How to run an AI agent in shadow mode before granting production authority
A six-stage, 16-week-or-shorter playbook for collecting production evidence while humans retain consequential decisions and rollback remains available.
Build the company that runs on AI.
A publication for founders and operators rebuilding how their companies work. Every claim carries the evidence a reader can check, with editorial responsibility stated plainly.
Featured
Latest
What a week of verifying our own verification cost
A founder’s decision framework for separating checks that work from checks that merely pass.
What to require before an AI coding agent can merge code
A minimum merge path for protected branches, bounded agent permissions, fresh checks, and recoverable releases
What Klarna’s AI numbers can and cannot tell you
The public record shows reported scale, service indicators, savings, and staffing—but not AI’s causal effect on any of them.
How the work is done
Company Rewrite
A specific company function, taken apart and rebuilt around what AI can actually do today. Not a prediction — a redesign, with the constraints and the failure modes left in.
Playbook
A procedure you can run, with the preconditions it assumes, the steps in order, and the way you check that it worked.
How to check us
Sources
Every article lists the public evidence behind its claims, closest to the fact first. Where a claim rests on our own measurements, it says so.
Editorial method
Agents assist research, drafting, and review. FounderCLI Editorial edits the work, and a governor is responsible for every final revision. The full method is on About.
Corrections
Material corrections and retractions are published at /corrections. A withdrawn article keeps its address and says what happened to it.
Follow the work
The newsletter is not open yet. Until it is, the RSS feed carries every published article, and the community page explains who this is for and what joining will involve.