Brendan O’Leary is a developer relations engineer at Kilo Code who helps software teams make agentic engineering practical without surrendering human judgment. His career spans healthcare software, GitLab, open-source security, data orchestration, and developer communities.
O’Leary spent a decade at healthcare-software company MagView, ultimately becoming director of product, before leading release management and software quality at Orbis Technologies. He joined GitLab in 2017, working in professional services, continuous-integration product management, and developer evangelism. His writing on GitLab’s contributor community emphasizes the influence of people beyond a company’s payroll on open-source products.
He subsequently led community at ProjectDiscovery, where he wrote about reusable Nuclei vulnerability-detection templates, and became vice president of developer relations at Prefect. He also served on the Cloud Native Computing Foundation governing board and founded Liscio Apps.
An early management mistake shaped his approach to AI: interns building a healthcare application faithfully copied the Comic Sans typography and placeholder icons from his wireframe because he had not explained which details mattered. O’Leary applies the same lesson to coding agents: execution without business context or architectural judgment can produce technically correct but inappropriate work.
His approach to agentic engineering
- Research, plan, implement. O’Leary separates understanding the existing system from designing changes and writing code. Kilo’s Ask mode supports research without immediately modifying files; explicit plans establish scope, affected files, and verification before implementation. His guide to prompting and context engineering develops this workflow.
- Context engineering requires restraint. He recommends concise AGENTS.md instructions, reusable task-specific skills, external notes, selective file references, and only necessary MCP integrations. When an agent accumulates irrelevant context or incorrect assumptions, he favors reviewing a concise handoff and starting a fresh session.
- Delegate only what humans can review. Small commits, explicit permissions, tests, build checks, and readable pull requests keep engineers accountable for architecture and correctness. His analysis of Kilo’s adoption also distinguishes downloads from active users and token consumption from actual software quality.
- Shared conventions make speed sustainable. His work on design systems for agentic engineering addresses how organizations maintain coherence when humans and agents ship rapidly. He favors inspectable tools, portable context, and workflows that leave consequential decisions under human control.
