How do AI agents help build a product portfolio?
The reason fifty properties is possible at all is that I stopped being the person who types the code. I am the person who briefs, dispatches, reviews, and verifies. The typing is done by AI coding agents — Claude Code and Codex, mostly — running in terminal sessions directly on the server and on my Macs, frequently several at once. On a good day the operation feels like running a small engineering department where every engineer is brilliant, tireless, instantly familiar with any codebase, and suffering from total amnesia and zero survival instinct. This chapter is about managing that department.
What the factory looks like
The unit of work is a brief. Not a vague wish — a written brief, the way you would brief a contractor: here is the goal, here is what already exists, here are the constraints, here is what done looks like, here is what you must not touch. The quality of agent output tracks the quality of the brief with almost embarrassing fidelity.…
The real failure modes
Let me tell you what actually goes wrong, because the marketing around AI coding never will, and every story here happened to me on production systems.
The discipline that fixed it
Every rule below is a scar with a date on it.
The honest summary
Agents multiplied my output by an amount I won't pretend to measure precisely — call it five to twenty times depending on the work. They also caused, by direct action, the worst outages this portfolio has suffered. Both facts are true, and the second is the price of the first only if you refuse to learn. The leverage isn't in the agents; it's in the management layer around them.…