You Don’t Need to Become an AI Expert to Ship Software
I built a serverless AI application last weekend. Not because I suddenly became an AI engineer — and not because I finally found a free weekend with nothing else going on. I built it between kids’ activities, nap times, and after bedtime. Maybe five to seven hours of actual work, spread across two days. The surprising part wasn’t that I built it. The surprising part was how little I actually needed to become an AI expert to do it.
The Advice I Keep Hearing About Becoming an AI Expert
If you spend any time online right now, it feels like everyone is telling you to learn AI. Learn prompt engineering. Learn agents. Learn MCP. Learn RAG. Learn vector databases. Learn — everything.
It’s overwhelming. And if you’ve already spent twenty years becoming an excellent engineer, it’s easy to feel like you’re starting over.
I don’t think that’s true. AI didn’t replace my expertise. It amplified it.
What Was Actually Hard (It Wasn’t Becoming an AI Expert)
The hardest part of building the application wasn’t writing code. Claude generated most of that. The hard parts looked surprisingly familiar:
- Deciding which problem was actually worth solving.
- Breaking the workflow into logical steps.
- Recognizing when AI misunderstood what I meant.
- Knowing which shortcuts were safe, and which would become technical debt.
- Debugging the inevitable failures when everything worked locally but not in production.
Those aren’t AI skills. They’re engineering skills — the same judgment I’ve developed over years of building software. AI simply let me apply that judgment much faster than I could have alone.
Put another way: it’s one thing to know how to prompt Claude. It’s another to know what to think through before you prompt it. It’s one thing to know how to deploy. It’s another to know that deployment is exactly where problems you couldn’t have predicted show up.
AI didn’t replace my expertise. It amplified it.
You Don’t Need to Become an AI Expert — Expertise Compounds, Knowledge Doesn’t
Knowledge is becoming cheaper every day. You can ask ChatGPT how to write a serverless function. Claude can scaffold an entire project. But none of those tools know whether you’re solving the right problem, what “good” looks like, or which tradeoffs matter. Experienced engineers do.
Here’s what that looks like in dollars and time. Twenty years ago, the first prototype of Mint.com — one that could download transactions and had a decent UI to show them — cost about $50,000 to build: three engineers (one acting as a designer), nine months. It was good enough to raise another $250,000 from investors.
Even the prototype I built for BizeeBee in 2010 — a CRM for fitness studios — took three months, $25,000, two engineers, a designer, and a PM.
This time, I could build a prototype by myself, in the cracks of a weekend with three kids. (To be fair: I’ve used AI tools for about a year now. But the whole vision, start to finish? I only had a weekend to execute it.)
That’s why I think we’ve been asking the wrong question. Instead of “How do I become an AI expert?” — try “How can AI help me apply the expertise I’ve already spent years building?”
Build From Your Expertise, Not an AI Expert’s Roadmap
The application I built wasn’t random. It solved a problem I’ve been thinking about for years: helping experienced technical leaders uncover expertise they’ve stopped noticing. That expertise came first. AI came second.
I didn’t wake up wanting to build an AI app. I woke up wanting to solve a problem I cared about, and AI happened to be the fastest way there. The opportunity isn’t to abandon everything you’ve learned and reinvent yourself as an AI expert. It’s to combine decades of experience with a new set of tools — a much smaller leap than most people think.
The Market Isn’t Calling for AI Experts — It’s Calling for This
There’s a term for this floating around right now: a High IC. Not someone who just writes good code — someone who can spot a real problem, build a proof-of-concept that proves it’s worth solving, push it all the way to production, and get actual people to adopt it.
That might mean shipping it across your org, if you’re inside a company. Or it might mean bootstrapping it yourself. Either way, you need something real to point to — not a resume line, an actual thing you took from idea to adoption.
What I Actually Built
I published a detailed, step-by-step walkthrough of exactly how I built the tool — from the first prompt to deployment, including the bugs, the real dollar cost ($0.57 in AI API usage + $100/month Claude subscription), and the mistakes along the way. It’s a prototype, a proof of concept — there’s still a long way to go before it’s a polished, finished product. But it’s a real start, and I’d rather share it honestly at this stage than wait for “done.”
If this resonated, you might also like a few related pieces on the same theme: just-in-time learning, why nobody’s coming to define your role, and what it actually took to propose and ship a project at Apple.
Expertise compounds. Knowledge doesn’t.
You don’t need to master every layer before you begin. You need enough expertise in one area to recognize what good looks like — and enough curiosity to learn the rest as the project demands it. If you’ve been waiting until you “know enough AI,” maybe this is your sign to stop waiting.
Try Uncover for yourself, or read the full build if you want to see exactly how it came together. And if you’re an experienced engineer thinking about turning what you know into something real, Ship It is built for exactly that.
