Engineering role
AI Prompt Engineer
An AI prompt engineer that turns a vague product requirement into a versioned prompt with an output format, examples and the test cases that prove it.
Give it the prompts in your repository and what the feature should do, and it does the careful work that keeps an LLM feature predictable: replacing “be concise” with a rule the model can follow, fixing the output format, choosing examples that cover the edge cases, and writing the test set that tells you when a prompt change or a model upgrade broke something. It treats each prompt as code, with a version and a changelog.
Things you could ask for
Written the way you would actually say them. The agent plans the steps itself.
- “Read the system prompt for our support classifier and rewrite it with an explicit JSON output, a fallback for out-of-scope messages and three examples.”
- “Write 30 test cases for this prompt in a Google Sheet — happy path, edge cases, prompt injection, empty and non-English input — each with the behaviour we expect.”
- “Review the prompt changes in pull request #88 and tell me what behaviour they could change that the description does not mention.”
- “We are moving this feature to a different model. List the parts of the prompt most likely to behave differently, and what to test first.”
- “These ten outputs are wrong. Read the prompt and tell me, for each, whether it is the instructions, the examples or the missing context.”
What you get back
Prompts as versioned files with a changelog entry, test suites as rows in a sheet or as test code in your repository, and review notes that tie each risk to a line of the prompt. Every rewritten prompt states its output format and its success criteria.
Tools it works with
A role is a job description; connectors are what let it do the job on your real work instead of on whatever you paste into a chat.
Where it is the wrong tool
- It does not run your prompts against your production model. A remote agent has no shell in the cloud and no connector to your LLM provider; it writes the test suite, and you run it — or a local agent with shell access in the desktop app runs it on your machine.
- Behaviour differs between models and temperatures. Its predictions about how a given model will respond are informed guesses until the tests have run on the model you ship.
- It cannot see production traffic. Paste or export the real inputs and the bad outputs; test cases built from real failures are worth more than invented ones.
- It will not write prompts meant to get around another model’s safety rules or to extract a system prompt that is not yours.
Starting an agent with this role
- 1Create a Burrak account — any plan works, and your first week is $1.
- 2In the Marketplace, open the Roles tab, find Prompt Engineer and choose “Start an agent as Prompt Engineer”. That creates a remote agent with the role’s instructions in its system prompt.
- 3Connect the tools it needs — GitHub, GitLab, Google Sheets, Notion, Linear, arXiv — from Connectors.
- 4Give it a brief. It keeps working in the cloud after you close the tab, and you are only charged while it is actually working.
The Prompt Engineer role, in practice
- Is prompt engineering still needed with newer models?
- Less of it is wordsmithing, more of it is specification and testing. Newer models follow instructions better, which makes a precise output format, clear scope and a regression suite worth more, not less — they are what let you upgrade the model safely.
- How is this different from the AI Engineer?
- The AI Engineer designs the whole feature around the model — retrieval, data, tools and evaluation. The Prompt Engineer works on the instructions themselves and the tests that pin their behaviour down. On a small team one person does both; the roles split the work where it naturally splits.
- Does it work with any model provider?
- Yes. Prompts and tests are plain text and code in your repository, so they are not tied to a provider. It will point out where a technique is specific to one model family so you know what to retest when you switch.
Capability reference
Summarised from the role’s instructions, which are adapted from the agency-agents collection (AgentLand Contributors), used under the MIT licence.
- System prompts with an explicit role, output format, length, scope and fallback
- Few-shot examples chosen to cover edge cases, not just the happy path
- Prompt test suites covering the happy path, edge cases and failure modes
- Versioned prompts with changelogs, treated like code
- Ambiguous product requirements translated into behavioural specs a model can follow
Your AI assistant is waiting.
Put it to work today.
Put an agent on the work that never needed a human in the first place — the research, the reports, the follow-ups — and take the week back.
$1 for your first week · Cancel anytime · Works on Mac, Windows, iPhone, Android & Web