A useful AI prototype test should include the part that doesn't go to plan.
Try this with a signup flow:
1. Enter an invalid email.
2. Submit twice.
3. Lose the connection.
4. Come back later.
Test the recovery, not just the happy path. That's the design work.
8
Day 28/100 of becoming an AI-native designer
AI made it free to look up one more thing before you decide. That makes tolerance for uncertainty worth more than it's ever been.
"The quality of your life is in direct correlation to your tolerance to uncertainty." It's an old line, usually credited to Tony Robbins. I read a passage this week that explains why it holds. What I took from it:
1. Uncertainty isn't the problem. Resisting it is.
The passage's claim: your resistance to not knowing, not the not knowing itself, is what freezes you.
2. You can't think your way into comfort with it.
You have to feel your way through. More analysis just feeds the loop.
3. Avoidance hides in small habits.
Checking your phone over and over. Over-preparing for a minor conversation. Asking someone to reassure you.
4. Start with small doses.
Have a conversation without planning your next line. Start a project before you feel ready. Take a new route without checking the map.
5. Every rep is evidence.
Each time you act while unsure and nothing terrible happens, you collect proof against your brain's worst-case predictions.
6. Lots of traditions found the same thing.
The Stoics called it voluntary discomfort. Therapists use exposure. Founders call it getting comfortable being uncomfortable.
7. It compounds.
After hundreds of "I felt unsure, I kept going, I was fine," your brain files uncertainty under normal instead of threat.
8. AI raises the stakes.
There's always one more question to ask a model, one more analysis, one more draft. If the need to feel certain runs you, AI will feed it forever.
One honest note: the passage reaches for big claims, quantum physics and "every wisdom tradition." The practice holds up without them.
How I'm practicing it is in the reply 👇
1
39
How I'd practice tolerating uncertainty in the AI era:
1. Catch your certainty loops.
For one day, note every time you check, re-plan or look for reassurance. Most people undercount by a lot.
2. Cap the research before you start.
Pick the limit first: three sources, or 20 minutes. Then decide.
Try: "Give me the 3 facts that would most change this decision. Then stop. I'll decide with what I have."
3. Decide in odds, not certainty.
When Ethan Mollick had Fable 5.2 pitch a business, it gave its own plan a 20-40% chance and still laid out the launch.
Try: "Before I decide, ask me what odds I'd put on this working and what I'd do if it fails."
4. Ask for the worst case, not comfort.
Try: "Don't reassure me. What's the realistic worst case, how likely is it, and what would I do the day after?"
5. Take one small dose a day.
An unscripted conversation. Shipping something before it feels ready. A walk with no destination.
6. Keep an evidence log.
Three columns: what I was unsure about, what I did anyway, what happened.
Try once a week: "Here's my log. Where did my predictions turn out worse than reality?"
7. Know what this isn't.
It isn't ignoring real risk. Decisions that are hard to undo, or that land on other people, still deserve the extra check.
(And if anxiety is running your life, that's a conversation for a therapist, not a tweet.)
Uncertainty is the normal operating condition. Not the emergency.
15
A smaller first build isn't a less ambitious product.
It's a cheaper way to find out which ambition survives contact with a user.
Ask AI for the smallest thing that tests the hardest assumption, not the smallest version of every feature.
1
1
19
John Ma | AI-native designer retweeted
Huge AI announcements from OpenAI, Anthropic, Meta, and Google in the past 48 hours.
Here's everything going on in AI right now:
1. The authors of AI 2027 just moved their timelines up. They now expect AI on par with the best human AI researchers to arrive in 2027, after METR estimated Anthropic is improving about 2x as fast thanks to AI. Daniel Kokotajlo gives it a 50-50 chance by May 2027 and says he's "scared, sad, and angry."
2. Anthropic opened the investor pitch for its IPO, with shares likely trading as soon as Thursday next week. An updated filing also says it has now crossed its own safety line for AI-driven progress.
3. OpenAI launched Businesses in ChatGPT, a beta hub where brands can set up AI agents for shoppers, claim a verified profile, and track how often ChatGPT mentions them. Paid ads can point users to those agents too. This is like the start of SEO for chatbots.
4. Anthropic rolled out Claude Sheets and Whiteboards in beta, made Claude Data generally available, and announced Claude Meetings, an assistant that sits in on your video calls. Add those to Slides, Design, and Docs. Seems like Claude is slowly rebuilding the entire office suite.
5. Anthropic also released Claude Fable 5.2 and Mythos 5.2 yesterday, with Fable debuting at #1 on Artificial Analysis' Intelligence Index, 5 points ahead of Google's Gemini 4 Carbon. It's also 20% cheaper and up to 30% faster than 5.1. Google lost the top spot before Carbon even shipped (!)
6. Meta released Muse Ember, its smallest model, built for high-volume agent work. It ties Claude Haiku 5.5 on the Intelligence Index at roughly one-third of the cost per task ($0.14 vs. $0.41 on an average test). Meta shipped its first Muse model in April and now has an answer for every tier OpenAI and Anthropic sell.
7. Google announced 2 new AI chips coming next year: the TPU 8tx, which keeps data encrypted even while a model is working on it, and the 8i Max, which joins 2 chips into one for agents that need to hold an entire codebase in view. Google already has nearly 100 TPU customers (including Anthropic and Apple) and every available chip is spoken for.
8. Nvidia's top lawyer reportedly emailed agents in August 2025 while they were investigating whether Singapore's Megaspeed smuggled its chips to China. The Commerce Department then halted the case and handed it to the Justice Department. Not a great week for Nvidia.
9. Adobe signed a multi-year deal with Disney to build custom AI models trained on Disney's characters and visual styles, with work already underway for theme parks, film, and ESPN. Pretty interesting to see one of the most protective IP owners in entertainment is now training its own AI models.
10. OpenAI acquired Odessia, a months-old AI trip-planning startup from Sonder founder Francis Davidson, who is joining OpenAI to work on travel. Between this and #3, ChatGPT is coming for shopping and travel.
11. France unveiled the Paris Call for AI Safety and Security, with OpenAI, Anthropic, and Mistral committing to build emergency off switches into their most advanced models. Interesting timing given #1 on this list.
Daily reminder that this is the worst AI will ever be.
John Ma | AI-native designer retweeted
If you design with AI, save this
1. modulify.ai — best vibecoding tool
2. landdding.com — top notch landings
3. modulify.ai/templates — free templates
4. motionin.design — 1 prompt 1 website
5. iconly.design — world best icons
6. fontinlogo.com — fonts behind brands
7. on.design — design community
8. godly.website — award websites
9. 60fps.design — mobile UI animations
10. footer.design — footer inspiration
11. recent.design — next level design
12. backgrounds.supply — backgrounds
13. posts.design — big brands posts
14. beui.dev — animated components
15. wonderlist.design — figma designs
16. mobbin.com — real app screens
17. sent.dm — almost free sms
20. paperanimation.ai — paper anim
18. logotouse.com — premium logos
bookmark this
The hardest part of a repeating AI workflow isn't run one.
It's run two: knowing what already happened, what changed, and what must not happen twice.
That's why Loop Engineering needs a state file, not just a better prompt.
A design-research example ↓
2
1
20
5/ Keep interpretation separate from bookkeeping.
The agent can assemble evidence. A person decides whether the theme is meaningful.
Try this on one recurring task: if you ran it twice tonight, what would duplicate?
Fix that before adding a schedule.
1
3
Day 27/100 of becoming an AI-native designer
Tell an AI the work belongs to someone else and it acts like a contractor. Tell it the work is its own and it acts like an owner.
Quick observations:
1. One sentence flips the behavior.
"This codebase is yours: what it's for and how it's built are your call, and you'll keep building on it."
With it, the model merged repeated code in about half of tries. Without it, under a tenth.
2. Without it, the model plays guest.
It adds to the existing structure, even when that means duplicating code, and reports nearby problems instead of fixing them.
3. The feature worked either way.
So your tests won't catch the difference. It shows up in the structure, months later.
4. Ownership costs something.
About 1.5x the output tokens. Unrequested tests and docs. Existing behavior changed in one try out of every six to ten. And its closing summary often skipped the small changes.
5. "No drive-by refactors" doesn't switch it off.
With that rule added, it still merged repeated code in about a third of tries.
6. Caution piles up the same way.
One of twelve notes in memory said never to retry. With it there, the model retried a busy tool in about four of ten tries. Without it, nearly every time.
7. Approvals get read literally.
A rule requiring approval for "sending messages to people outside this conversation" made it hold back about half its routine messages to colleagues.
8. Vague permission does nothing.
"Act more freely" changed nothing. Naming retries as a step that needs no approval took them from almost none to about half.
Yesterday's post was about AI that adds and never cuts. I wonder how much of that is a model that thinks it's a guest in someone else's house.
What this means for designers is in the reply 👇
When Anthropic release a new model it's always worth taking a look at their updated prompting guide - here's an entertaining tip to make Fable 5.2 less likely to be cautious in how it improves existing code: tell it "This codebase is yours" platform.claude.com/docs/en/…
ALT Say who owns the code Claude Fable 5.2 responds closely to what the prompt says about who owns the code it works on. In Anthropic’s testing on small coding tasks, when a prompt said nothing about ownership or presented the code base as someone else’s, the model tended to add to the existing structure, even when that repeated existing code. It also tended to report nearby problems instead of fixing them. When the prompt says the code base is the model’s own, the model often merges repeated code and makes related fixes without being asked. If you want the model to improve the structure as it works, say that the code is its own: This codebase is yours: what it's for and how it's built are your call, and you'll keep building on it. In Anthropic’s testing on small coding tasks, the model merged repeated code in about half of the tries with this sentence, and in under a tenth of the tries without it. The rate varied by task. The result held when the sentence stood inside a long system prompt
1
1
92
How to use this when you prompt as a designer:
1. Your design system has the same problem.
An agent that thinks the library belongs to someone else adds another button variant instead of merging the three you already have.
Try: "This design system is yours: what it's for and how it's built are your call, and you'll keep building on it."
2. Pick owner or contractor on purpose.
Owner for your own prototype or a sandbox. Contractor for a shared library other teams depend on.
Leave the sentence out when you want the smallest possible change.
3. Review the diff, not the summary.
Its summaries skipped small changes. For UI, that means before and after screenshots of every screen it touched.
Try: "List every file and screen you changed, including small ones, with one line on why."
4. Prune old cautions.
Rules files fill up with "never" lines written for older models. One stale note cut retries from nearly every time to four in ten.
My team's design brain reads its RULES.md on every run, so this applies to every line in it.
Try: "List every rule in this file that says never or always. For each, give the reason and whether it still holds."
5. Write the exception into the rule.
Under a flat "never" rule, the model took an approved step about one try in ten. When the rule said "get a go-ahead first," about nine in ten.
"Never change tokens" = "Ask before changing tokens."
6. Approve exactly.
Not "go." Say "Yes, update the button component and its three variants."
7. Name what needs no approval.
Try: "You don't need to ask before fixing contrast, spacing tokens or typos."
8. Ask for reports a busy PM can read.
Anthropic's own guide tells the model to put key points first and practice progressive disclosure. Designers have been saying that about interfaces for years.
Ownership is a scope setting. Not a feeling
1
29
I write about AI workflows for designers here: johnmaartifacts.substack.com…
3
John Ma | AI-native designer retweeted
STATE OF THE TECH JOB MARKET IN 2026
Twice a year, I dig into trends in hiring and job openings. This is the fifth report I've done, and the most surprising yet:
1. PMs are winning in the AI era
2. Engineering demand has stalled
3. Design openings are at a record low
4. Every role now requires AI skills
5. Tech jobs are concentrating in the Bay Area
6. Remote roles are at a record low
7. The tech job market is growing fast, but mostly outside of software (gasp!)
Full report here: lennysnewsletter.com/p/state…
