Mouad Belghiti
ENMe contacter ↗
← digest

The AI and Tech Weekly: June 7, 2026

Mouad Belghiti·2026-06-077 min read

The AI and Tech Weekly: June 7, 2026

Top market news, developer tips, open source picks, and startup ideas from the community

Market News

Anthropic Says Claude Writes 80% of Its Own Code

Anthropic published a detailed update on what it calls recursive self-improvement. The headline number: Claude now writes more than 80% of the code that lands in Anthropic's own repository. Six months ago, Claude succeeded on only 26% of the hardest, least-specified engineering tasks. In May 2026, that number reached 76%. Anthropic engineers are shipping 8x more code per quarter than they did between 2021 and 2025.

Claude is not just writing features. It is proposing training recipes, analyzing failure modes, and tuning hyperparameters. The company is clear that full recursive self-improvement — where an AI designs and trains its own successor without human input — is not here yet. But Anthropic says it could arrive before most organizations have a plan for it. The post is worth reading for the concrete metrics, not the headline.

Cloudflare Acquires VoidZero

Cloudflare bought VoidZero, the team behind Vite, Vitest, Rolldown, and Oxc. Together those tools see 130 million weekly downloads. Evan You (Vue.js creator, Vite author) stays on to lead them, and all four projects remain MIT licensed. Cloudflare is putting $1 million into an independent maintainer fund for contributors not affiliated with either company.

The goal is to embed this toolchain into Cloudflare Workers for AI-native web apps. For most frontend developers, nothing changes today. The long-term bet is that Vite and its siblings become the default development environment for apps built on Cloudflare's edge platform. Watch this over the next 12 months.

Meta Shipped Facial Recognition to 50 Million Devices

Wired found a feature called "NameTag" inside the Meta AI app. It can capture faces through Meta's Ray-Ban smart glasses, index them, and recognize them again when seen later. The feature is not active for regular users, but it already lives on over 50 million phones. Meta said it is internal exploration. Privacy researchers are not reassured, and the HN thread generated 189 comments.

The critical detail: the software is already distributed. Meta does not need to flip a switch at the app level. Any regulation of this feature will need to happen before deployment, not after.

GitButler Raises $17M to Rebuild Version Control

GitHub co-founder Scott Chacon raised a $17M Series A from a16z for GitButler. The thesis: Git's mental model was built in 2005 for large teams with long-lived branches. It does not fit how small teams or solo developers work alongside AI assistants today. GitButler rebuilds the assumptions from scratch, adding parallel branches, unlimited undo, and clean agent integration with Claude, Cursor, and Codex. The HN discussion hit 643 comments, the highest of the week, with a real argument about what is actually broken in Git.

Tips and Tricks from the Community

  1. Read the MCP vs. skills debate before committing to an agent architecture — A post called "I still prefer MCP over skills" hit 420 points and 337 comments on HN. The argument: MCP gives you typed tool definitions and automatic discovery; skills are flexible but require more custom wiring. The comments include real trade-offs from people who have built both. Read this before your next agent design session. (david.coffee/i-still-prefer-mcp-over-skills)

  2. Use Anthropic's open source vulnerability scanner in CI — Anthropic open-sourced a framework for AI-powered code vulnerability discovery called defending-code-reference-harness. It uses a model to scan for security issues and includes reference implementations you can adapt. It is used internally at Anthropic on production code. Clone the repo, point it at one service, and run it before adding it to your full pipeline. (github.com/anthropics/defending-code-reference-harness)

  3. Try marimo-pair to give AI agents a live Python environment — Marimo-pair makes reactive Python notebooks into execution environments where agents can write and run code. Unlike subprocess-based approaches, the notebook is reactive: the agent updates a cell and downstream cells update automatically. If you're building a coding agent, replace your current execution environment with marimo-pair on one project and compare. (github.com/marimo-team/marimo-pair)

  4. Quantize your KV cache to extend context on budget GPUs — Huawei's KVarN is a native vLLM backend that quantizes the KV cache to reduce memory use during inference. If you run models on consumer GPUs or budget instances, the KV cache is often your bottleneck. Quantizing it can extend your usable context window without buying more hardware. Run a memory benchmark before and after on your current setup. (github.com/huawei-csl/KVarN)

Open Source This Week

Startup Ideas for Inspiration

Drawn from YC's Requests for Startups and other leading accelerators

AI-Personalized Medicine

Most drugs are prescribed based on population averages. A patient's genome, microbiome, lifestyle, and prior drug responses are rarely factored in, even though they strongly predict whether a treatment will work. The result: patients cycle through medications for months before finding one that works, and many never do.

YC's bet is that genomic sequencing (now under $200) removes the data collection barrier that blocked this space for a decade. The most tractable starting point for a small team is drug response prediction: using a patient's genetic profile to rank likely treatments before the first prescription is written. This is not a consumer product. It starts as a B2B tool for clinics and hospital systems that already have patient data and need better decision support.

SaaS Challengers

Most SaaS products were built for a world where software was expensive to build. A feature that needed 6 engineers and 9 months now takes 2 weeks. YC is looking for startups that go after established SaaS categories — HR, project management, billing, CRM — and rebuild them from scratch with a smaller team and a lower price point.

The bet is not that incumbents are bad products. It's that their pricing reflects the cost structure of building software in 2015, and a new entrant can undercut them on price while matching 80% of the features. The opening move is usually a specific industry vertical that the incumbent serves badly.

Startups That Sell to Very Large Companies

Enterprise sales has a reputation for being slow and painful. YC is pushing back on that. Its argument: large companies have large budgets and low switching costs for point solutions. A startup that can close one $1M contract has more revenue than ten $100K contracts, with similar sales effort. The categories YC is most interested in are compliance, supply chain visibility, and internal developer tooling, all of which large companies buy but rarely build themselves.

Industrial Capabilities in Space

The cost to put a kilogram in orbit has dropped from $54,000 in 2010 to under $2,000 today. That changes what is worth doing in space. YC wants startups building industrial processes that either work better in microgravity (fiber optics, pharmaceuticals, certain alloys) or that use space-based solar and raw materials to produce things that are cheaper to make off-Earth than on it. This is early-stage and long-horizon, but YC is funding it because the infrastructure window is open now.

※
The AI and Tech Weekly: June 7, 2026 | Belghitis