Anthropic discloses Claude now writes 80% of its own codebase and delivered a 52× training speedup — with human judgment the last narrowing frontier.
Anthropic ships Claude Opus 4.8 with dynamic workflows: hundreds of parallel subagents plus adversarial judges — and a honesty-first architecture.
A 32,000-GPU-hour benchmark confirms the harness layer outweighs model choice — identical backbones swing 3× in accuracy depending on agent framework.
Menlo Ventures data shows Anthropic at 34.4% enterprise share vs OpenAI's 32.3% in April, triggering simultaneous free-trial counter-offers from both labs.
Anthropic engineer Thariq argues HTML beats Markdown for AI output; Karpathy backs it with a six-step format evolution toward interactive neural video.
Anthropic ships Agent View for Claude Code: a unified terminal dashboard for managing parallel sessions with status indicators, inline replies, and /goal.
Anthropic secures all of SpaceX's Colossus 1 (220K GPUs, 300 MW) and doubles Claude Code rate limits—sub-agent fan-out is now viable at $200/month.
Anthropic leases all of SpaceX Colossus 1 (300 MW, 220K GPUs), doubles Claude Code limits, and ships Managed Agents, Dreaming, and Outcomes in one day.
Nine sources across GitHub, YouTube, X, and newsletters converge on one finding: the model is no longer the performance frontier — the harness is.
Three independent research groups converge on April 29 to map a triple security exposure in agentic coding editors: an 81% permission gate FNR, 84% prompt injection success, and systemic plan compliance failures.
Matt Pocock's two-hour AI Engineer workshop argues 30-year-old software fundamentals matter more under AI, not less — and outlines a complete methodology to prove it.
Anthropic published a post-mortem on three sequential Claude Code harness changes from March–April that degraded output quality, fixed in v2.1.116+.
Curated AI insights, sent when there's something worth your inbox.