March 30, 2026
Mind the gap: the career ladder broke
The ladder is missing rungs – Engineering Progression When AI Ate the Middle
Devs clash over AI: 'We’re faster now' vs 'Who trains juniors?' as the ladder disappears
TLDR: A talk warns AI is doing the mid-level work that trains engineers, and the old path up is breaking. Commenters split between 'we’re faster now, your data’s old' and 'we don’t need juniors anymore,' with some predicting a tiny elite of engineers and others betting demand will keep training alive.
Engineers are sparring over a talk that says AI has eaten the “middle” of software work—the grind that used to turn rookies into pros. Remember the claim that AI would write “90% of code”? The receipts are messy: Redwood Research pegs real code closer to ~50%, with big companies like Google and Microsoft reporting ~25–30%. Meanwhile, a METR study found seasoned devs got 19% slower with AI tools—and then couldn’t even run a follow-up because developers refused to work without AI. Dependency or power-up? The room is divided.
The comments are on fire. One camp, led by wolttam, calls the study old news—“models got way better late 2025”—and swears AI now makes seniors fly. Another camp, like ngburke, says the ladder’s missing rungs: “we just don’t need junior engineers,” floating apprenticeship-style training as the new on-ramp. The doomer take? DGAP: “very very very few engineers.” The optimists, like datadrivenangel, clap back with econ vibes: cheaper code means more software, which means more people will need training. Sci‑fi flair appears too—bearfox cites a story about societies stuck on a plateau. Memes abound: “Mind the gap,” “bring your own ladder,” and “onboarding = prompt‑prenticeship.” The only consensus: the job changed—fast—and nobody agrees who gets left behind.
Key Points
- •A 2026 QCon London talk argues AI now performs many early-career engineering tasks, disrupting traditional pathways that build judgment and supervision skills.
- •Dario Amodei predicted AI would write 90% of code in 3–6 months and nearly all within a year; a year later, broader data shows lower shares.
- •Redwood Research estimated AI accounts for ~50% of committed code, while Google (~25%), Microsoft (~30%), and GitHub Copilot (~30% acceptance) report similar ranges.
- •Anthropic reports engineers use Claude in 59% of daily work with 50% productivity gains; Claude Code authors ~4% of public GitHub commits.
- •A METR randomized controlled trial found experienced developers were 19% slower using AI, and a 2026 follow-up study failed because developers refused to work without AI for controls.