Andy University coding agents for real beginners

Guide Note 4

Improve One Thing At A Time

Improve one workflow with evidence instead of vague intuition.

Open the short practice promptBack To Start Here

Short Version

What this means if you are still new.

Choose one thing that matters right now: call prep quality, outreach clarity, or how fast you can turn notes into action.

Example

Sales example: define what a strong opening message means, generate ten versions, score them, and keep the winner instead of trusting vibes.

Open a short practice prompt Use this to feel the lesson without reading the full notes first.

Pasteable Prompt

Try this today

Help me write a tiny scoring rubric before we make variants.

The workflow I want to improve is: [fill this in]
The outcome I care about is: [fill this in]

Keep the rubric short, concrete, and usable by a beginner.

This is the shortest possible action that lets you feel the lesson in practice.

Longer Notes

Open the full guide notes only when you want more depth.

This keeps Andy's original longer write-up available without forcing all of it onto a first-time visitor.

Open the full notes for guide note 4 Andy calls this a quality loop. This is the longer original version.

You care about quality. You iterate. But if you don't track what changed, why, and whether it was actually better — you never develop calibration. You get better at producing outputs but not at knowing which outputs are better.

Quality loops fix this.

The Structure

Every metric that matters should have an active loop:

seed → instrument → variants (N≥10) → score → promote winner → re-seed

The critical step most people skip: instrument before you run variants. Define the metric BEFORE you try to optimize it. If you choose the definition after seeing the results, you'll unconsciously pick the one that makes your favorite version win.

A Complete Loop

1. Seed — pick one piece of work where you want to improve a specific quality. "The opening hook of my weekly summary."

2. Instrument — define the metric before writing any variants. "A good hook makes the reader want to know one specific answer. I'll score each variant on: (a) does it ask a specific question? (b) is the answer one only I could give? Scale: 0-2 for each."

3. Variants — write 10+ versions, each trying a different approach. Stay generative. Don't evaluate while you're writing.

4. Score — apply the instrument to each variant. The highest scorer wins, not the one that feels most polished.

5. Promote winner — use it. Let it run.

6. Re-seed — after the winner runs and you can see how it performs, bring what you learned back to the instrument. Does the metric still make sense? Update and run again.

The No-Backlogs Principle

Don't maintain a list of things to improve. If a problem matters, it resurfaces — that's the signal to run the loop. If it doesn't resurface, it wasn't worth fixing.

A backlog is a graveyard of things that felt urgent once and weren't. One active loop on one metric at a time, done well, is more valuable than a list of fifteen half-started ones.

What This Builds

The calibration record — what worked, why, over how many cycles — becomes the compounding asset. One loop is a test. Years of loops is a defensible position.

Practice

Identify one quality metric that matters to you right now. Write the instrument. Run 10+ variants. Score them. Promote the winner.

Completion criteria: An instrument file exists (written before variants). 10+ variants scored. One winner promoted.