I Over-Engineered Everything at Apple. AI Made It Worse.

If you've been living under a rock: AI got really, really good. We're in 2026, and a whopping 95% of engineers use AI tools weekly. Nearly half of all code being written comes from models. Writing code by hand is heading the way of punch cards. Life should be easier now.

Should be.

AI Can Do WHAT Now?

Agents can plan, edit, test, and deploy features (across entire repos) while you're grabbing coffee. Claude Code is barely a year old and it's already the #1 dev tool. Harvard Business School found that AI cuts writing tasks from 87 minutes to 22 minutes. That's 75% faster. And podcasts? Tools like NotebookLM and ElevenLabs' GenFM can turn a 20-page PDF into a polished two-host audio discussion. In minutes.

The ceiling for what one person can ship has never been higher.

For most people, that's liberating. For perfectionists? It's gas on a fire.

Enter the Perfectionist

I spent the first 14 years of my career at Apple. If you know anything about Apple's engineering culture, you know the phrase: "paint the back of the fence." The idea being that even the parts nobody sees should be beautiful. I once worked with a team to shave 10ms off a function call to save a few seconds of test time. Because that was the standard. And I loved it.

That mentality works great when you're shipping millions of products with world-class quality. It works less great when an AI agent is whispering "I found something else" every few minutes.

The Loop of Doom

Here's the thing about language models that perfectionists need to understand: they're non-deterministic. Every run takes a different path through your codebase. Every run surfaces different findings.

Let’a say you finish a feature and run a code review agent. It catches a few bugs, flags some missing tests. Great! You fix everything. Clean bill of health, ready to ship. But better run it again, just in case.

New issues. Different paths. A refactor suggestion you hadn't considered.

Run it again. More warnings. An edge case three layers deep.

For a normal person, this is mildly annoying. For someone with years of "paint the back of the fence" burned into their DNA, this is a slot machine that never hits the jackpot. There are always more gremlins. The orange is never fully squeezed.

And then come the five deadliest words an AI agent can say: "Now I've got the full picture."

You know what that means. It means the agent just found a new thread to pull. It means another hour. Another mass of tokens burned. Another round of "there's a better way to do this." Each cycle feels productive. Each cycle is technically an improvement. And each cycle is pulling you further from the only thing that matters: shipping.

The Result

Thanks to AI, my code has never been more over-engineered. The back of the fence has never looked so good. Most of those code paths will never be exercised. But hey, the agents think it's in great shape.

Burning tokens? Fine if you're on a monthly plan or OPM (other people's money). On usage-based billing? Forget it.

The line between stable software and software that has error-handling routines for error-handling routines has completely blurred.

Embracing the Chaos

Perfection is an asymptote. You never reach it. But AI makes the asymptote feel so much closer: there's always something to fix, always within arm's reach. Just one more review. One more pass. One more step toward an ideal that doesn't exist.

Here's what I'm learning: the essential skill in 2026 is no longer "how do I use AI?” It's knowing when to stop.

Use AI to do this, not that. Ship the feature. A few bugs keep things interesting. Actual issues will be found…and they'll get fixed once somebody notices enough to complain.

Will I still catch myself painting behind the fence because it feels like the right thing to do? Probably. Am I practicing restraint and working on leaving well enough alone? Trying.

Steve Jobs also said: "Real artists ship." Same guy who built the culture that made me this way. Turns out both lessons were always there.

The back of the fence still deserves a coat of paint. Maybe just one coat instead of three.

Want to get the most out of ChatGPT?

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

Download the free guide