About
I help engineering teams put AI coding agents to work without betting the codebase on them.
I am an engineer who builds agentic development workflows for a living: Claude Code setups, multi-agent teams, and the guardrails that make them reliable enough to trust with real work. If you are weighing whether to bring agents into your stack, this page is here so you know who you would be working with before you fill in a form.
How I approach the work
The hype around AI coding moves fast and most of it does not survive a production codebase. These are the principles I keep coming back to.
Agents do the grunt work, you keep the judgement
AI is good at the repetitive, well-scoped parts of shipping software and bad at owning ambiguous decisions. The setups I build keep humans on the calls that matter and hand the rest to agents.
It has to survive contact with your codebase
A demo that works on a toy repo is not the goal. The work is judged against your real CI, your review gates, and the legacy modules nobody wants to touch.
Boring plumbing beats clever prompts
Most of the reliability comes from the unglamorous parts: issue queues, role boundaries, guardrails, and clear acceptance criteria. I spend more time there than on prompt wording.
How an engagement runs
Most engagements follow the same shape, scaled to the size of the problem. No long discovery phase before anything ships.
- 01
Scope a real problem
We start with something concrete you are already trying to ship, not a hypothetical. I look at how your team works today and where agents would actually save time rather than add ceremony.
- 02
Stand up a working setup
I build a Claude Code or multi-agent setup against your codebase: prompts, skills, MCP servers, CI hooks, and the review gates that keep it safe to merge.
- 03
Hand it over with the reasoning
You get the working system plus the why behind each choice, so your team can extend it after I step back. No black boxes you cannot maintain.
What to expect
You can expect plain answers about what agents can and cannot do for your situation, including when the honest recommendation is to not bother yet. I would rather tell you a workflow is not worth automating than sell you one that creates more review work than it removes.
Named case studies are on the way as clients agree to be referenced. Until then, the proof is in the approach and the working setups you walk away with, not in a wall of logos.
Think your team would ship faster with agents in the loop?
Tell me what you are working on and what is in the way. I will tell you whether I can help and what a sensible first step looks like.
Start a conversation