Anthropic Unveils Orbit Assistant and Multi-Agent Features at Code With Claude 2026

Anthropic used its Code with Claude 2026 developer conference in San Francisco on May 6, 2026 to debut Orbit, a proactive AI assistant that produces personalized briefings by syncing with Gmail, Slack, GitHub and Figma, per coverage from Simon Willison’s live blog, TestingCatalog and Phemex News. Orbit is built on Anthropic’s internal MetaEra stack and is positioned as a developer-first counterpart to ChatGPT Pulse: rather than summarizing executive email, it builds briefings tuned to product creators — developers, designers and PMs — about their tickets, code, designs and on-call duties.

The conference also introduced a substantial expansion of Claude Managed Agents. Three new capabilities were announced: Multiagent Orchestration for breaking complex tasks into a fleet of specialized agents; Outcomes, a structured success-criteria mechanism so agents can iterate against measurable goals; and Dreaming, which lets Claude recall prior sessions and build on past work rather than restarting cold. Anthropic also outlined an “advisor strategy” in which Sonnet can dispatch hard sub-problems to Opus on demand — one customer reported “frontier-model quality at 5x lower cost.”

On the Claude Code side, Anthropic launched Claude Code on Desktop, a full-screen GUI surface alongside the existing CLI and IDE entry points, plus a Code Review product the company says is now used by every team at Anthropic. The announcements land just over a week after Anthropic doubled five-hour usage limits for Pro, Max, Team and seat-based Enterprise plans and raised Opus API limits after expanded SpaceX compute capacity came online. Code with Claude continues to London on May 19 and Tokyo on June 10, both livestreamed.