There's a pattern forming in this industry that I don't think we're going to feel for another five or ten years, and I think that's exactly why nobody is treating it with any urgency. The more we integrate agentic coding into how we build software, the more siloed our engineers are becoming. Less time in the room with each other. Less reason to turn to the person next to you and ask how they'd think about a problem. A culture of learning that has been central to this profession for my entire career — and, I'd wager, well before it — is quietly being switched off, and we're not going to notice what it cost us until the people who could have absorbed it are already gone.
// The Gray Beards
When I was starting out, I was lucky. I had a handful of older engineers around me — gray beards, we called them — who took the time to actually work alongside me instead of just assigning me tickets. They were an odd mix. Some were classically trained software engineers. A couple were musicians who'd found their way into code sideways and never left. A few described themselves, without irony, as blue collar engineers — and they meant it as a compliment to themselves. They did the unglamorous work. The systems nobody thought about until they broke, the plumbing that kept everything else running while getting zero credit for it.
I worked side by side with those people for years. Not in a mentorship-program sense, with a calendar invite and a rubric — just literally next to them, solving real problems, having real conversations. And it's through that proximity that I actually learned something. Not syntax. I could have gotten syntax anywhere.
// What They Actually Taught Me
What I got from them was much harder to pick up from a book or a doc or, frankly, from a model. I learned how they thought about systems design — what they chose to care about deeply, and just as important, what they deliberately let go of. Knowing where not to spend your time is its own skill, and it's one you mostly learn by watching someone more experienced make that call in real time and explain why.
I learned how to manage my manager from them. I learned how to prioritize when everything on the board looks urgent. And I watched how they kept themselves sharp — they tinkered. They built side projects out of pure curiosity, with no business case attached, just because something was interesting to them. Some of those side projects quietly turned into real products for the company. Not because anyone asked for them. Because someone was curious enough to go find out.
I am the engineer I am today because of those years. Not because of a course I took or a certification I earned — because of the people I stood next to.
// The Vibing Bet
Which is why the current conversation happening around junior engineers worries me. There's a growing sentiment that they're not worth investing in right now — that the new generation doesn't need to learn to code the way we did, that they'll just vibe their way to competence over enough time and repetition.
I don't think that's entirely wrong. There's probably some merit to it — plenty of people are going to build real skill through sheer volume of reps with these tools, and I won't pretend otherwise. But I think the people making that argument are quietly writing off something that isn't replaceable by volume: the human connection, and the education that only happens through a peer who's sitting next to you, invested in whether you get better. An LLM will answer your question. It will not notice that you've been avoiding the same category of problem for six months and gently push you toward it. It will not tell you a war story about the time it made the exact mistake you're about to make. That's not a knowledge gap. It's a relationship gap, and you can't close it with a better prompt.
// A Regret With a Long Fuse
Here's the part that makes this genuinely dangerous rather than just a little sad: we are not going to see the effects of this immediately. Abdicating mentorship doesn't blow anything up next quarter. The juniors will still ship features. The agentic tools will make them look productive fast, faster than gray beards looked at their age, probably. It's a slow leak, not a rupture — which is exactly the kind of problem organizations are worst at responding to, because nothing on the dashboard tells you it's happening.
But somewhere in the next five to ten years, the cohort that never had anyone sit next to them is going to be the cohort we're asking to be the senior engineers, the architects, the people who mentor the next generation. And I think we're going to find out, all at once, what was actually being transmitted in those conversations we stopped having.
It's a hard time to be starting out in this industry. If you're one of the people just getting in, find a mentor. Ask for their time directly — most people say yes more often than you'd expect. You will get things from that relationship you will not get from an LLM, no matter how good it gets.
The tools are only going to get better at answering questions. They are not going to get better at being the person who noticed you needed one asked.