The One Skill AI Can't Replace: Figure-Out-Ability
AI might write code, but it lacks figure-out-ability: the human grit and intuition to solve messy, real-world engineering problems.
You ask an AI for a complex SQL query and it returns one in seconds. Ask for a whole component and you get that too.
That’s enough to make you wonder whether it’s coming for your job. It’s a fair thing to wonder, and AI is missing something: the part that can’t be copied or scripted.
I call it figure-out-ability.
What Is Figure-Out-Ability?
Figure-out-ability isn’t a language or a framework, and you won’t find it on a certification. It’s the stubbornness that keeps you at a broken system at 2 AM because you’ve decided to work it out, the instinct for which question to ask when nobody else in the room has one, and the patience to turn a half-defined problem into something that works.
Writing code is the smaller part of the job. The rest is ambiguous requirements, a legacy system nobody documented, and finding out why production falls over while staging is fine.
AI answers a question well once you know which one to ask. It doesn’t know your codebase, your team, or why the database chokes on Tuesdays.
Why This Keeps You Relevant
If your job is boilerplate and simple CRUD screens, the worry is justified. AI does that work faster and cheaper than you.
Context decides more than code does. A model works in a vacuum: it doesn’t know that your marketing team rewrites requirements every week, or that the database struggles under a particular load. You know those things, and that knowledge is the difference between a solution that fits and one that is only technically correct.
Figure-out-ability is what puts the pieces together: taking the script AI generated, adapting it to your architecture, deploying it, and putting monitoring around it.
The tool you learn now will be replaced. Learning its successor is what carries over.
How to Build It
The next time something breaks, resist pasting the error into a model straight away. Read the stack trace, check the documentation, and form a hypothesis before you ask. The struggle is where the skill comes from.
Volunteer for the projects nobody has scoped yet. The ones where no one knows where to start teach you the most, and the discomfort is the price of learning them.
Use AI to argue with. Ask it questions, push back on its answers, let it sharpen your thinking. The decision stays yours.
What AI Won’t Do
Learn to use AI well. It is good at a lot of this work.
What it won’t do is stay with a problem that doesn’t want to be solved. It doesn’t keep going when the documentation is empty and search returns nothing, and it won’t hack something together from the scraps. Those are the moments that decide whether a system ships.
The job was always solving problems, with code as one of the ways to do it.