AI Prompt Engineer
Design and optimize prompts for large language models. Help organizations get maximum value from AI tools like ChatGPT, Claude, and Gemini.
What a typical day looks like
Prompt engineering as a standalone role is a 2024 invention. Most days I'm working at the intersection of product, AI, and content. Mornings start with reviewing prompt performance from yesterday — looking at outputs that users rated poorly, finding patterns, then crafting better prompts. By 10 I'm usually iterating on a specific prompt — running it through dozens of variations to see which performs best across a test set. Afternoons are more collaborative: working with product to define success criteria for a new AI feature, working with engineering to wire up new prompt versions, working with subject matter experts to verify outputs are actually correct. The field changes every week. A good prompt engineer is part writer, part data analyst, part product designer.
Hour-by-hour
Skills you need
Required
Nice to have
Build these to stand out
Hands-on projects beat any CV bullet point. Pick one and finish it.
Prompt Engineering Cookbook
Pick a real task (e.g. summarising news, writing job descriptions, explaining code). Try 10 different prompt structures (zero-shot, few-shot, chain-of-thought, role-based, structured output). Document results with concrete examples. Publish on GitHub.
Shows you understand prompting beyond 'be polite to the AI'. Standout portfolio piece.
Evaluation Harness for LLM Outputs
Build a small Python framework that takes a list of prompts, runs them against an LLM, and scores the outputs against criteria you define (accuracy, length, tone, factual correctness). Can use another LLM as judge.
Evaluation is the unsexy but critical part of prompt engineering. Most candidates miss it.
Specialised AI Assistant
Build a small Streamlit or web app that uses an LLM to do one thing very well: e.g. a code reviewer, a meeting note summariser, a recipe generator. Polish the prompt until outputs are consistently good. Deploy publicly.
Working AI products are more impressive than theoretical prompt knowledge.