Controlled AI shipping

Ship as Superman. Release as Kent.

AI writes the change; Kent gates it, rolls it out, and kills or heals it — no redeploy.

The gap between an AI coding tool and a flag suite. The model can write the diff. A flag can ramp a change later. Kent is where the change is born already gated.

Open the console Read the MVP memo

Sample readout

Widening

Checkout trust line

kent_one_line_trust_statement

src/app/checkout/page.tsx

of traffic

Born gated at 0%. This sample has been widened once, to a small rollout. The new line is inside the flag. Kill would send checkout back to the previous copy, with no redeploy.

--- a/src/app/checkout/page.tsx
+++ b/src/app/checkout/page.tsx
@@ -24,7 +24,15 @@ export function Screen() {
   return (
     <section>
-      <p>{copy.current}</p>
+      {flags.kent_one_line_trust_statement ? (
+        <p>encrypted, and you can leave and come back.</p>
+      ) : (
+        <p>{copy.current}</p>
+      )}
     </section>
   )
 }

Brief. Gate. Then widen, hold, or kill.

01
Brief
Say what should change, and the file or route where it lives. Kent starts from the behavior you want in production, not from a blank pull request.
02
Gate
The change comes back as a diff. The new behavior sits inside a flag. It is born at 0%. Shipping the code does not turn it on.
03
Widen, hold, or kill
Step traffic through 0, 5, 25, 50, 100. Hold leaves the slice where it is. Kill sends every request back to the previous behavior, with no redeploy. Heal restores the pre-kill slice.

The gap

Kent is not a better coding agent, and it is not another flag dashboard. It owns the moment between them.

Writing a diff

An AI coding tool is done when the patch exists. Merge it and the new behavior is on for everyone, or it stays on a branch. There is no 5%.

Flipping a flag later

A flag suite can widen, hold, and kill. It starts after someone has already wrapped the code, named the key, and shipped the guard. The gate is a second project.

Kent

The change and the gate are one object. The brief produces a flag key, a diff that checks it, and a rollout that begins at zero. Widen, hold, kill, and heal are the release.

Beachhead

AI-native Next.js and TypeScript teams

A small team that already lets a model write the change, shipping a real Next.js product, without a flag platform in the middle. The risk is not typing. The risk is a change that is right in the editor and wrong the moment it is on for every customer.

That is the first user. Platform teams that already run a flag service, and teams that do not let a model touch the repo, are not the beachhead.

Gate one change in this browser.

The console is the weekend demo. It gates a brief locally, then lets you widen, hold, kill, and heal. Nothing is deployed.