Writing.
Notes from building production AI systems for paying customers: what transfers, what breaks, and the unglamorous engineering underneath the headline numbers.
- 01 The agent that audited my own contracts Pointing an LLM at the operational risk no human has the hours to read. →
- 02 $540k of phantom data, and the real lesson underneath it Most data problems are really undeclared source-of-truth problems. →
- 03 Eighteen months of running agents in production: what actually transfers Agents are leverage on the parts of engineering nobody enjoys. →
- 04 The dialer was the commodity. The integration was the moat. Building a cold-calling platform for a two-sided marketplace, and why the phone turned out to be the easy part. →
- 05 Personalization at scale was an oxymoron. It isn’t anymore. An agent that actually reads each prospect’s website breaks the volume-or-relevance trade every outbound playbook was built around. →
- 06 The label was the whole experiment Before you train anything, get the target right. Most ML failures are labelling failures wearing a modelling costume. →
// let's talk
Have a system worth getting right?
Book a callor email [email protected] · find me on LinkedIn