How to Learn AI from Scratch: A Practical 28-Day Roadmap

AI feels overwhelming because the loudest voices are selling complexity. You don't need a CS degree, a Python course, or three months of Andrew Ng before you can ship something useful. You need a tight roadmap and a habit.

This guide is the roadmap we hand new students at Cyan. It's the same one our team would follow if we started over today.

What "learning AI" actually means in 2026

For 95% of working professionals, "AI skills" means three things, in this order:

1. Prompting with structure — getting useful output on the first try.
2. Building workflows — chaining prompts and tools into something repeatable.
3. Judging output — knowing when AI is wrong before your boss does.

Everything else — fine-tuning, embeddings, agents, evals — is downstream of those three. Skip them and you'll keep hitting the same wall in week six.

What beginners should not start with

• Python tutorials. They teach syntax, not AI judgment.
• Prompt-engineering certificate dumps. Most are PDFs from 2023.
• LLM theory courses. Useful in month three, not week one.
• Agent frameworks. Don't run before you can prompt.

The four practical skills to learn first

1. Structured prompting

Every good prompt has four parts: role, task, constraints, output format. Write a bad prompt, then refactor it into those four parts. Do this 20 times. You're now ahead of 80% of your office.

2. Workflow design

Pick one repetitive task you do every week — weekly status report, candidate screening, sales follow-up. Map the manual steps. Replace 3 of them with AI. Ship.

3. Tool fluency, not tool hoarding

Pick one chat model (ChatGPT, Claude, or Gemini), one note tool, one automation tool. Master them before adding more.

4. Output review

Build a simple checklist: "Is this factual? Is this in our tone? Is this the level of detail we ship?" Run every AI output through it before it leaves your laptop.

The 28-day beginner roadmap

Thirty minutes a day, five days a week.

• Week 1 — Prompting: learn the four-part structure. Rewrite 25 bad prompts. Save your top 10.
• Week 2 — Workflows: pick one weekly task. Automate three of its steps. Measure time saved.
• Week 3 — Tools: add one automation tool (Zapier or Make). Build one cross-app workflow.
• Week 4 — Project: ship one real artifact — a research brief, a sales sequence, an onboarding doc — that your team uses.

Beginner projects that prove you can do this

• Turn 30 customer emails into a tone-of-voice guide.
• Build a weekly competitor digest that runs while you sleep.
• Convert your meeting notes into next-step actions automatically.
• Draft job-spec → outreach sequence in under five minutes.

Common mistakes

• Reading tutorials instead of shipping artifacts.
• Chasing every new tool launch on X.
• Trying to learn alone without a deadline.
• Confusing "I used ChatGPT today" with "I built a workflow."

Next step

Pick the project, set the calendar block, and start week one. If you want the plan built around your role and tools, take the 2-minute quiz below.