The other 80% · AI Bits and Pieces

The other 80%

I have a joke at work. I call it "the other 80%."

Since I started using AI for documents, code, and other projects, I've noticed it scaffolds so fast — and so densely — that it's created a strange paradox around the old 80/20 rule. This joke applies to plenty of other things too, but let's stick with AI.

What does it mean?

a) It's amazing how quickly you can get almost there the 80%, yet how slowly you actually get there. Hence "the other 80%."

b) After scaffolding, it feels like you're 80% done. In reality you may be closer to 20%. What you made may look wonderful and seem wonderful, but hold it up to a quality, context, and clarity lens and often it's more sizzle than substance.

c) By the time you've reviewed, updated, and simplified everything, you're drained (but satisfied) — and your tokens are gone. 🙂

**Have I solved it?**

Not exactly. But here are a few patterns that are helping:

**1) Create lots of skills.** Skills travel better than memory. My rule now: anytime I have to haggle with the AI more than once or twice over the same thing — a missing font, the wrong tone in a document, an ugly table — I stop and say "record what you just did to fix that into a skill, so you know better next time."

**2) Create lots of support docs.** AI is great at this. If you're reading this here then you probably already know about markdown. When I'm working on something, or building a spec for an app, I'll tell it at frequent intervals to update the docs and the skills. That keeps those files relevant and lets the AI capture the small details as it goes.

**3) Build help and instructions into the thing itself.** When I scaffold, so much gets created that it overflows my mental model, fills me with doubt, and slows me down. Now, if I'm building an app, my pages get little "?" bubbles with the what and how of a given thing. For example, at work right now I'm analyzing phone data for abandoned calls — accounting for short abandons, filtering out internal calls, outbound, extension-to-extension, voice mails. Who can remember all that detail a month from now? Build it in. Then as you iterate, your own notes show up in easy-to-find places. If you're having AI build slide decks, ask it to populate the notes fields so later you can re-absorb your original intentions.

**4) Have AI scaffold a user manual** at the beginning of the project rather than at the end - read it and be surprised what your project can do, lol, and update it every time you add a new feature or update something. At the end, you'll have a nice feature complete guide. If that's not enough, have it create an outline or a short slide deck to explain what the heck it built that you are going to put your name on.

**5) Have AI track a change log.** If you're using a file system / second brain approach, use git for your files and have the AI make commits to keep the history.

**6) document your larger patterns and have it use those when scaffolding - for instance - "if you need to use a bar chart to represent trends, if it's a web page, make sure that each slice can be clicked on to show a drilldown of the elements that the slice represents.

**7) stay true to your original scope - my problem sometimes is building to 110% instead of the 80% first version or rough draft that I thought I was starting with. The AI can build so swiftly, it lets your mind really run wild and expanding my scope because - well because "I can" is hard to resist.

**8) maybe this should be first - start with a better plan or definition of what is the outcomes you want with specificity. If you have moments like me when you say, "I'll know it when I see it", then get ready for the other 80%.

Next up: refining "my style," "my voice," and my UI preferences so it gets those things right out of the box more often — and I spend less time refining.

curious if others work this way and have similar patterns/issues or if your work is reliably 80/20?

5 comments

The other 80%