AI career skills roadmap: from literacy to building

Employers increasingly expect people in every kind of role to use AI tools sensibly, and many want more than that. The trouble is that "learn AI" is too vague to act on. It could mean writing a better prompt in a chat assistant, or it could mean training models on GPUs. Those are different careers with different learning paths.

This roadmap breaks AI skills into four tracks that build on each other, then gives you a 90-day plan to move up at least one level. It also covers the part most courses skip: proving what you can do, describing it on a résumé without exaggeration, and keeping up with a fast-moving field without burning out.

You do not need to reach the last track. For most people, solid literacy plus power-user skills will make a bigger difference to their work than half-finished engineering knowledge.

The four skill tracks

Track 1: AI literacy for every role

Literacy means understanding what AI tools are, what they are good and bad at, and how to use them responsibly. Everyone who uses a computer at work benefits from this level.

• Explain in plain words how generative AI produces text and images, and why it can state false things confidently. Start with what generative AI is.
• Know the main tool types: chat assistants, writing and meeting tools built into office suites, image generators and coding assistants.
• Recognise sensitive data and know your organisation's rules about where it can go.
• Verify outputs: check facts, numbers and citations against original sources.

Track 2: Power user

Power users get consistent, high-quality results and have built AI into their weekly work. This is where most of the everyday productivity gain sits.

• Write structured prompts with role, context, examples, constraints and output format. The prompt engineering guide covers this in depth.
• Work in iterations: draft, critique, revise, rather than expecting one perfect answer.
• Use file uploads, long documents, data analysis features and reusable instructions or custom assistants.
• Choose the right tool for a job and understand trade-offs between models, covered in choosing an AI model.
• Maintain a personal library of tested prompts for recurring tasks.

Track 3: Automation builder

Builders connect AI to other software so work happens with less manual effort. This suits operations, marketing, analyst and IT-adjacent roles.

• Use no-code workflow platforms to chain triggers, AI steps and actions.
• Understand APIs, structured outputs such as JSON, and basic data handling in spreadsheets or databases.
• Design human review points, error handling and logging so automations fail safely.
• Grasp retrieval-augmented generation, where AI answers from your own documents. This is the basis of most internal knowledge assistants.

Track 4: AI product or ML engineer

This track is for people building AI features into products or working on models themselves. It is a genuine engineering career and takes considerably longer than 90 days.

• Programming fluency, usually in Python, plus software engineering practice: version control, testing and deployment.
• Calling model APIs, prompt and tool design for agents, evaluation methods and monitoring in production.
• For machine learning roles: statistics, linear algebra, data pipelines and model training and fine-tuning.
• Security, privacy and responsible AI practices, including frameworks such as the NIST AI Risk Management Framework.

Choosing your target track

If you are unsure where your organisation stands overall, the AI readiness score quiz can help frame the conversation with your manager.

A 90-day learning plan

This plan assumes roughly three to five hours a week alongside a job. It takes a literate beginner to confident power use, then into first automation projects. Engineers can compress the early phases and spend more time in phase four.

Phase 1 (weeks 1–2): foundations

1. Week 1: Learn how language models work at a conceptual level and read your employer's AI policy. Set up one chat assistant and use it daily for low-risk tasks.
2. Week 2: Start a learning log. For every task you try, note the prompt, the result, what you corrected and the time saved or lost. Practise fact-checking three outputs against sources.

Phase 2 (weeks 3–6): power-user habits

3. Week 3: Rewrite your five most common work prompts using a structured template. Compare old and new results.
4. Week 4: Work with documents and data: summarise a long report, extract a table from a PDF, analyse a spreadsheet and verify the numbers manually.
5. Week 5: Build one custom assistant or saved instruction set for a recurring task, and share it with a colleague for feedback.
6. Week 6: Try a second assistant on the same tasks and write down where each performs better.

Phase 3 (weeks 7–10): first automation

7. Week 7: Map one manual process step by step and mark which steps involve reading or writing text.
8. Week 8: Build a simple automation on a no-code platform using test data. The AI automation workflows guide has starting patterns.
9. Week 9: Add a human approval step, handle errors and run it for a week alongside the manual version.
10. Week 10: Document it: purpose, steps, risks, results and how to switch it off.

Phase 4 (weeks 11–13): proof and next steps

11. Week 11: Turn your two best results into short case studies (see below).
12. Week 12: Present one of them to your team or a peer group, and run a short demo.
13. Week 13: Review your log, decide whether to go deeper on the current track or move to the next one, and set the next 90-day goal.

Build a portfolio of proof

Certificates show you finished a course. A portfolio shows you can produce results, which is what hiring managers and clients want to see. Aim for three to five pieces, each following the same structure:

1. Problem: the task and why it mattered.
2. Approach: tools used, prompt or workflow design and how you checked quality.
3. Result: what changed, measured honestly. "Weekly report preparation went from most of a morning to under an hour, with a manual check of all figures" is better than a vague claim.
4. Limitations: what still needed human judgment or did not work. Including this signals maturity.

Good portfolio formats include a one-page write-up, a short screen recording, a public repository for technical work, or a reusable prompt set. Never include confidential data or client material without permission.

Talking about AI skills on your résumé

Recruiters see "proficient in ChatGPT" constantly, and it says almost nothing. Describe what you achieved with AI, not which logo you clicked.

• Weak: "Experienced with AI tools including ChatGPT and Claude."
• Stronger: "Designed a reusable AI-assisted workflow for drafting client proposals, adopted by a team of six, with a review checklist to verify pricing and terms."
• Stronger: "Built a no-code automation that summarises and routes incoming support requests, with human approval before replies are sent."

Put specific skills in your skills section using generic terms that age well, for example "prompt design, AI-assisted data analysis, workflow automation, output evaluation". In interviews, be ready to walk through one case study, including what went wrong. Do not overstate your level: claiming engineering skills you cannot demonstrate in a technical interview damages trust quickly.

Keep up without burning out

AI news moves faster than anyone can follow, and trying to track every release is a common cause of fatigue. Most announcements do not change how you should work this month.

• Choose a small set of sources: the official blogs or changelogs of the tools you actually use, plus one or two trusted newsletters or communities.
• Batch it: one fixed slot a week for reading, not constant checking.
• Test before you switch: run a new tool on three of your real tasks before adopting it.
• Focus on durable skills: problem framing, evaluation, data handling and clear writing outlast any single product.
• Use a reference: keep the AI glossary handy for unfamiliar terms instead of researching each one in depth.

This guide is general information, not professional advice. Spotted an error? Tell us.