Code with Claude London 2026

• Code with Claude London 2026 focused on one major theme: moving Claude from a coding assistant into a practical platform for agentic software development, managed agents, enterprise workflows, and AI-native engineering.
• Anthropic used the London event to show how developers and companies are using Claude across codebases, product workflows, agent infrastructure, and large-scale engineering operations.

• Claude Code is moving toward async engineering
• Anthropic showed how Claude Code can run multiple sessions, manage work in the background, fix CI issues, respond to code review comments, and create PRs through routines. Boris Cherny described this shift as moving from "I prompt Claude Code" to "Claude prompts Claude Code," meaning developers can create routines that trigger work automatically.

Anthropic highlighted Claude Code Desktop, a full-screen graphical interface for managing local and cloud sessions, plus an agents view in the CLI for developers who prefer the terminal. These interfaces help developers see which agents are running, blocked, waiting for input, or finished.

Routines let developers trigger Claude Code work through schedules, webhooks, or API calls. This allows tasks such as issue triage, pull request review, CI fixes, recurring checks, and scheduled code work to run without manually starting each session.

Auto mode allows Claude to continue working when an action is safe, instead of stopping for permission every time. The system checks whether an action is destructive or looks like prompt injection before deciding whether Claude can proceed.

Anthropic introduced Auto Memory for Claude Code. It lets Claude capture useful project context, coding style, architectural choices, and debugging notes in local memory files. These memory files stay on the user's machine and help future sessions start with better context.

Anthropic announced updates to Claude Managed Agents, including self-hosted sandboxes and MCP tunnels. Self-hosted sandboxes let teams run Claude's work inside their own infrastructure, while MCP tunnels let agents securely access private MCP servers without exposing them publicly.

Anthropic also discussed multi-agent orchestration, outcomes, memory, and dreaming. These features help developers build agents that can delegate work, improve across sessions, work toward success criteria, and operate with more production-level visibility.

The London event highlighted how companies are using Claude in real engineering environments. Spotify said more than 99% of its engineers use AI coding tools every week, 94% report higher productivity, and PR frequency increased by 76%. Spotify also said it has merged 2.5 million automated maintenance PRs through its fleet management system.

Anthropic and guests highlighted examples from Spotify, monday.com, Doctolib, Delivery Hero, Lovable, and Legora. These examples covered repo migration, healthcare workflows, enterprise agents, vibe coding, self-healing platforms, legal agents, and AI-native product development.

• Code with Claude: Extended London: Extended London happened on May 20, 2026. Anthropic added this second event for independent developers and early-stage founders after high demand for the main London event.
• Extended London focused on founder stories, builder deep-dives, practical demos, managed agents, memory, evals, multi-agent systems, and live agent challenges. Unlike the main Code with Claude London event, Extended London was not livestreamed, but Anthropic said sessions would be recorded and made publicly available after the event.

• Code with Claude London matters because it showed how Claude is moving from single-session coding help into a broader developer and agent platform.
• For developers, the event highlighted Claude Code updates such as routines, auto mode, work trees, memory, agents view, remote control, and improved desktop workflows.
• For engineering leaders, the event showed how Claude can support repo migrations, PR review, CI fixes, security checks, and engineering productivity at company scale.
• For platform teams, Claude Managed Agents, self-hosted sandboxes, MCP tunnels, outcomes, memory, and multi-agent orchestration showed how teams can build production-grade agents with better control.
• For founders and product teams, Extended London showed how Claude can help builders move from idea to working product through demos, founder stories, workshops, and practical agent workflows.