Notes on building enduring companies — product, engineering, AI, and the lessons in between.
The AI workflows worth automating first in a small business are the repetitive, rules-based, high-volume ones: quoting and estimates, customer support triage, lead qualification and follow-up, reporting and back-office data entry, and routine content operations. Start where a human is doing lookup-and-respond work all day — that's where an agent pays for itself fastest.
AI-generated apps usually break in production for the same handful of reasons: no real authentication, unhandled edge cases, secrets committed to the codebase, and no monitoring or backups. Here's how to triage what's broken, stabilize it fast, and harden it into a product that survives real users — usually without a full rewrite.
Before a vibe-coded or AI-generated app is safe for real users, it needs the things the demo never tested: real authentication and authorization, secrets out of the codebase, input validation, backups, error handling, monitoring, and a deploy pipeline. Here's the checklist we run every rescue against, so you can find the gaps before your users do.
In 2026, the cost of a minimum viable product is driven far more by scope, integrations, and production-readiness than by lines of code — because AI has made writing the code the cheap part. The honest answer isn't a single number; it's understanding the three things that actually move the price, and where a fixed scope protects you.
The hardest and most valuable thing a product partner can do is help you decide what not to build.
Execution is cheap and getting cheaper. Knowing what to execute is where the value moved.
Retention is a design problem long before it's a growth problem.
The goal of an early launch isn't to be finished. It's to start learning as fast as possible.
The best stack is rarely the newest one. It's the one your team can move fast in for years.
Agents are extraordinary at some jobs and quietly terrible at others. Knowing which is the whole game.
