An unofficial, lovingly-curated recap of Code with Claude San Francisco and the Extended workshop day. Two days of announcements, deep technical talks, and hands-on sessions — collected here as an interactive timeline. Click any session to dive in.
The conference inhabited a converted SoMa warehouse — cherry-wood paneling, coral-pink screens, and a small army of pixel-art Claudes drifting through the corridors. A few moments that weren't on the official program.
From the keynote stage: Anthropic announced a deal to take the entire compute capacity of SpaceX's Colossus 1 supercomputer in Memphis — 220,000+ NVIDIA GPUs (H100, H200, GB200), 300+ megawatts of new capacity. Same post: Claude Code limits doubled for Pro, Max, Team, and seat-based Enterprise; peak-hour reductions removed; Opus API limits raised.
anthropic.com/news/higher-limits-spacex ↗ Managed AgentsScale a fleet of agents and have them break complex tasks into coordinated pieces — first-class on the Claude Managed Agents platform.
platform.claude.com/docs ↗ Managed AgentsDefine success criteria for an agent and let it iterate autonomously to hit them — reframes agent design around what good looks like, not the steps to get there.
platform.claude.com/docs ↗ Managed AgentsBetween sessions, Claude inspects what it did and writes new memories. Agents that get better at your workflow the more they run it. Featured deep-dive talk by Mahesh Murag.
platform.claude.com/docs/dreams ↗ Claude CodeAutomated PR review built into Claude Code — the same tool every team at Anthropic uses internally, now available to everyone.
code.claude.com/docs/code-review ↗ Claude CodeWhen CI fails on a PR, Claude files an automatic fix back against the same PR. Less context-switching; more "the bot already pushed the patch."
code.claude.com/docs/whats-new ↗ SecurityClaude Code now runs security-focused review passes over changes — SQL injection, XSS, auth issues, secrets — before the PR ships.
support.claude.com ↗ Claude Code"Higher-order prompts." Define an async automation that wakes up on a schedule or trigger and produces ready-to-merge PRs without a human in the loop.
code.claude.com/docs/routines ↗ DeveloperThe same SDK Anthropic uses to build Claude Code itself, now public for outside developers building their own agents.
code.claude.com/docs/agent-sdk ↗Sources verified against the Anthropic newsroom, the Claude Managed Agents docs, code.claude.com, and Simon Willison's live blog · Ars Technica