Build It Wrong Before You Know It's Right
Use AI to test a hundred ideas a day as a failure machine, not to polish one idea for three months
- Difficulty
- Easy
- Time to result
- ~weeks to results
- Steps
- 4
- Confidence
- 90%
A testing philosophy that inverts how most teams use AI. Instead of using AI to build one polished idea over three months, build deliberately-wrong, low-cost versions whose only job is to generate signal — running more experiments in a week than your industry runs in a year. The goal of what you build is to learn, so it only needs to be good enough to test the specific idea, not to be right.
Origin
Mark Pincus's approach to product testing at Zynga, sharpened by the AI era. He warns that AI's speed makes it a 'dangerous drug' because it lets teams reach a viable product fast enough to fall in love with a single idea instead of testing many.
Core principles
- 01Build it completely wrong before you know it's the right product — the point of what you build is to learn
- 02AI should be a testing machine and a failure machine, not a way to lovingly build one idea in three months
- 03If you believe today's product is wrong, act on it: waste a day or a week, not three months
- 04Pull apart the pieces you can test cheaply — sometimes the test is just an ad
- 05Turn would-be marketing spend into a product-and-demand test on a surface you already own
How to run it
- 1
Assume the current product is wrong
Start from the belief that what you'd build today is the wrong product. That belief changes your behavior: you refuse to sink three months into it and instead build the lowest-cost version that can return signal.
Pro tip Ask 'what will I do differently because I believe this is wrong?' — the answer is build cheap and fast to learn.
- 2
Isolate the cheapest testable piece
Break the idea into components you can test independently and pick the cheapest one that yields signal. Often you can validate demand with just an ad long before building the product.
Pro tip Test your ads and messaging BEFORE building — teams routinely make ads an afterthought and lose the cheapest signal available.
Watch out Skipping ad/demand testing until after the product is built wastes your fastest, cheapest learning.
- 3
Run tests on a surface you already own
If you have an engaged audience, test new products and marketing on your existing surface rather than buying attention elsewhere. Put variants in front of live users and measure what draws heat.
Pro tip An existing user base clicks on anything new on their familiar surface — exploit that for near-free signal.
- 4
Scale the number of experiments, not the polish
Optimize for testing a hundred ideas a day rather than one idea in three months. Use AI/vibe-coding to spin up the lowest-cycle version that can return signal, then move on.
Pro tip Measure your process by experiment throughput per week versus your industry's per year.
Watch out AI's power tempts you to polish one idea; resist — its real value is volume of cheap failed tests.
In the wild
The team planned a $10M pre-launch ad budget on external channels despite having 25-30 million daily users. Pincus redirected them to place locked art variants of the expansion on the existing game board, testing look, feel, and message while offering early access via 'coming soon — click here to be a pre-user.'
→ Turned an advertising afterthought into product-and-marketing signal AND sold $19 million worth of early-access keys.
Common mistakes
Using AI to build one idea in three months
AI's speed lets you reach a viable product fast, which seduces teams into perfecting a single idea instead of testing many. This defeats the purpose — the leverage is in running a hundred cheap failing tests, not one expensive polished bet.
Treating ads and demand tests as an afterthought
Teams build the whole product first and only think about marketing at the end. The ad IS often the cheapest test of whether anyone wants the thing, so skipping it upfront throws away the fastest signal and risks building something no one will click.
Is it for you?
Best for
AI-era builders who can ship fast and want to convert that speed into learning velocity rather than one over-polished product
Not ideal for
Products where a rough or 'wrong' version would irreparably damage trust or safety, so cheap failed tests aren't acceptable
From the transcript
“I like say more ideas in a week than your industry tests in a year.”
“So build it wrong before you know it's right.”
“What if you just put something on the game board?”
“we ended up selling $19 million worth of keys to get early access to, you know, the new expansion pack.”
From the episode
The hidden pattern behind successful products
Mark Pincus (founder of Zynga)