Claude now authors 80% of Anthropic's code, doubling task capability every 4 months

Broadcom's AI guidance miss triggers $400B capital rotation from Bitcoin into infrastructure

Broadcom beat Q2 earnings but guided fiscal Q3 AI revenue to $16B versus $17.2B consensus and held full-year AI target at $56B rather than raising it, triggering a sell-the-news reaction that sent AVGO down ~14% Thursday. The miss spooked the entire semiconductor group, as Broadcom is the cleanest proxy for hyperscaler custom-silicon demand across six major buyers on locked 18-24 month design cycles. The selloff coincided with spot Bitcoin ETFs recording their worst week since February, with $4B in outflows since mid-May as BTC fell 17% to $59,000—a shift framed less as a fundamental verdict on Bitcoin than as capital migration into the ~$400B AI infrastructure investment trend, underscored by Michael Saylor's first Bitcoin sale in four years to fund Strategy's preferred dividends.

AI spending creates 5x productivity divergence; coding agents boost commits 180% but releases lag 30%

Ramp data shows American companies with the highest AI investment seeing revenue growth five times faster than non-spenders, while business applications have spiked since 2023 as cheaper AI-powered company formation tools lower entrepreneurship barriers. Anthropic disclosed that code contributed per developer is eight times higher in the current quarter than the average through end of 2024, with a meaningful uptick starting in 2025 as Claude 4 and Claude Code rolled out. However, a GitHub study of 100,000+ developers reveals a critical bottleneck: while agentic coding dramatically outpaces first-gen autocomplete in raw output (commits up 180%), actual shipped products lag far behind (releases up only 30%), indicating the constraint has shifted from writing code to integration, review, and deployment workflows. This creates a long tail of mediocre apps but doesn't prevent best apps from improving.

Recursive Self-Improvement: AI Systems Accelerating AI Development

Anthropic Institute published a report on recursive self-improvement and AI-driven acceleration of AI development, with previously unreported internal metrics. Main takeaways:

• Claude authored >80% of code merged into Anthropic's codebase as of May 2026, up from single digits before February 2025; engineers now ship 8x more code per day than in 2024.
• Claude's success rate on open-ended coding tasks reached 76% in May 2026 (up 50 points in 6 months); automated Claude code review would have caught ~33% of bugs in past production incidents.
• Task completion capability is doubling every 4 months (up from 7-month doubling): Claude Opus 4.6 handles 12-hour tasks vs. 4-minute tasks a year prior; trajectory suggests week-long tasks possible in 2027.
• Claude achieves ~52x speedups on code optimization tasks (April 2026) vs. ~3x a year prior; human researchers typically achieve 4-8x in similar timeframes.
• Median Anthropic researcher estimated 4x productivity output with Mythos Preview access; SWE-bench and CORE-Bench benchmarks saturated in 2 years and 15 months respectively.