AI and Employment: A 200-Year-Old Debate

Tech job boards in 2025 are schizophrenic. Traditional software engineering roles are shrinking. "AI"-prefixed positions are expanding. Same company, same quarter — cutting junior devs and project managers on one side, opening Agent orchestration engineers and AI application architects on the other.

AI is creating and destroying jobs at the same time. Not news. The real question: which side moves faster, and how long does the gap last between "old job gone" and "new job accessible"? Economists have been fighting about this for two centuries. Their conclusion, distilled: technology creates jobs in the long run, but the transition hurts. Ricardo showed in 1821 that a bigger pie doesn't mean bigger slices for everyone. Keynes invented the term "technological unemployment" in 1930 — specifically for this kind of pain.

Two hundred years later, the theory holds. But AI has changed what the transition looks like.

What AI Is Already Replacing

Not speculation. Already happening.

Translation took the first visible hit. Duolingo cut about 10% of its contractors in January 2024, saying "generative AI is accelerating our content creation." One year later, in April 2025, they went all the way — announcing a full phase-out of human contractors in favor of "AI-first." They're not alone. Routine business documents, product manuals, press releases — the bread-and-butter translation work is getting swallowed whole by AI.

Customer service moved even faster. Klarna dropped a striking number in February 2024: their AI assistant handled two-thirds of all support chats in its first month. That's the equivalent of 700 full-time agents. Satisfaction scores matched human agents. Resolution time fell from 11 minutes to under 2. When banks and telecom companies see those numbers, they don't wait. Front-line support headcount keeps shrinking. The humans who remain handle escalations and messy disputes — the stuff AI still botches.

Programming hasn't seen mass layoffs, but the hiring mix is quietly shifting. Senior engineer demand hasn't dropped. What's dropping is junior and mid-level hiring. A junior dev who's good with Cursor can produce what used to take a mid-level engineer. Companies do the math: five juniors' worth of work now takes three. Stack Overflow's 2024 survey backs this up — 76% of developers are using or planning to use AI tools, 62% already use them daily. Everyone gets faster, but entry-level openings get scarcer.

Design is adjusting fast too. Before Midjourney, a concept design exploration from sketches to final direction could take a week. Now a designer can run through dozens of directions in a single day. But who benefits from that speed? Senior designers who can steer the AI, not junior ones who only knew how to execute. The role has shifted from "person who draws" to "aesthetic decision-maker." Execution-layer work is drying up.

One thread runs through all of these: what's getting displaced isn't bottom-tier physical labor — it's cognitive work in the middle of the skill ladder. The Industrial Revolution ate from the bottom up. AI cuts in from the middle. And it's climbing. Technical architecture decisions that needed a senior human last year? Claude handles them respectably now.

What AI Is Creating

New roles are real. They just can't be filled by the people losing their old ones.

Agent orchestration engineer — this job title didn't exist before 2024. Now every major tech company has openings. The work: designing multi-Agent systems that automate complex end-to-end tasks. AI safety researcher is another one; OpenAI, Anthropic, and Google are all scaling their safety teams aggressively.

Then there's the less glamorous kind. AI trainers — human feedback annotation, model tuning. Lots of entry-level openings, but the pay is modest and the work is repetitive. Basically data labeling with a new name. AI product managers are genuinely scarce, though. You need someone who understands both product logic and model capability boundaries, and that person barely exists yet.

Upstream infrastructure is booming too. Data centers, chip design, energy supply. Nvidia's market cap speaks for itself. Supply chain jobs are genuinely growing.

Two problems, though, and they're hard to fix. Skill mismatch: a ten-year translation veteran can't become an Agent engineer next month. Geographic concentration: Silicon Valley is on an AI hiring spree while traditional white-collar workers in smaller cities just see layoffs.

Acemoglu's task-based framework nails this. He analyzes employment as "tasks" rather than "jobs" and finds that since the 1980s, technology's ability to create new tasks that boost labor demand has been weakening. New tasks still appear, but the compensating force is weaker each cycle. AI makes the imbalance worse.

Three Speeds Racing

The whole thing boils down to three speeds.

Displacement is the fastest. AI reaches "good enough" at an exponential pace — a new tier every six months. AI translation beat average human quality in most scenarios by 2024. AI code generation could independently handle medium-complexity features by 2025. Previous tech disruptions were linear. This one isn't.

Creation comes second. The steam engine took roughly 40 years to spawn the railway industry at hiring scale. AI moves faster — Agent orchestration went from concept to formal job title in two years. But new industries still need education pipelines, infrastructure, and market demand to mature. That part can't be rushed.

Adaptation is the slowest. Going from "I can translate" to "I can build AI products" isn't a bootcamp problem. It means rebuilding your entire cognitive framework. Universities are even slower — AI programs opening this year won't produce graduates for four years.

Displacement far outpaces the other two. That gap is an employment fault line. Not permanent mass unemployment, but a painful transition — probably 5 to 15 years.

Who Gets Hit Hardest

The old rule was that high-end and low-end jobs stay safe while the middle gets crushed. AI rewrites that rule.

Paralegals, junior financial analysts, medical imaging readers — these are well-paid positions, but their core work is information processing. AI is good at that. Plumbers, electricians, and care workers are barely touched, because their work demands physical-world dexterity. The dividing line has shifted from income level to a simpler question: information work or physical work?

Worse than losing your job is what you might call "downgraded employment." You still have a job, but your bargaining power is gone. AI handles 70% of your workload. Your value proposition shifts from expertise to "willing to stay late" and "can take the client drinking." Already happening in translation, design, junior legal.

Software development is in somewhat better shape — not because displacement is slow, but because adaptation is fast. AI-assisted coding went from skepticism to standard tooling in two years. The "cut juniors, hire AI engineers" pattern from the opening paragraph? That's the reinstatement effect kicking in. There will be pain, but the cycle might be 3 to 5 years.

What Can You Do About It

Macro trends aren't yours to change. But compensation mechanisms are real — they just lag. People who survive the lag benefit from the technology. People who don't become statistics.

Querying ChatGPT for information is "using AI." Wiring Agents into your workflow to automate end-to-end tasks is "collaborating with AI." Very different scarcity profiles. The second one won't be commoditized for at least five years.

Where does AI's displacement curve arrive slowest? Complex interpersonal dynamics. Cross-domain judgment calls. Decisions where someone has to be accountable. Anything requiring hands in the physical world. Move in those directions.

A decade ago, "knowing SEO" was a career. Five years ago, "knowing data analysis" was a career. AI does both passably now. The iron rice bowl as a concept is disappearing. The most durable skill is the ability to pick up new ones fast.

Keynes said humanity will be fine in the long run. He also said something more honest elsewhere: in the long run, we are all dead.