Flagship
Designing AI Products
Six weeks on the design decisions that keep an AI feature useful when the model gets it wrong.
$2,150
Cohort 11
January 19 to February 25
Enrollment closes January 12
Not ready yet? Get updates about this program
Full refund until the end of week two.
Your company can pay. Get the email for your manager
Next cohort
January 19 to February 25
Schedule
Tuesdays and Thursdays, 9 to 10:30 am PT (noon to 1:30 pm ET)
Length
6 weeks
Live sessions
12 live sessions, 18 hours
Hours a week
About 5 hours a week
Group
24 people, in critique groups of six
Tuition
$2,150
Taught by
Nadia Kessler and Theo Baptiste
What you’ll be able to do
Map what a model actually returns for your feature, including the failures the demo never shows.
Design the ways people ask: examples, constraints, and suggestions that show what the model can do.
Show evidence and limits, where you have them, so people can check an answer.
Design correction and undo that people find and use.
Build a test set with engineers and review traces with them.
Write a case study that explains your decisions to your team.
Who it’s for
Product and UX designers with two or more years of shipping software, now working on an AI feature.
Product managers who write the specs for AI features.
Design engineers and researchers who sit between design and the model team.
Who it’s not for
People new to design. We don’t teach the fundamentals.
Anyone looking for prompt-writing tips or for how to train models. We design the product around a model; we don’t build the model.
Anyone looking for a first job in design. We don’t do job placement.
Week by week
We start with the feature you brought and fifty real outputs from it, not the demo. You sort them by how they fail: wrong, correct but unhelpful, or not safe to show at all. Then you find where the design has to do the work.
An empty box assumes people know what the model can do. A rigid form hides it. This week covers the ground in between: examples, suggestions, constraints, and the moment someone rewrites a request.
An answer is only as useful as someone’s ability to check it. We work on showing evidence where it exists, marking what the model inferred, and giving a partial answer when a full one isn’t reliable.
Every AI feature ships with wrong answers. This week covers correction, undo, trying again with a hint, and feedback that reaches the team instead of disappearing into a thumbs-down.
We turn your failure map into a small test set, write a rubric in plain language, and review traces with a guest engineer. This is the heaviest week, closer to seven hours.
You present the reworked feature to your critique group and two guest critics. Then you write it up: what you changed, what you tried and dropped, and what you’d test next.
The project
You work on one feature all six weeks. Bring one you’re designing at work. If yours is under NDA, choose one of three course briefs: a support assistant for a regional bank, a search tool for a legal team, or a note-taking assistant for a clinic. The course ends with that feature reworked and a case study written about it.
Instructors
NK
Co-founder
Nadia Kessler
Fourteen years designing software, the last five on products built on language models. Between cohorts she still works with product teams.
TB
Co-founder
Theo Baptiste
Design engineer. Theo spent four years building prototyping and evaluation tools for an AI search team.
From past cohorts
“Dispatchers had stopped using our route suggestions. Accepting one replaced the plan they’d already made, instead of sitting next to it.”
Marcus Hale
Designing AI Products, Cohort 7
“My critique group saw the feature every week, so they caught what I’d stopped seeing: we never told anyone when a category was a guess.”
Ana Lúcia Ferraz
Designing AI Products, Cohort 9
“I used to write ‘the AI handles it’ in specs. Now every spec I write has a section on what happens when it doesn’t.”
Kristen Walsh
Designing AI Products, Cohort 6
What’s included
12 live sessions with Nadia Kessler and Theo Baptiste, with recordings and written notes
A critique group of six for all six weeks
Open office hours every Friday
Three course briefs, if you can’t use your own work
The Seamful field guide: worksheets for failure maps, test sets, and rubrics
Monthly open critiques for six months after the course
A certificate of completion for attending or watching every session and presenting at the final critique
Dates and tuition
Cohort 11 · January 19 to February 25 · Enrollment closes January 12
Tuesdays and Thursdays, 9 to 10:30 am PT (noon to 1:30 pm ET). Every session is recorded.
$2,150, one payment. No application or materials fees.
Your company can pay by invoice or purchase order. We send a W-9 on request. Get the email for your manager
Paying out of pocket, between jobs, or working at a nonprofit? Apply for a scholarship
Full refund until the end of week two. If work gets in the way after that, move once to the next cohort, at no cost, before week four starts.
Questions
Cohort 11 starts January 19
Enrollment closes January 12. Full refund until the end of week two.