Substance Over Optics
Leading a team through the AI wave is hard, but not always for the reasons people assume. Technology evolves, but the harder part is distinguishing substance from optics.
Over the past few years, the trend has shifted constantly. Copilot gave way to ChatGPT, then agents, then loops, and each wave brought the same pressure to adopt it, ship faster, and show we were keeping up. That pressure is real, but what I see on the ground is different. The people I work with are already using AI where it helps. They’re already at capacity, while maintaining systems built from yesterday’s quick wins.
AI matters, but the question is what we optimize for when that pressure is everywhere. As engineering leaders, our job isn’t to chase whichever narrative is loudest right now. It’s to make sensible decisions about what actually helps the team and what protects the culture that makes good work possible.
Trusting Their Judgment
Good leadership starts with trusting your team. Let them decide what they’re comfortable using, and give them room to change their mind when something better comes along. Different people work differently, and different problems need different tools. Nobody needs a mandated AI stack from above just because the latest demo looked impressive.
AI adoption doesn’t automatically mean better outcomes. The goal isn’t to extract more productivity out of people. What matters is whether the team still has the mind to make good decisions about design, trade-offs, and what deserves to ship. Forcing adoption only makes sense if you think people are holding back. If you’ve hired well, they aren’t.
Protecting the Culture
When you chase the narrative instead of making sensible calls, the damage doesn’t stay with leadership. It hits the team and the culture that holds it together.
AI can erode engineering culture quickly. That tends to happen when it’s used without thought, or when adoption is mandated from the top down and the people doing the work are pressured to ship more feature to highlight the increase in productivity. AI makes it easy to generate code, but generating code isn’t the same as designing architecture. It’s akin to building a skyscraper using only bricks. That might work for a small and well-defined SaaS product. In a large system, it eventually falls apart. You might move fast for a while, but you’ll always end up paying the tech debt.
When output becomes effortless, teams start flooding themselves with more PRs, more changes, and more solutions than anyone can properly review or reason about. I call it an AI DDoS. The volume overwhelms the discipline that keeps a team strong. Code review gets rubber-stamped while design discussions get skipped, and nobody has to think through the work. So the knowledge are lost forever.
Leadership’s job is to hold the bar, not celebrate features shipped.
Outsource the Work, Not the Understanding
When institutional knowledge isn’t kept within the team, decisions take longer because nobody fully understands the system. Eventually the team stops being able to operate on its own. That’s the real risk of getting AI wrong.
AI can take on the work, like the boilerplate, the drafting, and the repetitive parts. What it can’t take on is the understanding. Institutional knowledge needs to stay with the people doing the work, not inside a tool. It doesn’t matter which AI they used to get there. What matters is that they can still explain what was built, why it was built that way, and reason through problems when things go wrong.
If someone can’t defend a decision because “that’s what the AI said,” you’ve crossed that line. That’s not augmentation. That’s abdication.
For me, good AI in a team passes a simple test. Remove it, and you should still be able to operate the system. You’d be slower, but nothing falls over, because the understanding stayed with the people. If unplugging AI means critical workflows stop working or nobody can explain how the business runs anymore, you’ve outsourced too much.
Reflection
The hype will keep coming. The discipline is knowing what actually matters.
For me, engineering leadership in the age of AI comes down to trust and culture. Outsource the work, not the understanding.