$ ls ./writing
Notes from the build log.
Essays on shipping AI products, retrieval, agents, and the unsexy systems work in between.
- Oct 23, 2026· #engineering
How Anthropic designed Claude Duet: an engineering breakdown
For an agent that acts on its own, the hard decisions are not the model. Claude Duet shows two of them: how often to check your apps, which is a cost decision, and what the agent may never do, even with approval.
- Oct 17, 2026· #engineering
Why Anthropic replaced git with Scion: an engineering breakdown
Anthropic went from about 2,500 merged changes a day to more than 30,000. Git's assumptions broke first. This is what they built instead, and when it does not apply.
- Oct 9, 2026· #ai
How I handle retries in an AI agent without doing things twice
Retries are where agents cause real damage. I persist every step, put an idempotency key on every write, and only retry what is safe to retry.
- Oct 3, 2026· #seo
How I write posts so answer engines can cite them
I write posts as one question, a direct answer, and the check I actually run. I do not invent a statistic to sound quotable.
- Oct 3, 2026· #ai
Retrieval or fine-tuning? I start with retrieval
For products built on a user's own notes, I use retrieval first. I fine-tune only after the task is stable and the same error survives prompt and retrieval fixes.
- Oct 3, 2026· #ai
What I check before an AI feature is production
Before I call an AI feature production, I want a one-sentence task, a cost for a bad answer, a retrieval path I can inspect, ten frozen examples, and a cap on spend.
- Sep 26, 2026· #ai
What I Actually Look at When an AI Feature Feels Bad
When an AI feature feels inconsistent, I no longer immediately rewrite the prompt. I break the problem into measurable parts.
- Sep 26, 2026· #startups
The First Version of My Product Was Too Complicated
I kept adding architecture because I was solving problems that did not exist yet. Simplifying the system made it easier to ship and easier to change.
- Sep 26, 2026· #ai
My AI Agent Does Not Need to Be Autonomous All the Time
The more I build AI agents, the less interested I am in making them fully autonomous. Control, visibility, and good handoffs often matter more.
- Sep 26, 2026· #saas
Building SaaS Taught Me That Code Is the Easy Part
I started by thinking the hard part of SaaS would be engineering. Then users, billing, support, infrastructure, and edge cases showed up.
- Sep 26, 2026· #ai
I Spent Too Much Time Making My AI Agent Smarter
A lot of AI engineering is not about finding a better model. Sometimes the real problem is that the system around the model is badly designed.
- Jun 11, 2026· #ai
How to build production-ready AI agents: Beyond the deterministic loop
Most agent tutorials stop at a while loop that calls a model and some tools. That is a demo. I care about retrieval, a runtime that can resume and stop, evals, and limits on what a run can do.
- May 10, 2026· #linkedin
I Stopped Treating LinkedIn Like Social Media. It Became a Growth System.
I stopped posting on LinkedIn when I felt like it. Consistency, a clear niche, and a content system beat occasional viral posts.