disrupt 2026. Early Bird pricing ends tomorrow, May 29, at 11:59 p.m. PT. After that, prices for the highly anticipated tech conference will increase. Unlock Savings of Up to $410 By registering now, you can lock in savings of up to $410 on your pass or up to 30% on group passes of 4+. Why Attend TechCrunch Disrupt 2026? TechCrunch Disrupt 2026, taking place from October 13–15 at San Francisco’s Moscone West, is a premier event for startups, investors, and tech enthusiasts. Here’s what you’ll gain by attending: Founder Pass: Accelerate growth with the right insights, tools, and connections. Meet investors aligned with your startup. Investor Pass: Discover standout startups and expand your portfolio with curated access. Use matchmaking tools to make every conversation count. Don’t Miss Out The window to the lowest ticket rates of the year is closing at 11:59 p.m. PT tomorrow, May 29. Register now to secure your ticket with up to a $410 discount.

General Compute, a new inference neocloud, raised a $15 million seed round at a $60 million post‑money valuation and booked a $300 million order for SambaNova’s upcoming SN50 chips. The company promises 600‑700 tokens per second per chip and a deployment model that fits into existing, air‑cooled data‑center infrastructure. General Compute’s Funding and Strategic Partnerships Seed round led by FUSE VC with participation from Carya Venture Partners and Village Global Ventures. Co‑founders Finn Puklowski (CEO) and Jason Goodison (CTO) partnered with SambaNova, an Intel‑backed chipmaker focused on inference. General Compute will be the first neocloud to deploy SambaNova’s SN50 chips, ordering $300 million worth of hardware. Colocation strategy includes traditional data‑center providers and repurposed crypto‑miner facilities. Financial Snapshot: $15 Million Seed and $300 Million Chip Order Seed funding: $15 million raised, valuing the company at $60 million post‑money. Chip commitment: $300 million of SN50 chips on order, enough to power a large inference fleet. Comparable market moves: Nvidia’s $20 billion acquisition of Groq (Dec 2025) and Cerebras’ $57 billion IPO (May 2026) illustrate the scale of inference‑focused investments. Implications for the AI Inference Landscape The shift from GPU‑centric training to specialized inference hardware is accelerating. SambaNova’s memory‑rich, flexible architecture claims to outperform GPUs, Groq, and Cerebras on token‑throughput, delivering 600‑700 tokens/sec versus ~250 tokens/sec for GPUs. Air‑cooled, low‑power chips lower the barrier to entry for colocation, enabling rapid deployment in existing facilities and even in repurposed crypto‑mining sites. This could democratize high‑speed inference, pressure pricing, and spur a wave of niche cloud providers focused on agent‑to‑agent workloads. What the Next Year May Hold for Inference‑First Cloud Providers When SambaNova releases its next‑gen chips later in 2026, General Compute’s early access positions it to capture a sizable share of the fast‑inference market. Expect: Increased competition among inference‑only clouds (e.g., CoreWeave, OpenRouter) to offer multi‑model routing and token‑cost optimization. More venture capital flowing into inference‑focused startups, mirroring the recent $113 million Series B for OpenRouter. Potential consolidation as larger players (Nvidia, Intel) seek partnerships or acquisitions to secure the most efficient inference stacks. Speed and cost efficiency will become the primary differentiators, shaping the architecture choices that dominate the AI future.

The Lead: Vertu's AI-Powered Foldable Targets Executive Market Luxury smartphone brand Vertu has unveiled the Alphafold, a foldable phone powered by an AI agent designed specifically for executives managing business operations on the move. The device represents Vertu's latest attempt to reinvent itself for the AI era, combining luxury materials with enterprise-focused AI capabilities to target the high-end business market. The Event Details: Luxury Meets AI: The Alphafold's Enterprise Capabilities The Alphafold features Hermes Agent, built on the open-source Hermes project by Nous Research, which can connect to enterprise systems like ERP and CRM. The AI agent coordinates tasks such as approvals, scheduling, sales tracking, travel planning, and operational reporting through natural-language prompts. The device can route requests across multiple AI models including OpenAI's GPT, Anthropic's Claude, Google's Gemini, and selected open-source models, while integrating with more than 80 apps and dozens of native phone functions for cross-platform workflows. Vertu has emphasized the device's privacy-focused architecture featuring a proprietary A5 security chip designed to isolate authentication keys, biometric credentials, and sensitive enterprise information from the main operating system. The company states that commercially sensitive data can be processed locally on the device, while prompts sent to external AI models are redacted or tokenized before leaving the phone. The Data Analysis: Premium Pricing Strategy in the Smartphone Market The Alphafold starts at $6,880 for the calfskin version, with higher-end models featuring bespoke finishes including alligator leather, 18K gold, and natural diamond accents. Vertu's highest-end standard model is currently priced at $46,800, with further customization options available. This pricing strategy positions Vertu firmly in the ultra-premium segment of the smartphone market. While foldable smartphones remain a niche segment globally—with IDC data showing approximately 20 million units shipped in 2025, accounting for less than 2% of total smartphone shipments—Vertu is betting that the combination of luxury materials and AI capabilities will justify its premium pricing. The average price of foldable smartphones was about $1,300 last year, roughly three times the price of non-foldable smartphones. The Impact Analysis: How AI is Transforming Executive Productivity Vertu CEO Molly Ma highlighted that existing AI features on smartphones from major manufacturers remain focused largely on consumer tools such as image editing and voice assistance, leaving room for more advanced AI-agent workflows tied to enterprise systems. The Alphafold aims to address this gap by providing executives with a device that can seamlessly integrate with their business operations and workflows. The device's larger foldable display (8.05-inch inner screen and 6.53-inch outer screen) is better suited for multitasking and productivity-oriented experiences, according to Kiranjeet Kaur, associate research director for mobile phones research at IDC. However, she noted that enterprise AI adoption on smartphones still lags behind computers, with most enterprise smartphone decisions continuing to be driven by ecosystem integration and device management support rather than AI capabilities. The Prediction: The Future of Luxury AI-Powered Mobile Devices The Alphafold represents Vertu's significant step forward from its previous AI-focused device, Agent Q, with Ma noting that AI-agent technology has matured rapidly over the past year, with improvements in memory, automation, and app integration. While the company has not yet undergone third-party security audits for the device, it has confirmed that independent audits and certification remain on its security roadmap. As the first 115-unit batch of Vertu's Alphafold begins shipping across major markets including the U.S., the device will serve as a test case for whether there's a market for luxury smartphones with enterprise AI capabilities. If successful, Vertu's approach could inspire other manufacturers to develop similar devices targeting the executive market, potentially accelerating the integration of AI agents into mobile workflows.

Google’s AI Overview Stumbles on Simple Letter Counting Google’s newly rolled‑out AI Overview feature in Search incorrectly counted letters in everyday words – claiming there are two “P”s in “Google”, one “r” in “poop”, and even misspelling “journalism”. The blunders highlight a long‑standing weakness of large language models (LLMs) when it comes to exact spelling. The Miscounted Letters Behind the New Search AI “Google” – AI said 2 Ps (actual: 0) “poop” – AI said 1 r (actual: 0) “journalism” – AI said 2 d’s (actual: 0) U.S. President’s last name – AI reported 1 P but rendered “t‑r‑p‑u‑m” Quantifying the Miscounts: Numbers Behind the Errors Beyond the anecdotal examples, the AI also produced a faulty definition for the word “disregard”, responding with “Understood. Let me know whenever you have a new prompt or question!” This illustrates that token‑based encoding can produce nonsensical outputs even when the input is a single word. Implications for Search Trust and AI Adoption Google’s AI‑driven overhaul aims to make generative responses the centerpiece of its 29‑year‑old search product. Repeated factual and spelling errors risk eroding user confidence, especially after earlier AI Overviews cited satirical sources and gave absurd advice such as “eat rocks”. Trust in AI‑generated answers remains a critical hurdle. What’s Next for Google’s Generative Search? Google told TechCrunch it is “working to fix this particular issue” and will likely refine its tokenizer and post‑processing pipelines. Industry observers expect incremental improvements rather than a complete architectural shift, meaning users may continue to see occasional glitches while the broader AI‑search strategy matures.

The Strategic Shift Toward Custom Silicon Snowflake's decision to deepen its reliance on AWS is driven by the explosive demand for AI processing power. The deal specifically targets AWS's proprietary Graviton ARM-based CPUs, which are increasingly vital for the inference and agent phases of AI workflows that GPUs cannot handle alone. By integrating Snowflake's Cortex AI tool, the partnership aims to streamline data operations, allowing enterprises to query databases using natural language and generate automated reports more efficiently. Financial Implications of the AI Boom This contract represents a massive financial milestone. While AWS has generated $7 billion from Snowflake since 2012, this new deal brings the total value to nearly the same level in a single contract. Furthermore, Snowflake reports that AWS spending has doubled in 2025 to $2 billion annually, highlighting the rapid monetization of AI tools. This data confirms that enterprises are aggressively accelerating their cloud spending to stay competitive in the generative AI era. Disruption in the AI Chip Market The move signals a broader trend where cloud providers are weaponizing their own hardware to undercut Nvidia. By offering "better price-performance," AWS aims to capture market share from Nvidia, a strategy already seen with Meta. This creates a bifurcated market where companies can choose between Nvidia's training dominance and AWS's cost-effective inference capabilities. The reliance on Graviton chips offers a more affordable option for cloud providers, allowing them to pass savings directly to customers. The Future of the AI Compute War As AI agents become more prevalent, the demand for high-performance CPUs will skyrocket. We can expect more multibillion-dollar contracts like this one, forcing Nvidia to innovate aggressively with its own Vera chip. The cloud giants are effectively building their own ecosystems, making it harder for third-party hardware vendors to maintain a monopoly. The winners in this space will be the companies that can optimize their data infrastructure for the specific chips they are using.