Agents don't need a dashboard. They need to reach you.
Run a fleet of Claude Code agents and the bottleneck becomes you. Why a passive dashboard isn't enough, and how a reach-back channel closes the loop.
Writing
Notes from running this in production. How I think about engineering, AI delivery, and building products solo. The method, the misses, the fixes.
Building Products
How I design, ship, and grow products. Mostly solo, mostly in production.
Engineering Leadership
Running delivery so teams ship predictably without the heroics.
-
The executor changed. Become the machine.
Everyone tells engineers to move up as AI takes the code. Wrong: the spec and the review get eaten first. Here is where the work actually went.
-
You govern machines. You manage people.
40% of agentic AI projects die by 2027, not from weak models but from missing operating discipline. Here is the managed autonomy frame that actually holds.
AI Engineering
Using AI across the delivery lifecycle. The method, not the hype.
-
Formwork for agents: a reusable shape you can pour into any stage of your SDLC
LLMs moved the verifiability line into natural language. Why the requirements gate is the sharpest first pour for AI agents, and how to build the formwork.
-
Clean code is copper. You're stepping over the gold.
In an agent fleet, quality lives in the frame you write before execution starts. The QA engineer role isn't support anymore. It was always the whole job.
-
They Called QA Second-Class. Then AI Made It the Job.
The skills that got QA engineers dismissed now run every AI agent. Framing and validation are the whole loop. You did this before it had a name.