Most developers are using AI to write a SQL query.
Maybe clean up a function, or knock out a unit test they didn't feel like writing.
It saves them ten minutes and they feel like they're living in the future.
Ten minutes! Congratulations.
That's the whole problem. We have a tool that can architect, plan, and build an entire project, and most developers are using it like a fancy autocomplete.
Small Questions Get Small Answers
There are two kinds of people at every job.
One kind shows up every day trying to get a little bit better. A little faster, a little cleaner, shave a few minutes off the thing they already do. The other kind stops and asks how they could do the whole thing 10x better.
Those are completely different questions, and they lead to completely different careers.
You don't get a 10x leap by tweaking. Tweaking gets you 10%. To get 10x you have to reimagine the whole thing, rip it up, and ask what you'd build from scratch today knowing what AI can actually do.
Most developers are asking AI the 10% question. "Write me this query." The 10x question is "how would I build this entire feature if I let AI do the whole thing?"
Ask a small question, get a small answer.
Stop Micromanaging the Machine
Let me give you a real example.
At Full Scale, our own team spent six weeks trying to build the "perfect" Model Context Protocol (MCP) server for one of our internal tools. Six weeks. When they finally handed me something to try, it didn't work right.
So I rewrote the entire thing myself in two days.
I didn't really write it. I asked AI to scan our codebase and tell me the top things we should expose through the MCP server. It came back with a list. I spent a few minutes deciding how I wanted it architected and which of those endpoints made sense. Then I told it to go build the whole thing.
The first version was buggy. So I told it to test every single endpoint, find all the bugs, and fix them. It did.
I never even looked at the code.
Read that again. Our team spent six weeks and shipped something broken. I spent two days making decisions and let AI do the building, and it worked. The difference wasn't talent. It's that I stopped trying to micromanage the machine and asked it to do something big.
Most people are still handing AI tiny tasks and checking its homework. Ask it to do big things.
Let AI Interview You
The best part of that project was what my job shrank down to: making decisions.
That's the whole shift. Instead of writing a perfect spec up front, tell AI what you want to build and have it interview you. Let it ask the questions. How should it handle errors? What happens in that weird edge case? Which of these two approaches do you actually want? You answer, it builds.
You'd be amazed how many decisions you never wrote down because you were carrying them around in your head. AI drags them out into the open in twenty minutes.
Your job as the engineer is the decisions only you can make. The architecture, the tradeoffs, the calls about how the thing should actually behave. Let AI handle everything downstream of that.
Ship code as fast as you can make the decisions and validate the work.
That's the goal. If writing the code is still your bottleneck, it's time to rethink how you're using AI.
You Don't Like How AI Writes Code? That's On You
Half the developers I talk to swear AI writes garbage code.
Odds are that's your fault.
If you don't like how AI writes code, it's almost always because you never told it how you want your code written. No files describing how you architect things. No rules for how you organize a project, name things, format, which patterns you use and which ones you ban. You gave it nothing, and now you're mad it guessed wrong.
This is the most valuable thing your team can build right now, and almost nobody treats it seriously. Your coding standards, your architecture patterns, your conventions, all written down in markdown files that AI reads every single time it touches your code.
Treat those files like real code. Version them. Improve them every week. When AI does something you don't like, don't just fix the output, fix the rules so it never does it again.
I've personally built 43 skills for the marketing content and research I do. Every single day I add new rules and guidelines to them.
Improving 1% a day adds up fast.
Remember my MCP server? If we'd already had our standards written down, AI would have just followed them.
Build a Wiki Your AI Can Read
Standards tell AI how to write code. Documentation tells it how your system actually works, and this is the investment almost every team is skipping.
Most codebases have the real knowledge locked in three people's heads: why this service talks to that one, why someone made that weird call back in 2023, and what actually breaks when you touch the billing code. New developers spend months learning it the hard way, and it's even worse for offshore or remote engineers.
Write it down. Build up real wiki-style documentation, in your codebase or in a system AI can reach through its own MCP server. Then let AI both read it and add to it. Every time it figures something out about your system, it writes it down for the next person, or the next agent.
This takes real work up front, and it pays back every single day after.
You have to slow down to go faster.
The Job Changed
Every hour your best engineers spend hand-writing code AI could have written is an hour they didn't spend on the stuff only a human can do.
The real job now is building the standards, the docs, and the context, then handing AI something big and getting out of the way so it can do the writing.
Stop writing code. Start building the systems that write it.