Claude Code Just Got WAY More Powerful

Claire’s quick take from Anthropic’s first dev event

Claire opens straight from Code with Claude, Anthropic’s first developer event, with a promise: five launches, what they do, and what she’d actually build with them. Her framing is practical from the jump — not “mindblowing,” but useful right now if you build with agents.

Routines bring cron jobs into Claude Code

The first feature is the one she’s clearly happiest about: routines. She compares it to what she already loved in OpenClaw — crons — and demos a weekly newsletter workflow that checks a changelog.md every Monday at 6:00 a.m. and drafts customer-facing copy, explicitly skipping internal stuff like tech debt unless customers would care. The key detail is flexibility: routines can run locally or remotely, and they can fire on a schedule, an HTTP webhook, or a GitHub webhook, with connectors like Slack and GitHub available inside the workflow.

Outcomes: agents that keep iterating until they hit the rubric

Next she points to Anthropic’s answer to OpenAI Codex’s /goal: outcomes. The core idea is simple but powerful — define what “done” looks like in a markdown rubric, add a grader, and let the agent self-check and retry up to 20 times. Her example is a “ship-ready PRD,” where the agent keeps revising against product priorities and technical constraints instead of handing you one draft and walking away.

Multi-agent teams are now a first-class API primitive

Claire is especially into the new managed-agent support for multi-agent systems. Through the API, you can define an orchestrator plus delegate agents — up to 25 total — all working in the same container and filesystem, but with different tool access and responsibilities. Her ChatPRD example makes it click: one master PRD orchestrator, plus a strategy agent channeling the CPO voice, a critic agent to poke holes, and an engineering review agent wired into GitHub.

Dreams turns agent memory into a deliberate process

Then comes “Dreams,” which she jokes is a funny name but a useful concept. Instead of only writing memory on a hook — like when a session ends — Dreams lets you explicitly review a pile of past sessions, say 50, and synthesize the important memories worth saving to disk. She leans into the metaphor: maybe this really is what we do when we sleep, silently review the day, and decide what sticks.

More usage, fewer limits, and a clear platform strategy

The last update gets the biggest practical cheer: higher usage limits. Claude Code’s 5-hour limits are now doubled for Pro, Max, Team, and seat-based enterprise users; peak hours are disappearing for Pro and Max; and Opus API rate limits are increasing too. Claire closes by bundling the whole event into one read: routines, outcomes, multi-agent orchestration, and memory primitives all point to Anthropic trying to become the agent platform builders choose first.