Week of June 1, 2026
8 stories tracked
Anthropic dropped Opus 4.8 on May 28 with Dynamic Workflows — Claude Code can now plan complex work, spawn hundreds of parallel subagents, and verify outputs before reporting back. The model is 4x less likely than Opus 4.7 to miss coding flaws. Fast mode (2.5x speed) is now 3x cheaper, and effort control is live on claude.ai.
Announced at Computex, Nemotron 3 Ultra is a 550B-parameter open-weights model using NVIDIA's hybrid latent MoE architecture — only 55B parameters fire per token. It scores 48 on the Artificial Analysis Intelligence Index (highest of any US open model) and runs 3–6x faster than competing Chinese open models.
On June 1, Anthropic submitted a confidential draft S-1 to the SEC. This follows a $65B Series H that valued the company at $965B — surpassing OpenAI's $852B private valuation for the first time. Revenue run-rate hit ~$47B in May, with Q2 expected at $10.9B (more than double Q1). An IPO could happen as early as this fall.
At Build 2026, Microsoft unveiled the MAI family — seven models trained entirely in-house with zero distillation. MAI-Thinking-1 is a 35B-parameter reasoning model with 256K context that independent raters prefer over Sonnet 4.6. MAI-Code-1-Flash is already deploying into VS Code and GitHub Copilot. The strategic signal: Microsoft is building every layer of its AI stack, reducing dependence on OpenAI.
In a Platformer interview, Boris Cherny said he ships 10–30 PRs per day with zero manual code edits, running five Claude Code agents simultaneously. He believes coding is effectively solved for the work he does, and that major job displacement is coming — but so is job creation. The interview is the most concrete first-person account of what fully AI-assisted development actually looks like at scale.
On June 2, Trump signed an executive order directing agencies to create a framework for voluntary pre-release review of frontier AI models — up to 30 days before public launch. The order also establishes AI cybersecurity benchmarks and a vulnerability clearinghouse. An earlier draft requiring 90 days was scrapped over concerns it would stifle innovation.
On June 2, Cognition retired Windsurf and relaunched as Devin Desktop with a Kanban-style Agent Command Center as the default surface. Devin Local replaces Cascade — rewritten from scratch in Rust with 30% better token efficiency. The big move: ACP (Agent Client Protocol) support, meaning Codex, Claude Agent, and OpenCode can all run inside the same editor.
GitHub Copilot moved to usage-based billing on June 1. Your subscription becomes a base credit pool (Pro: $10/mo gets 1,500 credits; Pro+: $39/mo gets 7,000) with temporary flex allotments through September. Code completions stay free, but chat, agent, and premium model usage all consume credits. No more fallback — once you're out, you're out until next month.