Sia Reads
What Matters

The Anthropic Institute published concrete internal data: over 80% of merged code is authored by Claude, engineers ship 8x more code per quarter than in 2024, and Claude's success rate on open-ended tasks hit 76%. The paper proposes a verifiable coordinated slowdown — Anthropic says it would pause if other frontier labs did so verifiably. The hardest part: training runs are far easier to conceal than missile silos.

On June 9, Boris Cherny shipped Claude Code v2.1.172 with nested subagent support — each subagent can spawn its own subagents, capped at depth=5. Each level runs in an isolated context window and returns only a summary, letting developers break complex workflows into layered delegations. This was previously forbidden in every version of Claude Code. Cherny is explicitly asking for feedback on whether the depth ceiling feels right.

On June 10, Dario Amodei published an essay proposing a tiered government response to AI-driven unemployment: better data collection at 5%, pro-employment incentives at 10%, and UBI funded by taxes on AI companies at unprecedented levels. Anthropic committed $200M to an Economic Futures Research Fund and $150M to a national fellowship program. He explicitly suggested financing UBI through capital gains taxes or levies on 'relevant companies.'

Representatives Obernolte (R-CA) and Trahan (D-MA) released the Great American AI Act on June 4 — the first serious attempt at comprehensive federal AI governance. It requires frontier developers (>$500M revenue) to publish risk frameworks and submit to independent audits, authorizes $100M/year for a Center for AI Standards, creates federal whistleblower protections for AI-related reporting, and preempts state AI development laws for three years.

On June 8, OpenAI filed a confidential S-1 with the SEC, one week after Anthropic filed its own. The targeted valuation sits around $1 trillion, up from $852B in March. Goldman Sachs and Morgan Stanley are managing the offering. Sam Altman said timing is undecided and could be 'a while' away. The parallel filings signal a new era: the two leading frontier labs are both heading to public markets.

On June 11, Google DeepMind and partners announced a $10M research funding call focused on multi-agent AI safety — how large-scale systems of agents built by different organizations behave when they interact. The call covers four areas: sandboxes and testbeds, agent network science, agent infrastructure, and oversight and control. Applications close August 8.