Stop Building in Secret: Why Your AI Ideas Need Real Users Now

Maaria Tiensivu shares five practical steps to test your AI ideas with real users instead of tweaking them to death in isolation

We've all been there. You have this amazing AI idea that's going to change everything. You spend weeks tweaking prompts, months refining features, and before you know it, you're three models behind and someone else has already built what you were "perfecting."

Maaria Tiensivu, founder of KNOW-ME and Four Languages, came all the way from Mexico to ChippCon 2025 to share a hard truth that every AI builder needs to hear: your ideas are dying of loneliness while you're busy making them "perfect."

The Facebook Disaster That Teaches Us Everything

Remember Facebook's first "Year in Review" feature? The company expected users to share happy memories and celebrate their best moments. Instead, they accidentally created a nightmare by shoving people's most painful memories—deaths, breakups, tragedies—right into their feeds with cheerful, celebratory framing.

Good intentions, terrible outcome. And it could have been avoided with one simple thing: testing with real users instead of assuming what people wanted.

"The solution is very, very simple," Maaria explains. "It's literally just testing. So stop imagining, stop building, stop hypothetically thinking about things and actually go out and test."

The Five-Step Reality Check for AI Builders

Maaria breaks down testing into five practical steps that any builder can follow:

1. Decide Your Risk Budget Not every AI project carries the same risk. A tarot card reader app isn't the same as a medical platform handling sensitive data. Be honest about what you're willing to invest in time, money, and potential mistakes.

2. Identify Your Key Assumption Here's where most people mess up. Don't ask users "Would you like to see your year highlights?" because of course they'll say yes. Instead, show them actual highlights—including potentially painful ones—and see how they react. Show, don't ask.

3. Find Your Edge Cases Don't just test with people who had a great year and post positive content. Include people who went through tough times, posted challenging stuff. These edge cases will teach you more than a hundred happy users ever will.

4. Just Run It Use Chipp's chat history and tags to see what's actually happening in conversations. For early testing, keep it simple—don't get distracted by fancy animations or complex features that might hide the real issues.

5. Learn and Iterate Look at your results honestly. What worked? What didn't? What surprised you? Then make changes and test again.

The Real Cost of Assumptions vs. Experiments

"There's a cost to assuming things and there's a cost to testing," Maaria points out. "In my opinion, personally, assumptions are quite expensive usually. And experiments are cheap in comparison."

Think about it: How much time have you spent building something that nobody actually wanted? How many features have you added based on what you thought users needed rather than what they actually asked for?

Your Chat History Is Gold (If You Actually Look at It)

One of the biggest missed opportunities Maaria sees? Builders who don't look at their chat histories. "Chat history is very underused for what it should be," she says. "That's where the answers are. You will see if people are engaging, are they dropping off?"

It might be scary to see how people really use your AI, but that's exactly why you need to look. Those conversations contain the insights that will make your next version actually useful.

The Permission to Ship Imperfect Things

Here's what Maaria wants every builder to understand: people aren't upset when something doesn't work perfectly, as long as you're honest about it. "When you just give them the actual story behind it, people are not gonna be upset if something doesn't work. They're actually just gonna be really happy that they get to be a part of the process."

Your users want to help you build something great. But they can't help if you never show them anything.

Ready to Stop Overthinking?

Maaria's final advice is beautifully simple: "Just pick your budget, question people and go. And that's it."

Your AI idea doesn't need to be perfect. It needs to be real, in front of real people, solving real problems. The sooner you embrace that, the sooner you'll build something that actually matters.