Zach Anderson
May 27, 2026 00:41
Anthropic unveiled key updates to Claude Managed Agents, including self-hosted sandboxes and MCP tunnels, at the London Code w/ Claude 2026 conference.
Anthropic’s Code w/ Claude London 2026 event, held on May 19–20, brought European developers into the fold of its ambitious global AI conference series. This marked the second stop in the inaugural three-city tour, following San Francisco earlier this month and ahead of Tokyo in June. The event showcased the company’s commitment to redefining coding workflows with its Claude AI models and developer platform.
A key highlight of the London gathering was the announcement of new capabilities for Claude Managed Agents, designed to give enterprises tighter control and enhanced security over agentic coding workflows. Two major updates were unveiled:
- Self-hosted sandboxes (public beta): These allow tool execution to occur within infrastructure configured by the user, such as their own servers or managed providers like Cloudflare or Vercel. This ensures network policies, audit logs, and security measures remain within the enterprise’s boundaries. This feature also enables users to manage compute-intensive workloads more flexibly.
- MCP tunnels (research preview): These provide secure, private connections to Model Context Protocol (MCP) servers within an enterprise’s network. By avoiding public internet exposure and encrypting traffic end-to-end, this feature enhances data security and simplifies integration for enterprise teams.
Anthropic executives highlighted the potential of these tools to streamline operations for industries ranging from software development to legal tech. Early adopters like Amplitude, Clay, and Rogo are already leveraging the new features to optimize workflows and maintain stricter security controls over their AI deployments.
Reimagining Coding with AI Agents
The conference’s keynote, delivered by Boris Cherny, Head of Claude Code, centered on the transformative potential of AI in programming. Cherny emphasized how complex layers in modern development—compilers, typecheckers, and build systems—have distanced programmers from the immediate feedback of seeing their ideas run. Claude Code, he explained, aims to collapse this gap by enabling developers to describe problems while AI agents generate functional solutions.
Breakout sessions included hands-on workshops and case studies from companies like Spotify and Base44, demonstrating how Claude Code is being used to scale engineering teams and automate repetitive tasks. One session explored how “thinking budgets” can help teams optimize resource allocation across AI models.
What’s Next
For those who missed the London event, Anthropic has made session recordings available online. The next stop for Code w/ Claude 2026 is Tokyo, scheduled for June 5–6, where keynotes and breakout sessions will be livestreamed.
Anthropic’s expansion of its developer conference to three cities in 2026 reflects the increasing global interest in AI-powered coding solutions. As enterprises continue to adopt AI agents, tools like Claude Managed Agents could play a pivotal role in balancing innovation with security and control.
Image source: Shutterstock