The AI-Powered Design Sprint Playbook (v0.1)

I recently posted about the “holy sh*t” moment I had during a 3-Day Design Sprint we ran that centered around AI prototyping tools.

• By Day 2 we had 12 prototypes.
• By Day 3 we had launched three unmoderated user tests with three completely different prototypes.

?? The Results

• We usually end a 5-day sprint with 1 prototype. This time, in just 3 days, we wound up with 15 user tests across 3 solid real-code prototypes, plus a bunch of extra “out there” concepts.

I’ve gotten a bunch of questions about what exactly we did, so here’s a naive attempt at a playbook. This tech is moving so fast that it might be irrelevant in a few weeks, but I’d love feedback either way. We are all learning this together!

?? What You’ll Need

Even novices can easily do one of these sprints. All you need is:

• A basic familiarity with Design Sprints. The Design Sprint Book is an awesome resource.
• A small team of 4-7 people
• A user testing tool (or easy access to internal users)
• Roughly 5-6 hours a day for 3 days

?? Table of Contents of this Article

1. What We Did
2. What Blew My Mind
3. Challenges We Ran Into
4. ?? The Sprint Schedule We’ll Try Next Time ???This is probably the schedule you should try!
5. Where to Start
6. Tips For Running Your Own AI-Powered Sprint
7. What’s Next?

1. What We Did: An AI-Powered Design Sprint

We started the the sprint with our normal design sprint process, but completely flipped things for Day 2. Here’s the quick rundown:

Day 1: Problem Definition, Framing, and Target User

This is basically following The Design Sprint Book Day 1 pretty closely.

?? Day 1: What we did in the morning ??

• The classic sprint discussions: goals, possible pitfalls, target audiences, and success metrics.

?? Day 1: What we did in the afternoon ??

• Identify problem areas where a prototype might help validate some solutions.
• Assign small groups or individuals to dive into different parts of the product. We broke these up by critical points in the user journey. (Example: Onboarding, Managing tasks, and Analytics/Insights)

??? Day 1: Tools We Used ???

• We used a cool Coda template, but Notion also has something similar.

Day 2: AI-Powered Prototyping & Iteration

This was the fun part. Everyone spent the day of prototyping.

?? Day 2: What we did in the morning ??

• No sketches or whiteboards.
• Each person individual built a prototype using AI tools (we used v0.dev, Bolt.new, and some Figma to feed in existing designs)
• After an hour, everyone did a Show & Tell of what they created, and we voted on which elements of each have the most promise.
• This felt like a very short amount of time, but we were shocked at how much everyone produced consider we were all still novices.

?? Day 2: What we did in the afternoon ??

• Pick which prototypes to refine. We broke into pairs to do this part. If we really want to just refine an existing prototype we forked them and kept iterating. In other cases we just started from scratch with a new prompt that incorporated ideas from multiple of the morning’s prototypes.

??? Day 2: Tools We Used ???

• v0.dev, Bolt.new (Lovable.dev and Replit.com likely would have also worked)
• Figma to push existing design components into the AI tools (if they existed).

Day 3: Culling and Launching User Tests

This day got a bit weird because we had SOOOO many prototypes, more than we could realistically test with users. Next time we’ll do this day differently (see tips below).

?? Day 3: What we did in the morning ??

• Another round of show-and-tell
• We tried to stitch things together into one super-prototype, but that proved to be really difficult. The prototypes were so wildly different that there wasn’t a clean way to “merge” them.
• We ended up choosing 3 prototypes that we wanted to get ready to test with users. If it’s truly new or challenges a risky assumption, test it. If it’s redundant or obviously off-track, skip it.

?? Day 3: What we did in the afternoon ??

• We wrote unmoderated user testing scripts for the prototypes
• Polish prototypes to match the script
• Launch the user tests and start getting feedback!

??? Day 3: Tools We Used: ???

• We used UserBrain.com for quick user feedback sessions (but UserTesting.com, Maze.co, and Lookback.com all likely would have worked)

2. What Blew My Mind

AI contributed more to solution ideation than I did

AI didn’t just turbocharge tasks; it actually contributed real ideas. It showed us new public APIs we hadn’t heard of, integrated fresh data sources, and stumbled on design concepts we’d never have considered otherwise. (Example: It found a free service to generate realistic user personas, with names, photos, backgrounds, bios. It made our mock data feel so much more realistic than it would have otherwise)

No attachment anxiety

Usually, if we decide to scrap a design that someone put a lot of time into and start over, it hurts. Here, we just re-prompted the AI or asked for a new angle, and voilà, we had a fresh approach within minutes.

So many prototypes, so fast

We normally finish a 5-day sprint with one kinda janky prototype (usually only in Figma). This time, in just two days, we had multiple prototypes running with real code and tested three of them with 5 real users each. The throughput of learnings was completely bonkers.

3. Challenges We Ran Into

Teams get too big real quick

AI is great if there are 1–2 people prompting it at a time, but more than that, and you end up with chaos (conflicting prompts, overwriting changes, etc.). Our solution: limit real-time AI collaboration to no more than 2 people. You can still collaborate and discuss in a larger group during the show-and-tell sessions.

Design Consistency? Nope.

We ended up with prototypes all over the style spectrum. In future sprints, we’re going to try to integrate our design system (or at least brand tokens) so AI outputs are at least somewhat aligned. In the meantime, if you have an existing style guide, try feeding it to the AI. If not, consider at least specifying color hex codes and fonts you’d like used if design consistency is important to your prototypes.

Endless Feature Creep

AI can be a bit too creative, cramming in features we really didn’t need. And it’s a little difficult to prune without messing up other features in the process. Don’t be shy about telling the AI to rip out features that ****bloat what you really set out to test. Example: If the AI adds a login feature, but you don’t need it, say: “Remove the login flow and any references to user authentication, and keep the rest intact.”

“Merging” prototypes is not really possible (yet)

If you have two great prototypes and want to merge the best of both, it would be awesome to just tell the AI to do that (and have it work). Not there yet, so you basically have to start over and use the learnings from the other prototypes or just fork one of them. If wish you could just say “Take the login screen from prototype A, and the dashboard from prototype B”. (Or maybe you can and we just haven’t figure it out yet?)

4. The Sprint Schedule We’ll Try Next Time

Day 1: Problem Definition, Framing, and Target User

Keep this the same we did above. The Day 1 from the Coda template is really great.

Day 2: AI-Powered Prototyping & Iteration (and User Testing!)

I’ll change Day 2 the next time we do this to get real feedback earlier.

?? Day 2: Morning next time (this is mostly the same) ??

• Each person individual builds a prototype using AI tools (we used v0.dev, Bolt.new, and some Figma to feed in existing designs)
• After about an hour, do a Show & Tell of what they created, and vote on which elements of each have the most promise.
• This is a change ?Pick 2 or 3 prototypes that we want to test TONIGHT

?? Day 2: Afternoon next time (This is a change) ??

• Write user testing scripts.
• Polish prototypes to match the scripts.
• Launch the tests so you have some results in the morning. If you launch tests by 3 p.m., you can often get 5 testers by the next morning. It just depends on how specialized of a user you need. If you need someone really specific, you might try to schedule them ahead of time.

??? Day 3: Tools you can use ???

• v0.dev, Bolt.new (Lovable and Replit likely would have also worked)
• Figma to push existing design components into the AI tools (if they existed).
• We used UserBrain.com for quick user feedback sessions (but UserTesting.com, Maze.co, and Lookback.com all likely would have worked)

Day 3: Analyze Feedback And Refine

(This day is completely different because you should have some real feedback by now) We didn’t have any real feedback until this end of this day when we did our sprint.

?? Day 3: Morning next time ??

• Watch user recordings
• Share insights with each other
• Determine what needs to be in an MVP of your actual product
• You can bucket features into categories like “Expected, Nice-to-have, Delighters, Etc.” or create a simple matrix of “Must-have” vs. “Can Wait” to help with decision making.

?? Day 3: Afternoon next time ??

• Revamp the prototypes based on the insights
• Determine some must haves for your real MVP (and things you can leave out)
• Prepare a shareout

5. Where to Start

The barrier to entry to getting started with these tools is fortunately extremely low.

• Just try a tool like v0.dev or bolt.new by yourself for free. All it takes is a single prompt like “build me a shopping list app” and you’ll understand how it works. You might want to play around with it for a bit to get comfortable.
• You’ll want a paid account before you start your sprint. You are going to be doing a lot of prompting.
• Practice deploying and sharing your app. Deploying is a single click with most of these tools, but it is good to understand how it works before you start a sprint.
• You might want a team account, but not necessary. Having a team account makes it slightly easier to work on prototypes together, but not required.

6. Tips for Your Own AI-Powered Sprint

• Limit the Prompting Crowd. Keep it to pairs or a single “prompt pilot.” Three’s a crowd.
• Provide a Style Reference. Give the AI a sample design (can be an image, Figma file, or even just HTML) or snippet of your UI library. It can help it keep things coherent. You can also just try prompts like “Use #0070F3 as the primary color and #FF4081 as the accent color.”
• Refine Prompts in Other Tools and Keep a “Cheat Sheet”. Document which prompts work best for animations, integrations, etc. You can have other tools like ChatGPT o1 refine and expand you initial prompt ideas. It saves a ton of time later. Ideally put these somewhere where the whole team has access like a Google Sheet or your Coda/Notion hub.
• Time-Box Divergence & Convergence. AI can produce an infinite stream of ideas. Decide when to explore freely and when to narrow down. We explore widely for about 20 minutes and spend the rest of the hour refining and iterating.
• Don’t Test Everything. If a prototype is too basic or too off-track, skip user testing unless you think it might reveal something truly new.

7. What’s Next

This experiment really changed how I think about product development, but it also raised some new questions that these tools are going to have to solve:

• How do we link AI-generated prototypes with our existing design system? It will be awesome when these prototyping can generate design and code using our own components. We have ideas here about how to potentially make this happen, we’ll share them if we get something compelling.
• What is the make-up of an empowered product team when AI is participating in such an integral way? This is an on-going debate about how teams will be structured in the future. It seems like the traditional team of a PM, Designer, Engineering Lead, and 3-7 engineers is going to change with AI as a “teammate”.
• How do we balance AI-driven exploration with real deadlines and shipping actual products? This Design Sprint is intended to be exploratory, it is far from delivering shippable products. So until AI gets to that point, we need to find easy ways to take these prototypes and smoothly transition them to into the normal development process.

Give It a Try & Let Me Know ??

This is all still very new, so I’d love to hear if any of this works for you:

• Try an AI-powered design sprint with your team.
• See how quickly you can spin up prototypes and get real user feedback. Can you get feedback by the start of Day 3?
• Share your results or feedback so we can make a Version 0.2.

Thanks for reading, and here’s to figuring this all out together! ??