Back in February, Paul Ford lit up The New York Times with his rave about AI’s arrival in coding. Faster. Cheaper. Vibecoding on the subway, dusting off dead projects. That’s what we all expected: a turbo-boost for devs, productivity through the roof, everyone jazzed.
But here’s the twist — and it’s a gut-punch. Developers aren’t just coding faster; they’re forgetting how to code at all. Pia Torain, software engineer at Point Health AI, nailed it after months of prompt-spamming:
“I started to lose my ability to code.”
She forces herself now to slow down, trace the whole architecture — flow, guts, everything. Because if you don’t use it, you lose it. Simple as that.
This isn’t hype. It’s happening now, raw and real across Hacker News threads, NYT essays, dev chats. A 60-year-old programmer posts about Claude reigniting his passion — 1,000+ upvotes. Old Mac shareware vets vibe-coding websites they’d shelved decades ago. Solo founders getting guts to launch. Cool, right?
Wait — Is This Passion or Deskilling?
Look closer. That same thread? Nearly 1,000 comments, half cheers, half dread. John Calhoun, Apple alum from ‘95, admits AI let him tackle a site he wouldn’t have touched. Reini Urban in Dresden: unarchiving hard projects with Opus. John Reine in Boston: solo-founder courage.
But flip it. An anonymous Apple engineer dodges AI entirely. “Having the computer do it strips the fun, the fulfillment.” Joel Dare, 50-year Utah vet with 40 years in: zero tolerance for “architectural decay.” He’s fixing vibe-coded spaghetti — functions fine, app a mess.
One Hacker News gem: “LLMs good at coding, terrible at software engineering.” Spot on. AI spits clean snippets. Humans architect systems that don’t crumble under load, scale without hacks, secure without backdoors.
And juniors? Screwed hardest. Microsoft execs warn agentic AI hollows the pipeline. No one’s hiring newbies when LLMs crank junior work faster, cheaper. How do you level up without reps? YouTube tutors, Gemini coaches — human mentors obsolete in two years, some say.
It’s like the autopilot paradox in aviation. Pilots lean on computers for routine; muscles atrophy. Rare crisis hits — hands frozen on sticks. Programming’s wiring the same. AI handles boilerplate, loops, CRUD. You forget the why behind the how. Next refactor? Panic.
My unique angle here — and trust me, the originals skim this — echoes the calculator’s curse in math classes, ’70s. Kids punched buttons, lost mental arithmetic. Teachers fought back with no-calc drills. Coding needs that now: deliberate practice zones, AI-free sandboxes. Companies spinning productivity gains ignore this; it’s PR gloss on skill erosion.
Why Are Junior Developers Getting Crushed?
Straight talk. Entry-level gigs? Vanishing. LLMs debug, scaffold, even spec. What’s left for juniors — prompt engineering? That’s not coding; it’s herding cats.
Hacker News commenter: “If nobody’s hiring junior devs because LLMs do it faster and cheaper, how does anyone become an expert?” Boom. No reps, no path to senior. It’s a feedback loop to nowhere.
Older devs? Mixed bag. Some resurrected passion — fine. But burnout looms. Faster output means more bugs to chase, more spaghetti to untangle. Thompson’s interviews: most devs still get the success jolt, even with AI writing lines. Weirdly jazzed, he says.
Yet quality craters. Vibe-coding prioritizes speed over structure. Architectural shifts? AI’s flattening the pyramid — solos thrive, teams bloat with prompt jockeys lacking depth.
Can AI Tools Actually Improve Code Quality?
Short answer: Not yet. They’re pattern-matchers, not thinkers. Feed ‘em clean specs, get okay code. Vague vibes? Dog’s breakfast.
But here’s the why: LLMs train on GitHub slop — real-world mess. They mimic mediocrity at scale. True engineering? Holistic. Context across modules, tradeoffs, future-proofing. AI’s blind to that without god-tier prompting — rare in wild.
Prediction: Firms mandating AI audits, human sign-off on arch. Or open-source collectives building anti-deskilling tools — AI tutors that explain, not just generate.
Burnout’s the sleeper hit. Crank out 10x code, fix 10x bugs. Passion reignites? Sure, till exhaustion.
Thompson captures it: developers happy to ditch hand-coding, but a minority resists, guarding the craft.
The reckoning’s live. Tools evolve — Claude, Gemini, Opus — boosting solos, sidelining juniors, risking all. Expected boon? Turning bittersweet. Architecture’s shifting: from deep coders to AI wranglers. Smart readers see it: reclaim the fundamentals, or watch skills evaporate.
🧬 Related Insights
- Read more: Agent Sprawl: The Tech Debt That’s Already Burying Your AI Dreams
- Read more: Unlock ChatGPT-Style Streaming: SSE Magic in Next.js
Frequently Asked Questions
What are developers saying about losing coding skills to AI?
Many report atrophying abilities after heavy AI use, like Pia Torain’s warning: slow down or lose the architecture grasp.
Will AI coding tools replace junior developers?
They’re hollowing the hiring pipeline — faster/cheaper than newbies, stunting expertise growth.
How to avoid deskilling with AI tools?
Practice AI-free: trace full flows manually, build from scratch weekly, focus on software engineering over snippets.
