Companies·3 min read·Simon Willison's Live Blog

Code with Claude 2026: Anthropic Reveals 'Dreaming' Agents, Routines, and a Rebuilt Claude Code

At its second annual developer conference, Anthropic unveiled a stack of agentic features — overnight self-improving 'dreaming' agents, scheduled Routines, multi-agent orchestration, automated Code Review and CI auto-fix — alongside doubled Claude Code rate limits.

Code with Claude 2026: Anthropic Reveals 'Dreaming' Agents, Routines, and a Rebuilt Claude Code
Share:

Anthropic used the San Francisco kickoff of Code with Claude 2026 on May 6 to roll out the largest batch of developer-facing features it has shipped since Claude Code's debut, anchored by a research preview called "dreaming" that lets agents replay prior sessions overnight and write new memories before the next workday begins. The conference, which travels to London on May 19 and Tokyo on June 10, was framed less around a new flagship model and more around getting existing Claude variants to do meaningfully more work without a human in the loop.

The marquee reveal was the Dreams memory system. In its current research-preview form, agents inspect their previous sessions while idle and synthesize playbooks — Anthropic showed an example where an agent that repeatedly fumbled a deployment task automatically generated a descent-playbook.md file after analyzing its own failed attempts. The pitch is that Claude stops re-learning the same lessons every morning, a long-running pain point for teams trying to use agents on multi-day projects. Access requires an explicit request, and Anthropic was careful to position the feature as exploratory rather than production-ready.

Around dreaming, Anthropic shipped a tighter agent runtime. Multi-agent orchestration moved into public beta, allowing a lead agent to delegate sub-tasks to specialized worker agents with their own tools and personas. A companion feature called Outcomes lets developers define rubric-based success criteria so an agent can iterate against them without prompting tweaks. Routines — described by Claude Code lead Boris Cherny as "higher-order prompts" — enable scheduled or webhook-triggered async runs; the demo line, "wake up to PRs that are ready to merge," is the headline use case.

Claude Code itself got a substantial product update. Code Review, which Anthropic says every internal team now uses, automates pull-request review with security and style checks. Remote Agents lets developers drive their laptop from a phone, kicking off long-running runs from anywhere. CI auto-fix files automatic remediation patches against failing pull requests, and a refreshed desktop app supports multiple concurrent sessions for parallel async development. The cumulative effect is that Claude Code is no longer just a terminal companion — it is increasingly an always-on coworker that reaches into CI, code review, and on-call rotations.

Anthropic paired the product reveals with a quieter but commercially significant change: the doubling of the five-hour rate limit on Claude Code for Pro, Max, Team, and seat-based Enterprise plans, the removal of peak-hour reductions on Pro and Max, and lifted Opus API ceilings. The capacity expansion is underwritten by the SpaceX Colossus 1 deal disclosed the same day, which adds more than 300 megawatts and 220,000 NVIDIA GPUs to Anthropic's footprint within the month, alongside the previously announced multi-gigawatt commitments with Google, Broadcom, Amazon, and Microsoft.

Customer name-drops did the work of the usual benchmark slides. Shopify and Mercado Libre told the audience they were aiming for "90% autonomous coding by Q3 this year," a number that would have sounded absurd at the same conference twelve months ago. Whether dreaming, Routines, and the rebuilt Claude Code are enough to get there is now an empirical question — but for the first time, Anthropic shipped a roadmap whose obvious endpoint is engineers managing fleets of Claude agents rather than typing prompts at one.

Comments

Share your thoughts. Be kind.

0/2000

Loading comments…

Related Articles

SPACEX · NASDAQ: SPCXLARGEST IPO IN HISTORY$1.75TTarget valuation · up to $75B raisedSurpassing Saudi Aramco's $35.4B record raiseJUN 4RoadshowJUN 11PricingJUN 12Trading · SPCXSource: SpaceX S-1 · CNBC
Companies

SpaceX Sets Terms for the Largest IPO Ever: ~$1.75T Valuation, Up to $75B, Trading June 12 as SPCX

AI SILICON · THE CHIP RACEMICROSOFT × ANTHROPIC4thSilicon vendor in Anthropic's stackif the Maia 200 talks with Microsoft closeStatus: early talks — nothing signed yetMICROSOFT SILICONMAIAMAIA 200216 GB HBM3e · TSMC 3 nm10+ PFLOPS FP4 · launched Jan 2026WHAT IT MEANSAlready on AWS, Google and Nvidia chipsMicrosoft holds a ~$5B stake in AnthropicSource: WSJ / CNBC
Companies

Anthropic in Talks to Run Claude on Microsoft's Maia 200, Adding a Fourth Silicon Vendor

TeamPCP Hackers Put 450 Mistral AI Repos and 5GB of Source Code Up for Sale at $25,000
Companies

TeamPCP Hackers Put 450 Mistral AI Repos and 5GB of Source Code Up for Sale at $25,000