AI Is Creating Jobs. They're Just Disguised as Old Ones.

Most of the new jobs created by AI don't arrive with a new title. It shows up as the augmented version of a job that already exists. The paralegal posting from 2025 looks identical to the one from 2019. The expectations buried inside it do not. If you're waiting for an "AI jobs" tab to appear on LinkedIn, you're waiting for a wave that already broke under your job's old name.

That's the part the career-advice industry keeps getting wrong. It keeps pointing people at a future that will announce itself with fresh postings and fresh titles. History says it won't. Technology destroys tasks and rebuilds jobs that look like reconfigured versions of the old ones. The title stays. The job description mutates. The person who rewrites their own job description wins.

Why doesn't new technology get its own job title?

Because the work gets absorbed into roles that already have names. When spreadsheets arrived, accounting postings didn't say "Excel accountant." They kept saying "accountant" and quietly raised the bar. VisiCalc and Lotus 1-2-3 didn't kill the profession. They thinned out the junior clerks doing manual ledger work and rewarded the accountants who could do strategic analysis. The numbers bear it out: there are roughly 400,000 fewer bookkeeping and accounting clerks than in 1980, and about 600,000 more jobs for higher-level accountants. The role split in two. The posting never changed.

The same thing is happening now, on a wider footing. About 60% of US jobs in 2018 were in occupations that didn't exist in 1940. New work is the engine of employment. But that work mostly grew inside the bones of older categories rather than as standalone, freshly-labelled jobs you could go apply for the week they appeared.

What does the bank-teller story actually teach?

Not that automation never costs jobs. The opposite, eventually. The teller case is the one everyone cites and half of them misread. Economist James Bessen documented the core finding, and the numbers are striking: from the 1980s to 2010, while roughly 400,000 ATMs were installed across the country, bank teller employment grew from around 500,000 to nearly 600,000. ATMs made each branch cheaper to run, so banks opened more branches, and each branch still needed people for the things a machine couldn't do: exceptions, complaints, selling, judgment.

Here's the part the optimists skip. Teller employment eventually fell, hard, once mobile banking did what the ATM couldn't. The machine didn't end the job in year one. The phone did, twenty years later. The lesson isn't "automation creates jobs forever." It's that the timeline is long and indirect, and the people who survived the in-between were the ones who moved toward the tasks the machine couldn't touch, fast, while the title still said "teller."

What does the data show is happening to jobs right now?

The shift is inside the role, not on the job board. Indeed's 2025 AI at Work report found that 46% of the skills in a typical US job posting fall into a hybrid or full AI-transformation category, meaning how the skill gets done is changing even though the skill's name stays put. LinkedIn puts a number on the broader drift: about 70% of the skills used in most jobs will change by 2030.

And yet the postings barely mention any of it. Generative-AI terms appear in only about 3 of every 1,000 job postings, even while AI literacy ranks as the single fastest-growing skill in the US. The DNA of the job is being rewritten in a font you can't see on the listing. That's the whole point. The change is real, and it's invisible if you're only reading titles. The same dynamic shows up when you trace the skills that quietly entered every job description: the requirements moved, the wording lagged. Which is also why the job description is a fictional document and a poor guide to what the work actually demands.

So why is waiting for "AI jobs" the wrong move?

Because the title that's supposed to save you might disappear before you can get hired into it. Prompt engineering was the hottest new job of 2023, a six-figure novelty. By 2025 it had collapsed as a standalone title and dissolved into the work of software developers, data scientists, and product managers. As Indeed economist Allison Shrivastava put it, "Prompt engineering as a skill is still definitely a good thing to have, but it's not an entire title." The skill demand went up. The job title went away.

Meanwhile the boring old titles are absorbing the value and growing. Software developer employment is projected to grow 17.9% from 2023 to 2033, well above the roughly 4% average for all occupations, even as AI shifts more of the developer's day toward design and review and away from typing out routine code. Radiologist demand has held up despite AI reading images: at the Mayo Clinic, radiology staff grew from about 260 in 2016 to over 400 today, a 55 percent expansion, with AI making human practitioners more efficient rather than redundant. The pattern is consistent. AI doesn't delete the role. It reweights it toward oversight, integration, and judgment. The economist David Autor frames this as AI letting "a larger set of workers possessing complementary knowledge" do the higher-stakes work once reserved for elite experts. Redistribution within a role, not a brand-new role.

What does the strong move look like versus the weak one?

The difference is who writes your job description first.

Weak: A paralegal waits for an "AI Legal Researcher" posting to appear before touching any AI tooling, on the theory that you learn the new thing once it's an official job.

Strong: The same paralegal starts using AI contract-review tools now, cuts document review from four hours to forty-five minutes, brings the time savings to the partners, and becomes the person who owns AI-augmented discovery at the firm. Their LinkedIn still says "Paralegal." Their leverage is in the gap between the title and what the job now requires.

Weak: A developer refuses to use AI coding assistants so they don't "train their replacement," and falls behind on output.

Strong: A developer uses AI to double their code output and redirects their own hours into architecture, code review, and edge-case judgment, the work the machine still can't do reliably.

Same title. Different person. One of them rewrote the description. Here's the same fork across three common roles:

Role	The weak move (wait for a title)	The strong move (claim the augmented role)
Paralegal	Apply only to "AI Legal Researcher" postings	Own AI contract review, show the time saved, redefine the role
Software developer	Avoid AI tools to "not train your replacement"	Double output, shift effort to architecture and judgment
Finance analyst	Wait for an "AI Finance Analyst" title to exist	Build AI scenario modeling into your FP&A work, become the rapid-analysis person

What's the honest trade-off here?

This is the part most takes skip, so here it is plainly. "Redefine your role" is not a guarantee, and it's not equally available to everyone.

First, the reinstatement effect may not hold this time. Autor's own research found that the job-eroding effect of automation was more than twice as large from 1980 to 2018 as it was from 1940 to 1980, while new task creation failed to keep pace. If overall demand for a role collapses, redefining your tasks inside it won't save the job. Task redefinition buys you position, not immunity.

Second, there's a speed asymmetry. A firm can restructure a team around two AI-augmented paralegals instead of eight before any single paralegal has had time to remake their own role. Employers redefine faster than individuals adapt. "Do it before someone else does" is right, but the clock is not on your side.

Third, baking AI fluency into every role raises the entry bar. That's harder, not easier, for people without the time, tools, or runway to retrain. "Claim the augmented version of your job" assumes you have slack to claim it with. Not everyone does, and pretending otherwise is the kind of BS this advice is supposed to avoid.

And the prompt-engineering story cuts both ways. The skill survived by getting absorbed. The $200K title did not. Anyone who bet their career on a shiny AI-native label as a hedge got burned. Betting on the durable old title, augmented, was the safer play.

What should you actually do this week?

Stop scanning for new titles. Start auditing the one you already hold. Three moves, in order:

1. Split your job into two columns. On one side, the tasks AI is absorbing in your field (the commodity reads, the first-draft code, the manual reconciliation). On the other, the tasks it amplifies (the judgment calls, the oversight, the integration, the relationship work). The amplified column is your leverage. Spend more of your week there starting now, and use it to filter the roles you actually chase.

2. Build a visible track record under your current title. Use the tools, measure the time or quality you save, and tell someone who matters. The paralegal who can say "I took discovery from four hours to forty-five minutes" has rewritten their description in a way HR can't ignore.

3. Update your resume and LinkedIn to the task mix you own, not the title you hold. Recruiters are screening for the augmented version of your role whether or not the posting says so, and a generic "proficient in AI tools" line won't get you there because those AI-skill claims are mostly theater now. Show them the specific work you do, so they can see you're already it.

Consistency beats the heroic sprint here. You don't need to reinvent yourself in a weekend. You need to move a little further toward the amplified tasks every week, while the title still says what it always said, until the day someone asks who owns the AI-augmented version of your job and the answer is obviously you.

Want to map which tasks in your exact role AI is absorbing versus amplifying, and rewrite your resume to claim the augmented version before someone else does? That's the kind of thing I work through with people on WhatsApp. Bring me your current title and let's find your leverage.