Michael Patterson

Michael Patterson focuses on enterprise coding-agent security: how organizations can adopt autonomous development tools while controlling their access to private data and outside services. At the AI Engineer World’s Fair in 2026, he was a staff solutions engineer at Coder.

The risk on developer laptops

Patterson applies Simon Willison’s lethal trifecta to coding agents: access to private data, exposure to untrusted content, and the ability to communicate externally. Agents such as Claude Code, Codex, and OpenClaw running on developer laptops can combine all three. Prompt injection connects these capabilities into a threat: malicious instructions encountered in content can redirect an agent that also has access to internal information and the internet.

His approach to securing coding agents puts controls around where they run, what information reaches the model provider, and which actions they can take:

• Cloud development environments: Isolate the agent’s execution from the developer’s laptop.
• Model proxy: Log model traffic and strip sensitive information before it reaches the LLM provider.
• Agent firewall: Deny commands and network destinations that the organization has not allowed.

These controls address different parts of the problem: execution isolation, visibility and filtering of model traffic, and restrictions on agent actions.

Moving from individual use to enterprise adoption

Patterson also examines the organizational obstacles to rolling out agents at scale. Personal AI subscriptions can put usage outside company oversight; alert fatigue can overwhelm security teams; and skeptical senior engineers can resist new workflows. His emphasis is on making agent adoption manageable for both developers and security leaders, with architectural guardrails that address the risks already present in everyday laptop use.