Join us for the first edition of Virtual Lunch & Hack, a biweekly online event where builders, founders, engineers, and AI enthusiasts come together to learn, build, and connect.

Each session starts with a short talk from industry leaders, followed by a hands-on hacking session where you'll build something in real time alongside the community. Whether you're an experienced developer or just getting started with AI tools, you'll leave with new ideas, practical skills, and something you've created yourself.

For our inaugural event, we'll explore how to improve AI coding workflows by hacking Claude Code memory with RoBrain.dev, an open-source shared memory system designed to help AI agents understand decisions, context, and prior work across sessions.

We'll wrap up with live demos from selected participants, and a few builders will win lunch on us, sponsored by RoryPlans.ai

On June 4, 2026, Anthropic published When AI Builds Itself, arguing that AI is already accelerating its own development and that the trend points toward recursive self-improvement, the point at which systems autonomously design and train their own successors. Anthropic engineers now ship 8× as much code per quarter as they did from 2021 through 2025, and Claude authors more than 80% of the code merged into the company's codebase, up from low single digits before Claude Code launched in February 2025. The company argues that a global ability to slow or pause frontier development would likely be good for the world, and that it would pause its own work if rival labs verifiably did the same. On whether models will build their successors, the post in reality commits very little.

The post arrived three days after Anthropic filed a confidential S-1 with the SEC on June 1, days after a $65 billion round set the company's valuation at $965 billion.

Researchers care about that sentence more than any revenue figure because recursive self-improvement is the mechanism behind the intelligence explosion that British mathematician I.J. Good described in 1965, in which a machine smart enough to build a smarter machine triggers a feedback loop that compounds toward superintelligence. A self-improving system that cleared the threshold could condense a decade of progress into a year, a takeoff fast enough that institutions would have no window to react.

Yoshua Bengio, Geoffrey Hinton, and Ilya Sutskever, the three most-cited researchers in the field, have all called that scenario credible, and it is chiefly their belief that is the foundation the capital sits on. The trillion-dollar valuations attached to Anthropic, OpenAI, and their peers are not underwritten by chatbot subscriptions but by the possibility that one of these labs lights the fuse first.

The evidence in the post needs to be exact for that bet to hold, and Anthropic concedes the central weakness itself. Lines of code measures quantity over quality, the company writes, so the 8× figure overstates the true productivity gain. Plus, Claude Code itself was built with Claude from the beginning, and these labs employ the most aggressive engineering talent on the market, so an unknown share of the acceleration is the most heavily resourced engineering organization in history getting better at using its own product rather than a clean leap in capability.

The external evidence carries a similar crack. Anthropic leans on METR's time-horizon metric, which measures the length of task a model can complete reliably and reports it doubling roughly every four months, but METR warns that its estimates above 16 hours are unreliable because the task suite is saturating, and the metric can swing on a handful of tasks at the high end, exactly where the steepest part of the projected curve lives. If the acceleration is anything short of the clean exponential the post implies, the distance between what is verifiably true and claimed is where skeptics will double down.

And by all accounts, the critics make some interesting arguments.

Bill Gurley of Benchmark fame, who said he spent a month reading everything Anthropic publishes, concluded that the company is not building software but "midwifing a deity," a Dr. Frankenstein impulse he finds more frightening than the familiar charge that Anthropic's safety advocacy is a strategy to lock out competitors. The deity framing is overcooked, but the structural reading beneath it has independent support.

Meanwhile, technology analyst Rob Enderle argued that the productivity gains are probably legitimate while promoting progress toward recursive self-improvement is a calculated move, one that positions Anthropic to investors as the bleeding edge and seeds the narrative that its technology is so advanced it requires ever-larger funding to manage safely.

Scientific American reduced the contradiction to a single observation, that Anthropic has called for the brakes in a race where it remains a front-runner. A pause that no rival has agreed to costs the company nothing, signals everything, and went out the same week as a near-trillion-dollar filing.

Our partners and sponsors get exclusive placements across the newsletter and access to AIC's in-person network — demo nights, dinners, hackathons, and forums across 180+ chapters.

For all inquiries, send us a note at [email protected].

The AI Collective is built by volunteers across 180+ chapters in 40 countries.

Thank you to the thousands of volunteers around the world who make this work possible. We truly could not do this without you.