The data analyst from three years ago gets posted now as an applied AI specialist. Next year, the same req might read AI workflow designer.
Titles are turning over fast enough that HR teams can’t keep pace with the names, let alone pin down what the person would do all day.
Some of this is genuinely work that didn’t exist before.
The confusion costs on three fronts.
A hiring manager can’t write a clean req for a job nobody has named. A qualified candidate scrolls past the opening because its title doesn’t match anything they’d search for.
And the person who does get hired often finds the day-to-day looks nothing like the posting that drew them in.
AI at Work Now
AI has moved from the sandbox to the shop floor.
Radiologists lean on it to flag what’s on a scan, insurers price risk with it, a factory runs it to catch defects coming off the line.
Marketing teams push first drafts through it, and anyone who’s sat in a recorded meeting has watched the transcript summarize itself before the call ends. This stopped being a Silicon Valley story a while ago. It shows up wherever work gets done.
Travis Lambert, general manager of Central Oregon Heating, Cooling, Plumbing & Electrical, runs a trades business where scheduling, dispatch, and estimating increasingly route through software.
He says, “AI turned up in our back office long before it turned up in anyone’s job title here. The dispatch tool learned our patterns, the estimating software started proposing the numbers, and a couple of people were suddenly doing work that had nothing to do with what we hired them for.
Naming that work took longer than adopting the tools did, and we are a small shop. For a company hiring hundreds, the scramble has to be worse.”
The World Economic Forum’s Future of Jobs Report puts AI and big data among the top drivers of job growth through 2027, with AI and machine learning specialists ranking among the fastest-growing roles anywhere.
The same report’s headline numbers are 69 million jobs created and 83 million gone by 2027 as tasks move around and roles get redrawn.
LinkedIn’s Economic Graph shows the same thing from the demand side: postings mentioning GPT or generative AI are up more than twentyfold since late 2022, and AI fluency now turns up in ads well outside engineering.
The Roles Showing Up
Webmaster used to be a real job. So did the first social media managers and cloud architects, back when those titles sounded invented. Every wave of tooling throws off roles nobody saw coming a few years earlier. This wave is doing it faster.
An AI product manager owns the lifecycle of AI features and whatever internal platforms those features run on. MLOps engineers, the people doing machine learning operations, build the pipelines that move a model out of the prototype stage and keep it standing once production traffic hits.
Around them sits a band of governance work: AI ethicists writing the guardrails and the fairness policy, and model risk managers taking the governance playbook finance has run for decades and bending it to fit enterprise AI and the regulations now landing on top of it. Prompt engineers handle the workflows inside generative apps.
The list keeps running: think data governance stewards, synthetic data engineers, AI solutions architects, applied AI engineers, each one carving off a slightly different piece of the same ground.
Electricity didn’t stop at putting candle makers out of work. It stood up whole industries that hadn’t existed. The internet went the same way: past digitized mail into e-commerce, into SEO, into a working discipline called UX design that nobody had a name for before. AI follows that arc.
Some jobs disappear. Far more get reshaped, and the genuinely new ones tend to show up several at a time.
Why Naming Is Hard
The work itself doesn’t sit cleanly in one department.
One AI role might sit in data engineering for most of a week, then get pulled into a security review, then into a legal argument over what the model is even permitted to do, and somewhere in there it also owns the slow work of coaxing a wary team into using the thing at all.
The title stretches to cover every bit of that, picking up an AI or ML or GenAI prefix along the way, and the job ad behind it ends up asking for fifteen must-have skills no single person actually has.
Inside the company, that lands as confusion. On the job board it’s noise.
‘AI engineer’ is the clearest tell.
At one company it’s a data scientist. At the next it’s an ML engineer babysitting training runs. At a third it’s someone wiring an LLM into a product, and the three share almost nothing day to day.
Bolting ‘AI’ onto an existing role lifts the click-through on a posting while wrecking the ladder underneath it, as pay bands stop lining up and nobody can say what the next rung is.
Then there’s the split between what’s precise internally and what’s findable externally.
‘Applied LLM Engineer II’ might be exactly right on the org chart and dead on a job board, where ‘AI Engineer’ pulls far more of the right applicants. Candidates search in plain English. A title they can’t guess is a posting they’ll never find.
The target also keeps moving.
Last year’s prompt engineer is this year’s retrieval engineer, and as the stack matures, that same person gets folded into a broader AI apps role.
Writing a stable description for something still forming is its own headache. One HR leader spent three weeks just working out who was even qualified to draft the job description for an AI governance role, because the role hadn’t finished taking shape.
What helps is writing for accountability rather than ambition.
A good description starts from outcomes: the results the role answers for, the systems it has its hands on, and some real measure of whether it’s working. The title comes out of that, last, rather than getting set before the work is even clear.
Jason Ledbetter, operator at Jason Ledbetter, spends much of his time helping companies hire for work that had no title a year earlier.
He says, “The fastest way to blow a hire is to name the role before anyone is sure what it does. I have watched a company post a title that sounded impressive, pull in forty applicants, and find that none of them fit, because the actual work was never pinned down first. Write what the person will be accountable for, and the title falls out of that on its own, and it falls out right.”
Who Defines the Terms
No one company owns the vocabulary. The language gets set by a handful of forces pulling on it at once.
The largest employers and the hiring platforms do most of the shaping. When a few big companies settle on the same title, the rest drift toward it, and LinkedIn’s taxonomy plus the plain behavior of job seekers nudges the whole market toward shared patterns.
Standards bodies pull in the same direction. A lot of compliance and product teams already treat the NIST AI Risk Management Framework as their reference point source, and the Partnership on AI and ISO/IEC SC 42 publish practices and standards that work their way into how job scopes get drawn PAI and ISO/IEC SC 42.
Regulation is where the language hardens.
The EU AI Act will spell out duties for risk management, documentation, and human oversight, and the roles attached to those duties: AI compliance, model risk, safety, multiplied because of it source.
The companies out front aren’t waiting for any of this to settle. They set up cross-functional AI councils and run their role definitions back against a framework like NIST. The internal role guides they write connect each title to a real rung on a career ladder. Getting there early spares them the mess everyone else inherits: a drawer of one-off titles, none of which has a clear owner.
The same thing is beginning to happen a level up, across institutions, as employers compare notes with universities and industry groups and inch toward shared definitions rather than each coining its own.
What Reskilling Requires
Changed roles demand changed skills, and the awkward part is how many of these jobs want depth and breadth at the same time.
The person who can evaluate a model also needs product sense. The one guarding data privacy has to carry a feel for how the thing behaves for whoever’s using it.
Running an experiment is one piece of the work; keeping it alive in production once real traffic arrives is just as much of it.
For a team that has never worked shoulder to shoulder with ML, that mix is hard to hire for and harder to grow in place.
What actually moves people into these jobs looks a lot more like the job than like a course catalog:
- Employee training works when it’s built on the exact stack and policies a team already runs, because a generic syllabus doesn’t survive contact with the real system.
- Apprenticeships and internal transfers help too, carrying someone over from an adjacent role so they aren’t starting from zero.
- Micro-credentials earn their keep only when they mean something in-house: model evaluation, RAG patterns, privacy-preserving ML, the things a team actually checks for, rather than a certificate nobody looks at.
The format that tends to stick is a short project, run by a real team and judged on whether it moved a metric that mattered. A finished course only proves someone attended.
Where Titles Go Next
Consolidation is coming, and some of the moves inside it are sharp.
As the stack matures, the sprawl of titles shrinks: prompt engineer narrows into AI applications engineer, while MLOps keeps folding back into ordinary platform engineering. Governance runs the other way and widens.
What used to be specialist corners, AI safety and model risk and compliance, turns into standard headcount, hitting regulated industries first and spreading out from there.
Evaluation looks like it’s turning into a craft of its own.
The job titles forming around it tell the story: model evaluation architect, red-teaming lead, an AI QA specialist whose whole remit is testing that holds up and repeats instead of getting improvised each time.
Data work drifts toward the center in step with that. Synthetic data, data contracts, roles built lineage-first- all of it stops being optional the moment AI is carrying a real production load.
Markets won’t move in sync.
Where regulation came early, or a strong industry group set the tone, standard titles lock in sooner; elsewhere the looser mix hangs on for years. The underlying work barely shifts through any of it. A business problem gets identified and mapped onto data and the right models, tested without breaking anything that’s live, then measured for what actually moved, and then the loop runs again.
The titles with staying power are the ones that stay bolted to that loop.
Naming a Role Well
A few patterns hold across almost any search. The title leads with the function the person will actually do, which is why “AI Product Manager” works and “Product Innovator, AI” says nothing. The prefix has to match the real stack, because AI, ML, and GenAI point at different work, and the wrong one pulls the wrong crowd.
Seniority stays plain: I, II, Senior, Staff, Principal, so candidates can size themselves against the role.
Public and internal titles don’t have to match. “Applied LLM Engineer II” can sit on the org chart while the posting reads “AI Engineer” and wins the search. The description earns its keep by naming outcomes rather than adjectives: the systems the role owns, the metrics it answers for, and the calls the person actually gets to make.
Getting every title and description on a careers page to read that clearly is the job Ongig was built to do.
