How to learn AI skills: a 30/60/90-day roadmap for busy professionals

What fluent looks like

Fluent does not mean you can explain how a transformer works. It means four things: you can get useful, reliable output from an assistant on your own work; you know when not to use it; you catch it when it is wrong; and you can show a colleague how to do the same. Everything in this roadmap serves those four.

You can get there in 90 days with about 20 minutes a day, as long as those minutes are spent on real tasks rather than on watching other people use the tools.

Before day 1: a 30-minute setup

1. Find out what your workplace allows. If there is a policy, read it. If there is not, ask your manager which tools are approved and read AI privacy at work so you know what never to paste.
2. Pick one assistant and stay with it for the first month. ChatGPT, Claude, Gemini, and Microsoft Copilot all have free tiers or come bundled with software you already use. The comparison guide helps if you are torn, but the differences matter less than your reps.
3. Take the AI readiness quiz and save your result. You will retake it on day 90.
4. Open a plain document called "AI log" with four columns: date, task, what worked, what I had to fix. This log is the single most important artifact in the whole plan.

Days 1 to 30: build the habit and the vocabulary

What to practice

One real task a day, done with the assistant, then verified. Not a demo, not a toy. The email you were going to write anyway, the document you had to read anyway.

Every prompt you write this month should contain five parts: who the assistant should act as, what you want, the context it needs, the format you want back, and any constraints. This is the structure the prompt builder walks you through; use it until the five parts are automatic.

After every response, do one of three things: accept it, ask for a specific change, or start over with a better prompt. Never accept a first draft without reading it.

Weekly routine

• Monday, 10 minutes: choose this week's task type (below) and one real task for each day.
• Daily, 20 minutes: run the task, verify the output, log one line.
• Friday, 15 minutes: reread the log. Write one sentence about what you learned. Look up any term you did not understand in the glossary.

Week by week

• Week 1, writing: emails, summaries of your own notes, rewording something for a different audience. Learn to ask for three versions and pick one.
• Week 2, reading: paste a long document (a public one, or an internal one your policy allows) and ask for a summary, then for a list of questions the document does not answer. Check the summary against the source. Notice where it flattened nuance.
• Week 3, thinking partner: use the assistant to plan a meeting, argue the other side of a decision, or list what could go wrong with a plan. Learn to ask "what am I missing?"
• Week 4, numbers: give it a small table and ask for a formula, a chart description, or a plain-English explanation of what the numbers say. Check the math yourself. This week teaches you the model's limits faster than any other.

Milestone at day 30

You should be able to write a five-part prompt without thinking about it, name three tasks where the assistant reliably saves you time, and name two where it does not. If you cannot, extend this phase by two weeks. There is no prize for rushing.

Days 31 to 60: build workflows and a first project

What to practice

• Reusable prompts. Take the three tasks that worked and turn each into a saved prompt with [BRACKETED PLACEHOLDERS] for the parts that change. Keep them in your log.
• Standing context. ChatGPT Projects, Claude Projects, and Gemini Gems let you store instructions and reference files so you stop re-explaining your job every time. The guide to custom GPTs, Projects, and Gems covers the setup.
• Examples. Show the assistant two of your best past outputs and ask it to match them. This habit, called few-shot prompting, fixes most "it does not sound like me" complaints.
• Structured output. Ask for tables, checklists, and numbered steps so results drop straight into your work.
• A second tool. Try one other assistant on the same tasks for a week and note the differences. Fluency includes knowing that the tools are not identical.

Projects to build

Pick one and finish it. A finished small project beats three half-built ambitious ones.

• A weekly report generator: a saved prompt plus a template that turns your raw notes into the report your manager reads.
• Meeting notes to action items: a workflow that takes a transcript from Otter.ai, Fireflies, or Zoom AI Companion and produces owners, deadlines, and open questions.
• A personal reference assistant: load your team's process documents into NotebookLM or a Project and use it to answer "how do we handle X?" questions, with citations you can check.
• A spreadsheet cleanup routine: a prompt that standardizes messy exports, following the spreadsheets guide.

Milestone at day 60

One finished project that you have used at least five times, three saved prompts, and a log with at least 40 entries. Show the project to one person.

Days 61 to 90: become the person others ask

What to practice

• Verification as a discipline. Read how to fact-check AI output and write your own checklist for each type of output you produce.
• Light automation. Connect one trigger to one action with Zapier or Make, or simply schedule a recurring prompt. The workflow automation guide starts small on purpose.
• Teaching. Explain your project to a colleague and watch where they struggle. Their confusion shows you what you have not yet understood.
• Risk. For each workflow you built, write down what happens if the output is wrong and who would notice. Adjust the checks accordingly.

Projects to build

• A one-page team playbook: the three workflows, their prompts, their checks, and the data rules. Offer it to your manager.
• A 30-minute lunch-and-learn. Demo one workflow live. Take questions. Keep the slides to five.
• One automation that removes a manual step from a workflow you already trust.

Milestone at day 90

Retake the AI readiness quiz. Compare your log from week 1 to week 12. You should see prompts getting shorter and more precise, fewer fixes per task, and at least one task where you no longer bother with the assistant because you learned it is not worth it there. That last one is a sign of fluency, not failure.

A worked example: a project coordinator's 90 days

Take a project coordinator at a mid-sized construction firm who has never used an AI assistant. Her days are status emails, meeting notes, schedule updates, and chasing subcontractors.

Month one, she uses the company's approved assistant for status emails and meeting summaries. Her log shows the summaries miss decisions that were phrased as questions in the meeting, so she learns to ask specifically for "decisions, including ones stated tentatively." Week 4 with numbers goes badly: the assistant miscalculates a schedule float, and she catches it. That log entry becomes the story she tells in interviews later.

Month two, she builds the meeting-notes workflow: transcript in, action items with owners out, reviewed against the recording before sending. She saves three prompts and starts a Project loaded with the firm's standard schedule terminology.

Month three, she writes a two-page playbook, runs a lunch-and-learn for the other coordinators, and sets up one automation that files action items into the team's task tracker. Her manager asks her to evaluate a vendor's AI scheduling add-on. She is now the person others ask, ninety days after starting from zero.

Free resources that are enough

You do not need to buy a course to get fluent. Everything below is free.

• Vendor documentation and prompting guides. Each major assistant publishes a help center with prompt examples and feature explanations. They are the most accurate source for what the tool can do this month.
• Free tiers. Every major assistant has one. Limits on usage or model choice matter less than you think during the first 60 days.
• The prompt library on this site, sorted by task: writing and email, meetings and summaries, data and spreadsheets, and more. Copy, adapt, log.
• The glossary for the vocabulary you will keep bumping into: context window, hallucination, few-shot prompting.
• University-hosted introductory courses and vendor academies. Useful for concepts in month two; skip them in month one, because reps matter more than theory at the start.
• Communities and newsletters. Pick one, not five. Use them to see what other people in your field are doing, not as a substitute for practice.

Paid help is worth it once you have done the free reps and want structure, feedback, and exercises. That is what the AI at Work course is for.

How to measure progress

Feelings of progress are unreliable. Use a scorecard every two weeks.

A prompt that turns the assistant into your tutor

Save this in a Project or as a custom instruction and run it once a week with your log entries.

Act as a patient, practical coach helping a working professional get fluent with AI assistants. You are not here to impress me; you are here to make next week's practice better.

About me: I am a [YOUR JOB TITLE] in [YOUR INDUSTRY]. I have been practicing for [NUMBER] weeks. My workplace allows [APPROVED TOOLS] and prohibits pasting [TYPES OF DATA I CANNOT SHARE].

My log from this week, one line per task:
[PASTE YOUR LOG ENTRIES]

Do the following:
1. Point out the two habits in my log that are working and explain why, one sentence each.
2. Identify the single biggest mistake or missed opportunity you see, with a corrected example prompt I could have used.
3. Suggest three specific practice tasks for next week, drawn from the kind of work in my log, in increasing difficulty.
4. Give me one verification check I should add to my routine.

Keep the whole response under 350 words. Use plain language. If my log is too thin to judge, ask me up to four questions first.

Where this site's free tools fit

• The prompt builder is for month one: it forces the five-part structure until it sticks.
• The AI readiness quiz is your day 1 and day 90 benchmark.
• The prompt library is the source for daily reps when you are out of ideas.
• The glossary answers the "what does that word mean?" moments on Fridays.
• The time savings calculator turns your log into a number you can show your manager.
• The AI at Work course is the structured version of this roadmap, with exercises and a 90-day plan module, for when you want guidance instead of self-direction.

Next steps

• Read the prompt engineering guide during week 1.
• Bookmark how to fact-check AI output for month three.
• See how this applies to your role in the AI for your job playbooks, such as AI for project managers or AI for marketers.
• When you are ready to turn practice into career evidence, read AI skills on your resume and the bigger picture in How to AI-proof your career.