Two years ago, automating a single ETL flow took me weeks. This week I built a full internal tool in five days. The real shift isn't better AI tools — it's the collapse of the cost of problem-solving, and what that does to the SaaS math.
#ai
6 posts
Essays
Buying a generalist chatbot raises the floor of competence. It doesn't build your edge. True value comes from bespoke systems that understand your specific business logic and data.
After nearly a thousand hours building with Claude Code, I've learned a few lessons the expensive way. These are my three non-negotiable rules for every AI coding agent.
AI has two modes: Cognitive and Operational. Work slop comes from treating everything like Cognitive Mode. The fix is knowing when to constrain AI to data-backed execution.
For weeks, the dream of bridging our underwriting models to the data warehouse felt like a fantasy. The breakthrough came when I stopped patching cracks and rebuilt the foundation.
The first version of our document routing agent hit 37.5% accuracy. The problem wasn't the AI — it was building probabilistic inference into a step that needed to be deterministic.