Claude Code adds Routines and Dreams at Code with Claude London
TL;DR
Anthropic introduced Routines for scheduled and webhook-triggered Claude Code automations, and Dreams — a research-preview self-improvement loop — at its first European developer conference.
May 19–20, 2026
Code with Claude London — first Anthropic developer conference in Europe
Routines
Productized — scheduled, webhook, and API-triggered automations running on Claude Code cloud infrastructure
Dreams
Research preview — self-improvement loop consolidating memory between sessions
~6x
Harvey's internal task-completion improvement after enabling Dreams — Anthropic customer claim, not independently verified
Anthropic announced Routines and Dreams for Claude Code at Code with Claude London on May 19–20, the company's first dedicated European developer event. The headline framing across the event was async agentic coding: Claude Code now runs scheduled jobs on Anthropic's cloud infrastructure and writes notes to itself between sessions. The features land at different availability levels — Routines is positioned as a productized feature, while Dreams launched in research preview alongside public-beta releases of outcomes and multi-agent orchestration.
What Routines actually is. Routines lets developers configure Claude Code to run automatically on a schedule, via an API call, or in response to external events such as GitHub webhooks. According to InfoQ's reporting on the announcement, scheduled routines cover recurring jobs like triaging bugs, scanning for documentation drift, or generating pull requests; webhook-based routines can launch a Claude Code session when a pull request matches specified conditions, then monitor updates, respond to PR comments, track CI failures, and continue across the lifecycle of the change. Routines run on Claude Code's cloud infrastructure rather than the developer's local machine — removing the need for self-managed cron jobs or servers.
What Dreams actually is. Dreams is a self-improvement loop: Claude inspects its previous sessions, consolidates a memory store, replaces stale or contradicted entries with the latest values, and surfaces new insights. The framing is that Claude "dreams" between sessions to consolidate what it learned. Anthropic shared one customer data point at the event — legal-AI company Harvey reported roughly 6x improvement in task completion rates in internal testing once Dreams was turned on, with previously-forgotten file-type quirks and tool-specific workarounds persisting across sessions. This figure is Anthropic's customer claim and has not been independently verified. Dreams is currently in research preview, not general availability.
Rate-limit changes. Press coverage from the event reported that Anthropic doubled the five-hour Claude Code limit for paid plan tiers (Pro, Max, and Enterprise). Anthropic's own news page does not yet list a standalone post announcing the rate-limit change, so the per-tier specifics are best read from official Anthropic channels before assuming the doubling applies uniformly to your plan.
Why this matters for AI tool users. For developers and small teams using Claude Code today, Routines is the practically meaningful release: it converts Claude Code from an interactive tool into a background worker. Scheduled jobs that previously needed a custom runner — overnight code reviews, scheduled refactors, on-PR triage — can now run inside Anthropic's infrastructure with no extra plumbing. Dreams is a more speculative bet: if it ships from research preview to a productized feature, it materially changes how Claude Code persists context between sessions. For now, treat it as an experiment to watch, not a feature to rely on. See our Claude Code review for the current capability baseline, and our Cursor vs Claude Code comparison for how the two agentic coding tools differ.
What to do this week. If you already pay for Claude Code on Pro, Max, or Enterprise, check your account for the new rate-limit allowance before assuming overnight Routines are within your budget. If you build any kind of GitHub-triggered automation today (cron jobs, bots, custom workflows), audit whether Routines could replace one workflow you currently maintain — start small with a single scheduled task and measure cost against your existing setup. Hold any production migration onto Dreams until it leaves research preview. If you're evaluating Claude Code against Cursor or GitHub Copilot for your team, our best AI coding tools roundup ranks the current options. For builders deciding what to ship next, the saas-with-cursor-claude-code/" class="text-blue-300 hover:text-blue-200 underline decoration-blue-400/40 underline-offset-2">micro-SaaS playbook uses both tools together.
Why It Matters
Routines turns Claude Code from an interactive tool into a background worker. That is the underlying shift. Scheduled or webhook-triggered agent sessions that previously needed a custom runner can now live inside Anthropic's infrastructure. Dreams is a more speculative bet — a memory consolidation loop currently in research preview — but if it productizes, it changes how Claude Code holds context between sessions. For developers running async automations against codebases, this is the most meaningful Claude Code release since the original launch.
Who's Affected
- — Developers and small teams already paying for Claude Code. Routines opens up workflows that previously required your own infrastructure: scheduled triage, on-PR review, doc-drift scans. The rate-limit increase reported at the event is a real input cost change — check your account.
- — Anyone maintaining custom GitHub bots or scheduled CI jobs. Some of what you maintain today could move to Routines. The trade-off is platform lock-in vs. eliminated maintenance. Worth a one-workflow pilot before any broader migration.
- — Teams evaluating Claude Code vs. Cursor or Copilot. Routines is now a meaningful differentiator on the asynchronous-agent axis. Cursor's positioning is interactive IDE-native; Claude Code is increasingly headless and background-friendly. The decision matrix has shifted.
- — Anyone counting on Dreams as a production feature. Don't yet. Research preview means subject to change, removal, or breaking behaviour. Treat it as an experiment, not a stable contract.
What To Do Now
- 1. Pilot Routines on one workflow before broader migration. Pick a single recurring task — overnight dependency triage, scheduled doc-drift scans, on-PR linting. Run it for two weeks and measure cost and reliability against your existing setup. Don't move multiple workflows at once.
- 2. Treat the Harvey 6x number as one customer's experience. Anthropic shared it as an illustrative data point. Your codebase, prompts, and session patterns are different. Do not use it as the basis for an internal ROI projection.
- 3. Verify the rate-limit doubling on your specific plan. Press coverage reported a uniform 2x increase for Pro, Max, and Enterprise. Confirm against your account dashboard before scaling Routines usage to the new ceiling.
- 4. Keep Dreams off your production critical path. Research preview is exactly that. If you find it useful, log what works and revisit when Anthropic promotes it. Don't engineer around it as if it were stable.
More on this topic — Best AI Coding Tools
Independent Review
Claude Code
Pricing, pros and cons, real-world verdict — no affiliate spin.
Read the Claude Code reviewMore from ToolNav News