IDF · Intent Driven Flow · 2026

A delivery framework where AI agents execute and humans govern outcomes. Work is organized around what the client needs to be true — not around what needs to be built.

Right Now
With IDF
Write a ticket describing what to build.
Write what the client needs to be true. Agents figure out what to build.
Estimate complexity before you start.
Learn it from the gate report after.
Deploy, then manage the panic.
Deploy behind a flag. Release when ready.
Measure velocity and story points.
Measure what the client actually received.
Sprint plan. Estimate. Stand up. Retro.
Intents close when the outcome is confirmed. Cycles close when the feature ships.

Agents execute.
Humans govern.

Feature flags separate "it's deployed" from "it's live." And an Intent doesn't close when the feature ships — it closes when the outcome is confirmed. Delivery and success are two different events.

The Driver and the Pilot

A seasoned Driver and a young Pilot stood before a gleaming, chrome vessel at the edge of the atmosphere.

The Driver held a leather-wrapped steering wheel, a heavy brake pedal, and a map of the local highways. "I've spent twenty years mastering the road," the Driver said. "I know exactly when to hit the gas to overcome friction, when to turn the wheel to stay in the lane, and when to slam the brakes to avoid a crash. I am ready to lead this mission."

The Pilot looked up at the black expanse of the stars and shook his head. "Friend," he said, "where we are going, there are no roads. There is no air. There is no friction."

What Actually Changes
01
You give outcomes, not instructions.
Tell the system what needs to be true for the client. Agents determine how.
02
Deployment and release are separate events.
Code ships behind a flag set to OFF. A human flips it ON when ready.
03
Intents outlive delivery cycles.
One goal can take multiple cycles to achieve. The Intent stays open — in monitoring — until the outcome signal is confirmed. The feature shipping is the middle, not the end.
How It Works

One goal. Six steps. Live feature.

01
Set a goal
02
Align first
03
Agents build
04
Human reviews
05
Go live
06
Did it work?
Step 01 — Intent
Set a goal, not a task list.
Write what needs to be true for the client, and how you'll know it worked.
Step 02 — Gate 1
Align before anyone builds.
Is this the right goal? Is the scope coherent? Nothing starts until everyone agrees.
Step 03 — Execution
Agents build. Humans govern.
AI agents execute. The human role is direction, not typing.
Step 04 — Gate 2
Read the report. Use the feature.
Human validates automated results and uses the feature — loop repeats until approved.
Step 05 — Gate 3
One decision: flip the switch.
The PO authorizes and the release switch activates.
Step 06 — Monitoring
Watch the signal. Close when it moves.
Value confirmed when the outcome signal moves.
What IDF Removes

What leaves your week.

Is IDF right for your team?
Good fit
  • ✓You want fast execution cycles — from intent to deployed, in hours
  • ✓You want to ship continuously, not in quarterly releases
  • ✓You want to achieve value faster and remove friction caused by human bandwidth
Not a fit
  • —Your stack requires all-or-nothing releases — incremental deployment isn't possible
  • —You're not willing to change the ways of working

"The artifacts carry the context. The agents carry the execution. The PO governs the outcome."

Intent. Cycle. Iteration. Three gates. One outcome loop. Read the framework →