Summary
Claireo from How I AI attended Anthropic’s first Code with Claude developer conference in San Francisco and covers five features that shipped for Claude Code and the Claude API. First is Routines — a built-in scheduler enabling Claude Code to run agent tasks on cron schedules, HTTP webhooks, or GitHub webhooks, with Slack and GitHub connectors included. Users can now automate recurring tasks like weekly newsletter drafts or PR reviews directly within the app without external orchestration.
The second feature is Outcomes, a Claude Managed Agents API primitive where developers upload a markdown rubric defining success, and the agent self-grades and iterates up to 20 times until it meets the target — conceptually similar to OpenAI’s `/goal` in Codex. Third is a formal multi-agent framework in the API supporting up to 25 agents sharing a container and filesystem, with explicit orchestrator-delegate hierarchy and per-agent tool sets. Fourth is Dreams, an agent memory primitive that writes structured markdown files on a configurable schedule rather than only on session-close hooks, giving developers more control over when and what agents remember across tasks.
The video rounds out with the rate limit increases tied to Anthropic’s SpaceX compute deal. Together, these five additions move Claude Code from a capable coding assistant toward a platform for building persistent, scheduled, multi-agent production systems — a significant shift in what developers can architect directly within the Claude ecosystem.
📺 Source: How I AI · Published May 07, 2026
🏷️ Format: News Analysis
