NNoona
The Flywheel · What happens when workflows run themselves

Climb the ladder.
The top step spins by itself.

From doing the work, to watching the work do itself. Six steps from a pipeline to a self-improving flywheel, explained for people who have never heard the words "closed loop" before.

Step 01 / 06
Pipeline
01
02
03
04
05
06
01 · Pipeline
A → B → C. A straight line.

A pipeline does the same thing every time, in the same order. Like a dishwasher cycle. No decisions, no surprises.

Real example: A nightly script that exports your sales as a CSV and emails it.

↓ Scroll to climb
What just happened?

Same idea, in a kitchen.

Pipeline
A recipe

Step 1, step 2, step 3. Follow exactly. Same dinner every Tuesday.

Workflow
A kitchen with choices

Out of basil? Use parsley. Pan too hot? Lower it. You still make every decision.

Flywheel
A kitchen that learns

Tastes the dish, rewrites the recipe, restocks the fridge, plans tomorrow's menu, by itself.

Glossary

Six words. Six pictures.

Pipeline

A fixed sequence of steps. Same input, same output, every time.

e.g. A nightly export job.

Workflow

A pipeline with forks. Someone (or something) chooses the path.

e.g. A support ticket router.

Loop

A workflow whose output feeds its own input. It runs in a circle.

e.g. A thermostat.

Verifier

The judge inside the loop. Decides if this turn passed or failed.

e.g. A test suite.

Flywheel

Generate → measure → decide. A loop that picks what to try next.

e.g. A coding agent retrying tests.

Self-improvement

The flywheel rewrites its own steps when the verifier says it can.

e.g. An ad system that invents new copy nightly.

The rule

Verification comes before autonomy.

A flywheel without a verifier is a runaway machine pointed at the wrong goal.

Every spin of a flywheel multiplies whatever is inside it. If the verifier is sharp, the system gets better. If the verifier is missing, or measuring the wrong thing, the system gets faster at being wrong.

That's why the order matters. First you make sure the system can tell good from bad. Only then do you let it run on its own.

In the wild

Three flywheels already spinning.

Coding agents
Write code → run tests → fix → retry
Verifier: the test suite (pass / fail).
Ad optimization
Generate variants → measure CTR → kill losers → scale winners
Verifier: click-through rate and conversion.
AI research labs
Propose experiment → run → read results → propose next
Verifier: benchmark scores, loss curves, eval suites.
Noona on this ladder

Noona lives between step 5 and step 6.

Most tools today are pipelines or workflows wearing AI hats. Noona is the layer that closes the loop on top of your computer, it watches what you do, encodes the verifier with you, and starts spinning the wheel on the work you already explained once.

The day the verifier is good enough, Noona stops asking and starts improving. That's step 6.