We put a room full of senior engineering leaders together, took away the slides, added beer, and asked them to say the thing they’d never put in a LinkedIn post. No PR-approved talking points, no vendor gloss, just people who ship for a living telling the truth about what AI is doing inside their orgs right now.
It got spicy. Here’s what actually came out of the room.
1. “AI is coming for junior engineers” is a sales pitch, not a forecast
The loudest claim in the market got some pretty quick pushback. The consensus: the job is changing like the way surgery changed with new tools, not disappearing. One leader caught the irony perfectly. The industry that spent a decade telling everyone else to “learn to code” is now the one squirming as automation turns to face it.
The real debate wasn’t if juniors survive, it was how the role shifts. Some think mid-sized companies stop hiring juniors because cheap code kills the old “hire cheap, grow them up” math. Others said it’s about domain, not headcount: a consumer app can tolerate a junior leaning on AI, a rocket cannot. And the plot twist nobody expected: the newest grads trained on these tools from day one might out-produce today’s mid-levels, because they never learned to work without them.
2. Should an AI agent review another AI agent’s code?
Nobody had a clean answer. One leader is already having agents pre-validate PRs against written standards, so by the time a human looks, most of the critical stuff is handled. But fully autonomous agent-reviews-agent for anything with real stakes? Not here yet, and maybe shouldn’t be.
3. Onboarding got 10x faster, and that cuts both ways
The good: a team mid-SOC 2 audit needed to investigate a possible vulnerability in a corner of the codebase nobody understood. Instead of hunting for the one nine-year veteran, engineers with almost zero context ramped up and shipped a fix in a week using AI.
The dark side: an engineer auto-generated a big batch of internal docs from the codebase, and it hallucinated in subtle, hard-to-catch ways. The only reason anyone noticed? People who’d actually worked in that system knew it was wrong. AI collapses onboarding time, but it doesn’t replace the person in the room who can smell when something’s off.
4. About those layoffs
This one had the most heat. Several big companies blamed AI for cuts after, as one leader put it bluntly, they “over-hired like crazy during COVID and never right-sized.” The AI story was often a convenient cover for a boring one: bad headcount planning finally catching up.
But the room didn’t go full cynic either. In copywriting, sales support, and customer support, there’s real evidence AI is trimming headcount while holding output. In engineering, it’s fuzzier: output per person may be rising, but nobody had seen a credible study showing a 10x cut in required engineers. The honest read is incremental gains today that could compound hard over the next few years. Think radiology: automation was supposed to end the profession, instead demand went up and the number of radiologists grew. The task changed. The job didn’t vanish.
5. Please! Stop building a dashboard for individual AI productivity
The most passionately argued point of the night. PRs merged, cycle time, tokens consumed (yes, someone raised Meta’s infamous token leaderboard) were all seen as a trap. Tie a number to performance and people optimize the number, not the outcome.
The sharper cuts:
• Team velocity is measurable and useful. Individual “AI productivity” is a much slipperier animal.
• Not all PRs are equal. One can be worth ten. The senior who spends the day unblocking and reviewing others looks “unproductive” on a naive dashboard while being the most valuable person on the team.
• Review time might be the metric that matters. As the bottleneck shifts from writing code to reviewing it, the best reviewers, the ones who spot what’s structurally off, become the highest-leverage people you have. No existing tool captures that well, and it deserves way more attention than it’s getting.
The through-line
One sentence captured the whole hour: the code was never the job. AI made code production cheap, but judgment, architecture, knowing where the risk lives, mentoring, reviewing, and talking to the business haven’t gotten any less important. They’ve gotten more important, because they’re the part AI still can’t reliably do.
Want to keep the conversation going?
This was round one. If any of these takes hit a nerve, made you nod, or made you want to argue, let’s talk. Book a 1:1 with our team and tell us what AI is actually doing inside your org, the good, the messy, and the “we don’t talk about that in standup” parts. Bring your hottest take. We’ll bring ours.