I’ve been watching something disturbing happen across the tech industry for the last eighteen months. And I’m not talking about layoffs. The real damage is invisible. It’s happening inside engineering teams that think they’re being smart. Here’s what I see: companies are rushing to put AI between their engineers and the work. Code reviews are automated. Architecture decisions are outsourced to LLMs. Junior devs are told “just ask Copilot” instead of being mentored. Senior engineers are spending more time writing prompts than writing code. And the result? The work gets done faster for about six months. Then the best people leave. I’ve been thinking about this since I read the data that came out last month. Engineering jobs aren’t dying. They’re actually more resilient than anyone predicted. But the quality of those jobs is being hollowed out. And that’s a problem no AI can solve. Step 1: Stop Using AI as a Replacement for Judgment The first mistake I see everywhere is treating AI like a senior engineer. You wouldn’t let a junior dev ship code without a human review. But companies are shipping AI-generated code with zero human oversight because “the tests pass.” Practical tip: Never let AI make a decision it can’t explain to a human in under 30 seconds. If your team can’t articulate why the AI’s suggestion is correct, they don’t understand the system. And that’s how technical debt compounds. Common pitfall: Thinking “it works” is the same as “it’s right.” It’s not. I’ve seen AI generate perfectly functioning code that creates security vulnerabilities, introduces subtle race conditions, or violates architectural principles. The tests pass. The system breaks six months later. Step 2: Protect the Craft, Not Just the Output Here’s the brutal truth: AI-mediated engineering is creating a talent gap in tech that nobody’s talking about. Not a gap in quantity of engineers. A gap in quality. When I built the Mac team, I didn’t hire people who could write code faster. I hired people who cared about the code itself. The elegance. The structure. The way a function reads like a sentence. AI is destroying that. Engineers are becoming prompt engineers, not software engineers. They don’t learn to debug because the AI writes the fix. They don’t learn to design because the AI suggests the architecture. They don’t learn to feel when something is wrong because the AI never hesitates. Practical tip: Mandate that every team member spends at least 20% of their week writing code without AI assistance. No autocomplete. No Copilot. Just them and the problem. This is non-negotiable for anyone under five years of experience. Common pitfall: Assuming that AI-assisted coding is “the same as” learning to code. It’s not. It’s like learning to drive with a chauffeur. You’ll get where you’re going, but you won’t know how to handle a flat tire. Step 3: Measure Learning, Not Velocity The second thing I see is how companies measure engineering productivity. Lines of code? Story points? PRs merged per week? All wrong. The only metric that matters for long-term health is: is your team getting better? AI makes teams faster in the short term. But it makes them dumber in the long term. Because the learning loop is broken. When a human writes bad code, they feel the pain. They debug it. They learn. When AI writes bad code, the human just tweaks the prompt. Practical tip: Run a “no AI” sprint once per quarter. Full week. No generative tools. Watch what happens. The team that struggles is the team that’s been outsourcing their thinking. Common pitfall: Celebrating velocity while ignoring atrophy. Your team ships faster this quarter. Next quarter, they can’t debug a production issue without asking an LLM. That’s not efficiency. That’s dependency. Step 4: Build the Culture That AI Can’t Replicate I spent a decade at NeXT learning this lesson the hard way. You can’t automate trust. You can’t automate mentorship. You can’t automate the moment when a senior engineer looks at a junior’s code and says “this is beautiful, but here’s why it’s wrong.” AI-moderated engineering is a tool. But if you let it become the culture, your best people will leave. Not because they’re afraid of AI. Because they’re bored. The engineers who stay at a company for ten years aren’t staying for the salary. They’re staying for the craft. The challenge. The feeling of building something that matters. Practical tip: Create spaces where AI is explicitly banned. Design reviews. Architecture discussions. Postmortems. These need to be human-only conversations. Common pitfall: Assuming that because the AI is “helpful,” it’s harmless. The most dangerous tools are the ones that make you faster while making you worse. The Bottom Line I spent my career trying to build tools that amplified human creativity, not replaced it. The Mac. The iPhone. These weren’t machines that did your thinking for you. They were machines that let you think better. AI can be that. But only if you’re intentional about protecting the craft. The talent gap in tech isn’t about finding more engineers. It’s about not destroying the ones you have. Don’t let the machine eat your team’s brain. Because once it’s gone, no amount of prompts will bring it back.