Building AI Operations from the work upward.
I joined a net-new function with a broad mandate: find where AI could create meaningful operational leverage across Honeycomb.
I started by listening. That changed the work.
- 36
- Discovery sessions
- 23
- Experiments identified
- 4
- Active pilots
- 100+
- GTM tasks decomposed
- 400+
- Subtasks analyzed
Engineering AI operating model
Human review, accountability, risk, code review, pairing, and which AI-assisted workflows are actually acceptable. The debates were symptoms of an unresolved operating model.
AI governance
Separating principles, norms, guidance, product policy, employee policy, and the operating rules a person can actually act on at their desk.
Tool evaluation
Evaluating AI products against real workflows instead of feature lists.
Agentic architecture
Claude, MCP servers, skills, context, tools, notifications, write-back, and scheduled execution wired into one path.
Reusable AI Ops methodology
Turning a team-of-one practice into frameworks other people can run without me.
Keep a human in the loop is not a design. Which human? Doing what? At what point? Because of what risk?
Contact
Let's get into it.
Ambiguous problem, AI adoption that stalled, an operating model that stopped scaling, a support function that should be a product. That is the conversation I want.