Building an AI product comes with a painfully familiar problem: the product ships fast, but the first user is nearly impossible to find.
Code done. Site live. Features working. Open the dashboard: zero users.
So I started asking myself non-stop: shouldn't I build something else? Find another channel? Jump on another trend? For a while, a new idea popped up almost every day, and every one of them sounded reasonable.
Eventually I learned that the hardest part of a startup isn't deciding what to build — it's deciding what not to build. This week I turned down three opportunities, each of which looked "directionally correct." Saying no to all of them is exactly what helped me do one thing right.
A while ago someone floated an idea: plenty of indie developers ship products with no real users testing them. Why not build a test-and-try site where people put their products and real users experience them?
I asked ChatGPT whether similar channels existed. Turns out: most of them cost money.
That brought back a habit from my e-commerce days — the "traffic magnet." Its job is to pull in your target users. For us, small tools are traffic magnets, because tools have natural demand. But the key question isn't what tool to build; it's whether the people that tool attracts match your main business.
So we chose AI resume building and AI resume screening. The reason is simple: the people those tools reach are exactly the people we want to serve later.
But then a thought hit me: why couldn't a test-and-try site itself be a good tool? If most testing channels charge money, why not make a free one that helps indie developers find their first users? It even aligned perfectly with our direction — people using Claude and Codex are the new-generation developers of the AI era. If they need talent later, they'll have outsourcing needs too.
So I started rethinking. A direction being right doesn't mean it's right to do now.
Question one: can we actually gather a large community of indie developers right now? Clearly not — we don't have the brand influence. Question two: if anyone can publish products for free, where's our business model? No revenue means we'll have to commercialize eventually. And once users get used to free, charging becomes a massive barrier.
This is the story of the dragon-slayer becoming the dragon. We start by building a free platform to help people, and end up forced by commercial pressure to become a platform we dislike.
My conclusion: yes, we'll build it. But not now. In the future it fits better as added value to the main product — once the main product works, it can become part of the ecosystem and form a new value loop.
A right direction isn't the same as the right timing. Knowing when to postpone an opportunity is also a decision-making skill.
No. 2: The Only10 projects that "are worth doing"
Only10 has run for six editions now, and it has surfaced some genuinely good projects: the furniture database from the first edition, the ESP32 hardware project from the second, and more. But I've realized a lot of people may not know how Only10 is actually meant to be used.
Only10 and the GitHub trending list are two different things. Only10 was never about "what fun new thing can we all play with this week." It's about this: can people take one or two of these projects and actually turn them into something with revenue?
And no — you don't need to build all 10 projects each week. Their shared premise: keep costs under $10K, and make it replicable by a single person. That leaves room for indie developers. But when it's your turn to choose, the real question appears: it being worth doing doesn't mean it's worth YOU doing.
A project might work great in India and find no market in Australia. So what Only10 gives you is just a pool of opportunities. You still need to re-filter it against your location, resources, skills, industry experience, and channels.
So the next three projects aren't me telling you "these are all worth building." They're me showing you how I find my own opportunity inside one project.
From edition one, the furniture database looked odd. It wasn't pretty, and it ranked low. But it was the first project where I thought "when I have time, I might actually build this." Why? Because it has one obvious advantage: it's ugly. The creator is a furniture enthusiast without a strong commercial instinct, which means huge room for visual improvement and a moderate difficulty to replicate. More importantly, it doesn't attract vanity traffic — it attracts people with furniture needs and furniture lovers. These people either just spent money or are about to. That traffic is more precise than raw search traffic. So what I saw wasn't "let me copy a furniture database" but "if I rebuilt this, could I make it better?" That's the first mindset: optimization.
The ESP32 hardware project from edition two fit my location pretty well. ESP32 is cheap — in a way it's low-cost embodied intelligence. With AI companion apps being so hot, emotional companionship is an obvious use case. But it spans software and hardware. People are willing to use AI to write code, but not necessarily to learn hardware. If someone already spends a couple hundred yuan a month on APIs, would they buy a ~100-yuan device that gives their AI a physical body? Can an existing thing be combined with another growing industry? That's the second mindset: weighting.
Edition four had too many good projects. I had to pick one that was simple, low-cost, and had a clear monetization path. And this one, I think, shouldn't be copied directly — but its approach is worth stealing. It solves a very real problem: kids have too much screen time. Is there something that's fun for kids, cuts device time, and adds hands-on and social activity? So they made stickers. Can I not make stickers? Change the material — blocks, fuse beads; change the activity — drawing, singing. Take the path it provides, swap out a few steps, and you have a new project. The monetization path stays clean: a physical add-on product after interactive entertainment. So the lesson isn't "children's stickers"; it's that the project offers a path you can recombine. That's the third mindset: decomposition.
The GitHub trending list gives you answers. Only10 gives you thinking opportunities. What truly matters is never "which project did I copy this week?" but: "after seeing this project, did I come up with one of my own?"
No. 3: Building products that only I think are cool
I've seen so many people doing vibe coding — typing code happily, launching the product, and nobody shows up on launch day. Let's talk about something uncomfortable: why, in the AI era, are there so many self-pleasing products?
Here's an example from us. We build resume screening. Companies do this for a living, but there's almost no competition at the tool level — why? Because resumes are sensitive data. Big companies would rather deploy locally than hand candidate information to an external tool.
So our first hard rule is PII desensitization: the model never sees candidates' contact details. Compliance isn't a bonus here; it's the ticket to enter this market.
That taught me something: why would anyone pay for your product? Not because you think it's cool, but because users find it useful. Only with that sentence can you have revenue, and avoid being self-pleasing.
How do you avoid building self-pleasing products? Three moves.
One: build for the needs of your own industry. You are your own user. You naturally have the user mindset of that industry, and what you build carries other-serving value by default.
Two: ask people. Especially treasure the contrarians around you. Ask them: if I build this product, would you need it? And if you don't have anyone like that nearby, ask me — I have no stake in your project, and I'll tear your product apart without mercy.
Three: find similar products and read the reviews and complaints. Real users' bad reviews are the best demand research. No similar product? Then find the audience. Say you want to build a digital-human live-streaming tool with no direct competitor — go read the comments under live-streaming plugins. Wherever users are complaining, that's where the demand is.
The takeaway
This week I declined three opportunities: a platform that was directionally right but wrongly timed, a batch of projects worth building but not worth me building, and a product that made me feel good about myself.
They look unrelated, but they're the same judgment: refuse to let "I think" replace "users think"; refuse to let "it makes money now" replace "it holds up long-term"; refuse to let "the direction is right" replace "is the timing right?"
A right direction isn't the same as the right timing. Something being worth doing isn't the same as it being worth you doing. You thinking it's cool isn't the same as users finding it useful.
Subtraction is much harder than addition in a startup. But it's exactly the opportunities I refused that kept my attention on the things that were actually right.