Prompt engineering has quietly become one of the most in-demand AI roles at UK tech companies. But the job title still confuses people — partly because it covers meaningfully different work depending on where you work. This guide explains what prompt engineers actually do.
What Prompt Engineers Actually Do
The popular image — someone typing clever questions into ChatGPT all day — is a long way from reality. At a UK product company, prompt engineering involves:
- Designing system prompts that control LLM behaviour across an entire product surface, not just individual interactions
- Building evaluation harnesses — automated test suites that measure prompt performance across hundreds or thousands of representative inputs
- Collaborating with product teams to define what "correct" LLM behaviour looks like for a given feature, which is harder than it sounds
- Managing edge cases and safety failures — what happens when users try to manipulate the model, use it in unexpected ways, or ask questions outside the intended scope
- Optimising for cost and latency — at scale, token efficiency matters; a prompt that uses 30% fewer tokens on a system handling millions of requests per day has real financial impact
The split between writing prompts and evaluating them is roughly 40/60 at most product companies. The evaluation side — building systematic ways to measure quality — is where a lot of the genuine skill lies.
A Day in the Life at a UK AI Company
In practice, a prompt engineer's day might look like this: reviewing the results of overnight evaluation runs to see how last night's prompt revision performed against the test suite; writing new adversarial test cases for a jailbreak scenario the safety team flagged; a meeting with the product team to agree expected behaviour for a new feature; then several hours of hands-on prompt iteration, testing each change against a 500-case eval set before documenting the change in a prompt version registry.
The work is iterative and empirical. You form a hypothesis ("I think adding this instruction will improve refusal accuracy"), test it systematically, measure the result, and decide whether to ship. It's closer in spirit to scientific experimentation than traditional software development.
What separates good prompt engineers from great ones
The ability to systematically measure quality. Anyone can tweak a prompt until it produces better output on one input. Good prompt engineers build evaluation infrastructure so they can make changes confidently without regressing on cases they can't see.
How the Role Differs by Company Type
The job title "prompt engineer" covers quite different roles depending on company size and maturity:
At an early-stage startup
You're likely the only specialist. You set up evaluation infrastructure from scratch, own multiple product features, and work directly with founders and engineers on LLM product strategy. Expect to do a lot of things that aren't strictly "prompt engineering" — documenting model behaviour, training the team on LLM capabilities and limitations, fielding questions from across the company. High autonomy, broad scope.
At a scale-up
More specialised. You might own a single product area — the customer support bot, the search feature, the document summariser — and go much deeper on it. There's usually an established eval infrastructure to work within, other prompt engineers to collaborate with, and clearer product requirements coming from a dedicated PM.
At an enterprise (bank, law firm, consulting firm)
Governance-heavy. Enterprise prompt engineers spend significant time on compliance documentation, prompt libraries with approval workflows, audit trails, and internal tooling. The output quality bar is often lower than at a consumer AI company, but the risk tolerance is also much lower — a hallucinated output in a legal or financial context has real consequences. Expect more stakeholder management and fewer opportunities to experiment freely.
Skills You'll Need
Technical: Python at scripting level (calling APIs, processing JSON, running evaluation pipelines), familiarity with LLM APIs (OpenAI, Anthropic Claude, Google Gemini), prompt management tools, version control (Git), and basic data analysis to interpret evaluation results.
Non-technical: Strong writing ability — not just prose, but the ability to write precise, unambiguous instructions that a language model will interpret correctly. Systematic thinking: you're essentially running experiments, so rigour matters. Communication skills: you need to explain model behaviour to stakeholders who don't understand why the model does what it does.
UK Companies Hiring Prompt Engineers
The clearest signal that prompt engineering is a real discipline: companies are advertising dedicated roles with specialist requirements. UK companies that have publicly advertised prompt engineering or LLM specialist roles include Cleo (personal finance AI), PolyAI (conversational AI for enterprises), Intercom, and a growing number of fintech, legaltech, and healthtech companies deploying AI-powered features.
Beyond AI-native companies, professional services firms — the Big Four accountancies, major law firms, management consultancies — are building internal AI teams that include prompt specialists to manage their LLM deployments.
Explore the full Prompt Engineer career guide
Salary tables, required skills, UK career paths, and how to get hired — covered in full on the Prompt Engineer role page.
Frequently Asked Questions
What does a prompt engineer do all day?
The work divides roughly into prompt iteration (designing, testing, and refining prompts) and evaluation (building and running test suites to measure prompt quality). At most product companies, evaluation work takes more time. You'll also spend time in meetings with product and engineering teams, documenting prompt changes, and managing edge cases.
Do you need to code to be a prompt engineer?
Basic Python is increasingly expected — enough to write evaluation scripts, call APIs, and work with JSON. You don't need ML expertise, but you need to be comfortable in a development environment.
Is prompt engineering the same as AI engineering?
No. AI engineers build and deploy full systems, typically with strong software engineering backgrounds. Prompt engineers specialise in LLM behaviour and output quality. The roles overlap but are distinct — prompt engineers tend to have stronger writing and evaluation skills; AI engineers stronger infrastructure and systems skills.
What tools do prompt engineers use?
The typical stack: LLM APIs, prompt management tools like LangSmith or PromptLayer, evaluation frameworks like Promptfoo or DeepEval, and orchestration tools like LangChain or DSPy.
Where do prompt engineers work in the UK?
Primarily in London, though remote roles are common. AI-native companies, scale-ups building AI features, enterprise companies deploying internal AI tools, and increasingly professional services firms running AI transformation programmes.