If AI Can Build Any App, Why Is OpenAI Selling It to You?

If AI can build almost any app, why is OpenAI selling it to you instead of building the apps itself? That question took over Hacker News last week. The answers explain where your edge is now that everyone has the same models.

What happened

On October 22, 2026, someone posted an Ask HN titled "If the models are so good, why doesn't OpenAI just start 100 companies?" It pulled in hundreds of comments. When I checked it this week it had 671 points.

The most-replied answer came from a user named epsteingpt, and it's the one I keep thinking about:

"The sort of obvious answer is the same reason McKinsey hasn't taken over the Fortune 500 despite consulting and having deep knowledge of the operations of most of those companies. AI is brains. Companies need arms and legs and relationships."

The second answer that kept coming up was the gold rush. In a gold rush, you sell shovels. One commenter, adelekeabdulgan, spelled out why that works for a model lab: "whether you succeed in building the 'next big thing' with [their] tokens or not, they get paid." Others pushed back, arguing the real shovel sellers are the chip makers, or that the cliche skips over how much these companies spend.

And a few commenters pointed out that OpenAI is in fact moving into applications. A user named CoolestBeans opened with "They are starting to do it," pointing to OpenAI's growing product lineup. So the honest read is that the labs are doing some of both.

The comment that matters most for builders was a short one from a user named atraac: "unless you have a very specific niche, creating the product is no longer an issue, finding clients who are willing to pay is."

Why it matters for builders

Put those answers together and you get a clear picture. Building got cheap. You can ship a working app over a weekend with an AI coding agent. That's the same for you, for me, and for everyone else with a subscription.

What the model can't hand you is the part epsteingpt called arms, legs and relationships. Knowing a specific industry's daily headaches. Having the phone numbers of people who trust you. Being the person they'd actually pay. None of that ships in a model update.

So the bottleneck moved. It used to be "can I build this?" Now it's "who will buy this from me, and why me?"

Try it in 15 minutes

Here's an exercise I'd do before writing any code.

1. Write down three groups of people who already know and trust you. Your old industry, your hobby community, your local business network, your clients from a past job.
2. For each group, list the five most annoying recurring tasks you've heard them complain about. Use their words.
3. Circle the one task where you know at least three people by name who have the problem.
4. Write a two-sentence description of a small tool that fixes it.
5. Send that description to those three people today and ask one question: "Would you pay for this if it worked?"
6. Only start building when at least one says yes and tells you what they'd pay.

That's fifteen minutes of writing and a few messages. It saves you from building something nobody asked for.

Here's how you get paid

My opinion: building the product got easy, and finding customers who'll pay is the hard part now. So flip the order.

Pick a niche where people already trust you. Find out what they need. Build it with AI, fast, and sell it to them directly. Your first version can be rough. What matters is that you're solving a problem you heard from a real person, and you can call them when it breaks.

This is the one advantage the model labs can't sell you. Everyone gets the same models. Your relationships are yours alone, and they compound. Every customer you serve well in a niche makes the next one easier to win.