Product Artificial Intelligence Workflow Specialist

This role sits between a product team and the AI tools that power or support a product. You work out where an AI model (usually a large language model - the technology behind chatbots like ChatGPT) can be slotted into a product feature or an internal process, then design, build and test the step-by-step "workflow" that gets it to produce useful, reliable output. The work is done for software companies, digital teams inside larger businesses, agencies and consultancies, and usually involves working alongside product managers, engineers, designers and the people who actually use the tool day to day.

Approximate graduate salary

Typically around GBP 30,000-42,000 to start, though this varies widely - London and well-funded technology companies sit at the top of that range or above it, while smaller firms, agencies and regional employers often start lower. Because the job title isn't standardised, two adverts with the same words can pay very differently depending on how technical the work actually is.

What you'd actually do

Writing and rewriting prompts - the instructions given to an AI model - and testing how small wording changes alter the output, then saving the versions that work in a shared library or repository
Building multi-step workflows in automation or orchestration tools (things like no-code automation platforms, workflow builders, or Python scripts) that chain together a data source, one or more AI calls, and an action such as drafting a reply or updating a record
Running evaluations: putting together a set of test cases, scoring the AI's answers against them, and tracking whether a change actually made things better or just different
Sitting with the teams who will use the workflow - support agents, sales, operations, or external customers - watching where it breaks or gets ignored, and adjusting it
Writing up documentation and guidance so other people can use or extend the workflow without you, including notes on what the model is bad at
Joining product planning sessions and sprint ceremonies to argue for or against building something with AI, including saying when a simple rule or form would do the job better
Checking outputs for things like hallucination (the model stating something confidently that is false), leaked personal data, or tone that doesn't fit the brand, and flagging anything that needs a human review step

How graduates get in

Moving sideways from a graduate product, operations or analyst role once AI work appears in the team - this is probably the most common route, because many of these jobs are created inside existing teams rather than advertised as graduate posts
Direct application to a junior or associate-level advertised role, which does happen but is still relatively unusual at true graduate level; most adverts with this kind of title expect a year or two of prior work
Graduate schemes in technology, product or consulting that include AI or automation rotations - the scheme title usually won't mention AI, so you have to read the rotation detail
Internships or placement years in product management, data, or technical support at a software company, then converting to a full-time role
Conversion from a technical background: computer science, data science or software engineering graduates who find they prefer the product and workflow side to writing production code
Building a visible portfolio - working AI tools, automations or evaluation write-ups published on GitHub or a personal site - and applying directly to smaller companies and startups, where this carries real weight

What employers ask for

Degree subject is genuinely flexible. Computer science, data science, engineering, maths, physics and linguistics are all common, but so are humanities and social science degrees where the person has clearly taught themselves the tooling. Some employers want a technical degree; many don't.
A 2:1 is the usual filter at larger employers, though smaller companies and startups often care far more about what you can demonstrate than your classification.
Some hands-on familiarity with AI tools is close to non-negotiable - not academic knowledge of machine learning theory, but evidence you have actually built something with an LLM API or automation platform.
Basic scripting ability, usually Python, is asked for in most versions of this role. A minority of more product-leaning roles accept no-code tooling instead.
No formal professional qualification exists for this. Short courses and vendor certificates in prompt engineering, cloud AI services or product management can help a CV but none are required or expected.
Experience requirements vary a lot: some adverts are genuinely open to graduates, others use 'specialist' to mean two to four years in product, data or operations. Read the advert body rather than the title.

Skills that matter

Prompt design and iteration

The core craft of the job is getting consistent, usable output from a model that behaves differently each time you ask it something.

Evaluation and testing discipline

Without a structured way to measure whether a change improved things, AI work becomes guesswork, and this is the skill that separates serious practitioners from people who just fiddle with prompts.

Process mapping

Before you can automate a workflow you have to describe the existing human process accurately, including the exceptions nobody documented.

Working Python or scripting ability

You will need to call APIs, clean up input data, and glue steps together when no-code tooling runs out of road.

Scepticism about AI's limits

A large part of the role is spotting where a model will fail quietly and designing a human check or a non-AI fallback around it.

Explaining technical behaviour to non-technical colleagues

You constantly have to tell product managers, legal teams or frontline staff why the system did something odd and what you're doing about it, without jargon.

Where it leads

Start out owning one or two defined workflows under supervision, mostly prompt writing, testing and documentation.
Move to owning a product area or a set of related AI features end to end - deciding what gets built, running the evaluation process, and briefing engineers.
From here the path usually forks. One branch is product: AI product manager, then senior product manager, where you own commercial outcomes and roadmaps rather than the workflows themselves.
The other branch is technical: AI engineer, applied AI or machine learning engineer, which means deeper coding and systems work and often requires deliberately building that skill while in the role.
Longer term, routes include heading an AI or automation function inside a company, consulting independently, or joining an early-stage startup in a founding product role. Timelines vary a great deal - this is a new enough area that people move faster than in established professions, but there is no conventional ladder, and job titles in this space are not yet standardised.

What people get wrong

“It's a prompt engineering job - you sit around writing clever prompts all day.”

Prompt writing is often a minority of the work. Far more time goes on understanding the business process you're trying to improve, building the plumbing around the model, and testing. Models have also got better at handling imprecise instructions, so the value has shifted towards evaluation and system design.

“You need a machine learning degree or to be able to train models.”

Most of these roles use existing models through an API and never train anything. Understanding how models fail in practice matters more than understanding the maths of how they're built - though a technical background certainly doesn't hurt.

“The job is about adding AI to as much of the product as possible.”

A good specialist spends a lot of time arguing against AI where a form, a rule or a search box would be cheaper, faster and more reliable. Knowing when not to use it is part of the role's value.

“It's a purely technical role tucked away from users.”

It's heavily collaborative and often user-facing. You'll spend a lot of time watching people use the thing you built, which is where you find out it doesn't work the way you assumed.

Where this varies

This job title is new and used inconsistently, so check what the advert actually describes. At a software company it usually means building AI features that customers use, and leans technical. At a large non-tech employer - a bank, retailer, insurer or public body - it more often means automating internal processes, with more emphasis on process mapping, governance and change management than coding. In consultancies and agencies you'll work across several clients at once and spend more time presenting and scoping than building. Startups will expect you to do everything yourself with little supervision; large organisations will wrap the work in approval processes, data protection reviews and risk sign-off, which slows things down considerably. Regulated sectors such as financial services and healthcare add substantial compliance work around anything that touches customer data or decisions.

General guidance about the role across the UK market, not about any specific employer. Entry routes and requirements vary — always check the individual job advert.