Tau
AI for product & design · 11 lessons

AI for product managers and designers

Use AI to make sense of user interviews, write specs, build quick prototypes and check designs for accessibility. For product managers and designers, including those building AI features.

Who it’s for: Product managers, product owners and UX/UI designers.

AI can help product managers and designers move from a vague problem to a working prototype in a day. That speed is useful, but it doesn't replace talking to real users. This course shows you where AI helps, and where only real people and data will do.

You'll use AI to synthesise user interviews, with a word-for-word quote behind every theme. You'll draft a PRD, a one-page product requirements document, then have AI attack it as a sceptical engineer and designer would. You'll run a competitor teardown where every claim links to a source. You'll turn a problem into a clickable prototype in about an hour. You'll also write clear interface copy for error messages and empty screens.

Later lessons sharpen the craft. You'll generate design variations, then judge them against the user's goal rather than polish. You'll catch accessibility issues such as poor contrast, focus order and missing labels before handoff. If you build AI features, you'll plan how they behave when they fail and how they tell people AI is involved. You'll also build an evaluation set, a list of real examples with pass and fail criteria, that decides whether a feature ships.

By the end you’ll be able to…

  • Synthesise interviews into themes where every insight cites a quote
  • Turn a problem statement into a clickable prototype to test with users
  • Check a design for contrast, focus order, target size and labels
  • Build an evaluation set that decides whether an AI feature ships

Short on time? Start with the essentials

These 4 lessons cover the most useful ideas in this course. Each one takes about a minute.

Essential 1

Use AI for speed, keep real users for truth

Separate the work AI speeds up (synthesis, drafts, prototypes, variations) from the evidence only real users and data can give, and never present AI personas as research.

Essential 2

Synthesise interviews with a quote behind every theme

Run consented, de-identified transcripts through AI to tag and cluster findings, require a verbatim quote and participant ID for every theme, and check the quotes and the counter-evidence yourself.

Essential 3

Draft a PRD, then have AI attack it

Turn research and a problem statement into a one-page PRD with users, success metrics, scope, non-goals and open questions, then have AI critique it as a sceptical engineer and designer.

Essential 4

Turn a problem into a testable prototype in an hour

Prompt an AI builder for one key flow with realistic content, error and empty states, so you can put a working prototype in front of users the same day.

Every lesson in this course

This course has 11 video lessons of about a minute each, from the basics to expert habits. Each one ends with a question, three takeaways and its sources.

Foundations 2 lessons

  1. Use AI for speed, keep real users for truthEssential

    Separate the work AI speeds up (synthesis, drafts, prototypes, variations) from the evidence only real users and data can give, and never present AI personas as research.

    “AI 'synthetic users' never get confused, bored or lost. Your real users do.”

  2. Write UX copy for errors and empty states

    Give AI the user's situation, your voice guide and the character limit to draft microcopy for errors, empty states and confirmations, then pick the version that says what happened and what to do.

    “'Error 403' → 'You need edit access. Ask the file owner to share it.'”

Building skills 4 lessons

  1. Synthesise interviews with a quote behind every themeEssential

    Run consented, de-identified transcripts through AI to tag and cluster findings, require a verbatim quote and participant ID for every theme, and check the quotes and the counter-evidence yourself.

    “12 interviews → 5 themes, each backed by quotes you can find in the transcript.”

  2. Draft a PRD, then have AI attack itEssential

    Turn research and a problem statement into a one-page PRD with users, success metrics, scope, non-goals and open questions, then have AI critique it as a sceptical engineer and designer.

    “Ask AI to tear up your spec like a tired senior engineer. Fix what it finds first.”

  3. Run a competitor teardown from public evidence

    Use AI to turn a competitor's pricing page, onboarding, release notes and reviews into a feature and positioning table where every cell links to a source and guesses are labelled.

    “Pricing page, changelog, 200 app-store reviews → one teardown, every cell sourced.”

  4. Turn a problem into a testable prototype in an hourEssential

    Prompt an AI builder for one key flow with realistic content, error and empty states, so you can put a working prototype in front of users the same day.

    “From a whiteboard sketch to a clickable flow you can test with users this afternoon.”

Advanced 4 lessons

  1. Generate design variations, then critique them hard

    Generate several layout or flow variations, then judge each against the user goal, usability heuristics and your design system, choosing on evidence rather than polish.

    “Ten polished variations in a minute. The skill is knowing which nine to bin.”

  2. Catch accessibility issues before handoff

    Combine AI review and automated checkers to test a design for WCAG 2.2 contrast, focus order, target size, labels and alt text, then confirm with a keyboard and a screen reader.

    “4.5:1 contrast, 24px targets, visible focus: three checks before any handoff.”

  3. Ask product analytics questions in plain English

    Turn a product question into a precise metric definition, have AI draft the query or use your analytics tool's assistant, and sanity-check the result against a known baseline.

    “'Did the new onboarding work?' Define 'work' first, then let AI write the query.”

  4. Design how an AI feature fails and discloses

    Specify what an AI feature shows when it is wrong, unsure or unavailable: clear AI disclosure, sources or confidence cues, easy undo, a human fallback and a way to report errors.

    “Every AI feature is wrong sometimes. Design that moment before launch, not after.”

Expert 1 lessons

  1. Build an eval set that decides if an AI feature ships

    Collect 50 to 100 real examples, write pass/fail criteria from reading actual failures, grade the feature against them and set the launch threshold with engineering.

    “'It looked great in the demo' isn't a launch bar. 60 graded real examples are.”

Apps we recommend

These are the tools this course uses, and what each one is best for. We have no partnership with any of them.

Dovetail

Best for: Research repository and AI synthesis of interviews

Transcribes and tags interviews and links AI-suggested themes back to the highlighted quotes, so every insight can be traced.

Freemium

Figma Make

Best for: Prompt-to-prototype inside your Figma files

Turns a prompt or an existing Figma design into a working interactive prototype you can test and hand to engineering.

Freemium

Lovable

Best for: Full-stack app prototypes from a description

Builds a working web app with real data and logins from chat, for testing ideas that a clickable mock-up can't show.

Freemium
Hands-on project

Take one real problem from evidence to tested prototype

Pick a real problem on your roadmap. With approved tools and de-identified research, synthesise the evidence, write a one-page spec, build a prototype of the key flow, check its accessibility and test it with three people. If the feature uses AI, add its failure states and a small eval set.

What you’ll make: A one-page spec, a testable prototype, an accessibility check, and notes from three test sessions with the changes you made.

Check your own work

  • Every problem claim in the spec links to a research quote or data point
  • The prototype covers the main flow plus at least one error and one empty state
  • Contrast, focus order, target size and labels were checked and issues listed
  • Findings from three test sessions are recorded with what changed as a result
  • If AI is involved, failure states, disclosure and a 20+ example eval set exist

Further reading and viewing

We’ve chosen these articles, research papers, videos and guides from other trusted sources.

Article · Nielsen Norman Group · 12 min

Synthetic Users: If, When, and How to Use AI-Generated "Research"

Where AI-simulated users help and why they can't replace talking to real people.

Guide · Google PAIR · 45 min

People + AI Guidebook

Patterns for designing AI features: user needs, trust, explanations, errors and feedback.

Research · Microsoft HAX Toolkit · 20 min

Guidelines for Human-AI Interaction

18 research-tested guidelines for how AI features should behave, fail and recover.

Article · Hamel Husain · 25 min

Your AI Product Needs Evals

A practitioner's guide to building evaluations from real examples and error analysis.

Guide · W3C Web Accessibility Initiative · 30 min

How to Meet WCAG (Quick Reference)

The official, filterable checklist of WCAG 2.2 success criteria and techniques.

Questions about this course

How can product managers use AI?

AI helps with synthesis, drafting and prototyping. You can turn interview notes into themes, draft a PRD and ask AI to critique it, and question your product analytics in plain English. Tools such as Dovetail help with research, while Figma Make and Lovable turn a description into a prototype. Decisions about what to build should still rest on evidence from real users and data.

Can AI build a prototype from a description?

Yes. AI builders such as Figma Make and Lovable can turn a prompt into a clickable or working prototype. The trick is to ask for one key flow, with realistic content, error states and empty states. That way you can put it in front of users the same day. The course shows you how to write that prompt and test the result with three people.

Can I use AI personas instead of user research?

No. AI personas can help you brainstorm, but they aren't evidence of what real users think or do. When you use AI on actual research, use consented, de-identified transcripts in a tool your company has approved, and follow its rules on research data. Check every quote against the source transcript. The course also covers accessibility checks against WCAG 2.2, confirmed with a keyboard and a screen reader.

How long are Tau's product and design lessons?

Each lesson takes about a minute. You watch a short video, answer one question, then get three takeaways and the sources. This course has 11 lessons, and many let you try the skill on a realistic task with live AI feedback. Your first 5 lessons in Tau are free, and Tau Pro gives you every course and lesson.

Get early access to this course

Tau opens soon. Join the waitlist and we’ll email you as soon as you can start this course.