What the job looks like day to day
Job titles vary: prompt engineer, AI engineer, LLM application engineer and conversation designer all overlap. The daily work usually includes:
- Writing and revising system prompts for features such as support assistants, document extraction or coding helpers.
- Building test sets of real inputs with pass rules, then running every prompt change against them.
- Reading failures. Most of the job is studying wrong outputs and deciding whether the fix is a clearer instruction, a better example, a tool or a different model.
- Managing cost and length. Trimming prompts, using caching and picking the cheapest model that passes, as covered in our guide to how AI API pricing works.
- Versioning. OpenAI's guide recommends pinning production apps to specific model snapshots and building evaluations, because new snapshots can behave differently.
- Working with product and legal teams on tone, refusals and what the assistant must never do.
In smaller companies this sits inside a software or product role rather than a separate title.
Skills table
| Skill | Why it matters | How to show it |
|---|---|---|
| Precise writing | Ambiguous instructions produce inconsistent outputs | A before and after prompt with the pass rate for each |
| Evaluation design | Without a score, you cannot tell if a change helped | A test set of 30 or more cases with written pass rules |
| Few-shot example design | Google recommends always including examples | A prompt whose examples fixed a named failure |
| Basic Python and APIs | Real prompt work runs in code, not chat windows | A script that runs a prompt over a dataset and scores it |
| Long-context handling | Document order and size change recall and cost | A summary of a long file with quotes tied to sources |
| Model selection and cost | The cheapest passing model saves real money | A table of cost per passed task across two models |
| Domain knowledge | Knowing a correct answer in law, health or finance | A project in your own field with expert-checked cases |
Learning path with free official resources
- Start with the vendor guides. Read Anthropic's prompt engineering overview, which assumes you have success criteria and a way to test against them before you tune anything.
- Work through Anthropic's interactive tutorial. The prompt engineering interactive tutorial on GitHub is broken into 9 chapters with exercises plus an appendix of advanced methods.
- Read OpenAI's guide. The OpenAI prompt engineering guide covers message roles, Markdown and XML structure, and why to pin snapshots.
- Read Google's strategies. The Gemini prompting strategies page explains zero-shot and few-shot prompts and parameters such as temperature.
- Take a short course. DeepLearning.AI's ChatGPT Prompt Engineering for Developers, taught by Isa Fulford and Andrew Ng, is listed as free, beginner level and 1 hour 40 minutes.
- Learn evaluation. Follow OpenAI's Evals guide and our AI evaluation learning path, because testing is where prompt work becomes engineering.
- Practice on long inputs. Our guide to context windows shows how order and length change results.
Anthropic's free prompt engineering interactive tutorial on GitHub, checked 17 October 2026, has 9 chapters with exercises plus an appendix of advanced methods.
Pay: which BLS occupation applies
The US Bureau of Labor Statistics does not track prompt engineer as an occupation, so pay data must come from the occupation the work falls under.
- Software developers. When prompting is part of building and shipping AI features in code, the role sits here. The BLS Occupational Outlook Handbook page gives a median annual wage of $135,980 for software developers in May 2025, and projects 10% employment growth for software developers, quality assurance analysts and testers from 2025 to 2035.
- Data scientists. When the job is mostly building test sets and measuring model quality, it sits closer to data science. The BLS data scientists page gives a median of $120,230 in May 2025 and projects 35% growth from 2025 to 2035.
BLS reports a $135,980 median annual wage for software developers in May 2025, the occupation most prompt engineering roles fall under.
These are national medians for broad occupations; they do not describe any single prompt engineering job.
Portfolio ideas
- Support reply assistant. A prompt for one public company's FAQ, a 40-case test set including policy edge cases, and a log showing the pass rate rising across five versions.
- Invoice field extractor. Pull ten fields from 50 sample invoices into JSON, with accuracy per field and the examples that fixed the worst field.
- Model cost comparison. Run the same task on a small and a mid-tier model, report pass rate and cost per passed task, and recommend one.
- Long-document Q&A. Answer 20 questions about a 200-page public report with quoted evidence, and show how moving the question to the end changed results.
- Prompt injection test. Write ten hostile inputs against your own assistant and document which defenses held.
Keep each project small, reproducible and honest about failures. The AI workflow design path is a natural next step, and our careers and skills hub lists related roles such as the AI product manager guide.
How to get your first role
Most people move into prompt work from an adjacent job rather than applying cold. Three routes come up again and again:
- From software development: volunteer for the AI feature on your current team and own its test set.
- From writing, support or operations: build the prompt library your team already needs, measure time saved and show the numbers.
- From a domain specialty: law, medicine, finance and education teams need people who can tell a right answer from a plausible one, and that judgment is hard to hire.
In each case, lead with a project that has a test set and a pass rate, not a list of clever prompts.
How we researched this
We read Anthropic's prompt engineering overview and interactive tutorial, OpenAI's prompt engineering and Evals guides, Google's prompting strategies page, the DeepLearning.AI course page, and the BLS pages for software developers and data scientists on 17 October 2026. Pay figures are national medians for related occupations.
Frequently asked questions
Is prompt engineer a real job title? Yes, though it often appears as AI engineer or LLM application engineer. BLS has no separate occupation for it, so pay is reported under related occupations.
Do prompt engineers need to code? Most do. Production prompt work runs in code with test sets and APIs, and basic Python is enough to start.
How much do prompt engineers make? BLS does not publish prompt engineer pay. Software developers, the closest occupation, had a $135,980 median annual wage in May 2025.
What is the best free course for prompt engineering? Anthropic's interactive tutorial and DeepLearning.AI's short course are both free. Pair either with the OpenAI and Google guides.
Will prompt engineering disappear as models improve? The narrow skill of clever phrasing matters less as models improve, but defining success, building test sets and choosing models remain core engineering work.
What should a prompt engineering portfolio include? A prompt, a test set with pass rules, results across versions and the failures you fixed. Cost per passed task is a strong extra.
Do I need a degree to become a prompt engineer? Not always, but BLS says software developers typically need a bachelor's degree in computer and information technology or a related field.