AI careers: twelve roles, and which ones are real jobs
Twelve job titles now circulate around getting AI into organisations, and they are not twelve jobs. Some are established roles hired consistently across many companies. Some are real work whose scope differs so much between employers that the title alone tells you almost nothing. And some are contested: the same words attached to several different jobs, or being quietly absorbed into the roles beside them. Sorting them by what each is actually judged on is more useful than defining them, because the judgement criterion is what your performance review will be about and it is stable even when the heading is not. Four criteria cover almost all of it: model quality against an evaluation set, a system running inside an organisation, an operating model that changed, or an estate somebody can account for. Those are four careers.
Established roles
Hired as a distinct position across many companies, with a recognisable scope.
AI engineer
Established roleBuilds the system around the model: retrieval, agents, evaluation, guardrails, latency and cost.
- Judged on:
- Whether the system is accurate, fast and cheap enough, measured against an evaluation set that can be run on demand.
- Fails when:
- The evaluation set stops resembling what users actually send, and the numbers keep improving while the product gets worse.
Machine learning engineer
Established roleTrains, tunes and serves models, and owns the pipeline that keeps them fed and fresh.
- Judged on:
- Model quality against a held-out set, and whether the training and serving pipeline holds under real volume.
- Fails when:
- The training distribution drifts away from production and nobody is watching the gap.
AI solutions architect
Established roleDesigns the target system: how the pieces fit, what the constraints are, what should be built and in what order.
- Judged on:
- Whether the design survives the security review and the customer reality it was drawn against.
- Fails when:
- The design is correct and unaffordable, because its cost was never priced by anyone who had to build it.
Emerging roles
Real work, but the scope differs enough between employers that the title alone tells you little.
AI implementation consultant
Emerging roleTakes a decision that has been made and gets it running inside the organisation, tools, process and people.
- Judged on:
- Whether the thing is live and in use at the end of the engagement.
- Fails when:
- The engagement ends at go-live and the organisation was never made able to run it alone.
AI product manager
Emerging roleDecides what an AI product should do, with the unusual complication that its behaviour is probabilistic.
- Judged on:
- Whether the product is used and whether its failures are the ones the team chose to accept.
- Fails when:
- It is run like ordinary product management, with a roadmap of features and no position on acceptable error.
AI governance manager
Emerging roleOwns the rules: what may be built, on what data, with what approval, and how it is evidenced afterwards.
- Judged on:
- Whether the organisation can answer, on demand, what its systems do and on what basis.
- Fails when:
- Governance becomes a gate rather than a service, and teams route around it.
AI security engineer
Emerging roleSecures systems whose inputs are untrusted text and whose behaviour cannot be fully enumerated.
- Judged on:
- Whether an attacker can make the system act outside its intended permissions.
- Fails when:
- The threat model is written for software and the system is an agent with a write credential.
AgentOps engineer
Emerging roleKeeps agents running in production: tracing, cost, failure modes, and knowing when one has quietly got worse.
- Judged on:
- Whether a degradation is caught before a user reports it.
- Fails when:
- Observability is built for services and the thing being observed makes non-deterministic decisions.
AI operations manager
Emerging roleRuns the portfolio: which systems exist, what they cost, who owns them and which should be switched off.
- Judged on:
- Whether the organisation knows what it is running and what it is paying for.
- Fails when:
- It becomes an inventory exercise with no authority to retire anything.
Contested titles
The title is used for several different jobs, or is being absorbed into neighbouring ones.
AI consultant
Contested titleAdvises an organisation on where AI is worth applying and in what order, usually without building it.
- Judged on:
- Whether the advice survived contact with the organisation, which is rarely measured and should be.
- Fails when:
- The recommendation is correct and nobody can execute it, because the constraint was never technical.
AI transformation consultant
Contested titleWorks on the operating model rather than on any one system: who does what, measured how, once AI is in the process.
- Judged on:
- Whether the organisation works differently a year later, which almost nobody measures.
- Fails when:
- It produces a target operating model that no existing team has the authority to reach.
Prompt engineer
Contested titleWas a standalone job for about eighteen months and has largely been absorbed into the roles around it.
- Judged on:
- Output quality on a defined task, when the role exists at all as a separate one.
- Fails when:
- It is hired as a permanent position, because the skill is now a component of several other jobs.
Why the titles are this unstable
Two forces are pulling in opposite directions. Demand is real and growing: the Stanford AI Index records AI skills explicitly requested in 2.5 % of all United States job postings, up 297 % over a decade. When demand moves that fast, companies invent titles faster than any shared meaning can settle.
At the same time the technology keeps absorbing work. Prompt engineering was a standalone position for roughly eighteen months and is now a component of several other jobs rather than one of its own. That is not a failure of the people who held it; it is what happens when the thing a role exists to do gets built into the tools.
The practical consequence for anyone planning a career is that betting on a title is riskier than betting on a judgement criterion. Someone who is good at making systems work inside organisations will be employable under whatever the heading turns out to be in three years.
How to read a posting in this space
Three questions do most of the work, and they are the same three whatever the heading says.
What does this person produce? A document, a design, a running system, or a changed process. Each implies a different job and a different set of skills, and postings are usually honest about this even when the title is not.
Who is accountable when it does not work? Roles that hand over before the consequences arrive are legitimate and are a different career from roles that are still there at the incident review.
How much of the week is writing code? This separates the engineering roles from the advisory ones more reliably than any title, and the answer changes what you will be able to do next.
What is missing from this page
Compensation, and it is the omission people notice. The reasoning is set out in full on the methodology page: the public figures for these titles disagree by roughly a factor of four because they average postings that share a heading and nothing else, and a single number per role would rank well and mislead everyone who used it.
Also missing: any claim about which of these roles will exist in five years. Several of them will not, and we do not know which. What we can say is that the underlying work, which is getting systems to change how organisations operate, is not going anywhere, and that is what the forward deployed engineer pages describe.
