Anthropic's Code w/ Claude 2026: no new model, but Routines, Dreaming, and a SpaceX deal

Anthropic’s Code w/ Claude 2026 keynote skipped a model launch and instead focused on tooling around the existing Claude lineup. Chief Product Officer Ami Vora announced doubled Claude Code five-hour limits for Pro, Max, and Enterprise tiers, backed by a capacity deal with SpaceX’s Colossus data center in Memphis — the same facility with a notable environmental record. API traffic is reportedly up 17x year-on-year.

Claude Managed Agents picked up three additions: multi-agent orchestration (public beta) for fleets of cooperating agents, Outcomes (public beta) that let Claude iterate against a success definition in a Ralph-loop style, and Dreaming (research preview) where an agent reviews prior sessions overnight and writes new memory files like a descent-playbook.md to self-improve. An ‘advisor strategy’ was pitched where Sonnet calls Opus on demand, with one customer claiming frontier-quality output at 5x lower cost.

Claude Code now spans CLI, IDE, and a Desktop app — all built on the same Claude Agent SDK exposed to external developers. Boris Cherny demoed multiple async sessions running in parallel, with Routines acting as higher-order prompts that produce ready-to-merge PRs and auto-fix CI failures so reviewers ‘never see a red X.’ Shopify and Mercado Libre’s 23,000-engineer org were cited as customers chasing 90% autonomous coding by Q3.