Should I build AI software or sell AI-powered services?

On a day when my own products were frustrating me, I said it out loud: software is just not the answer, at least not right now. Marketing is perfect, because you're still using the tech.

I spent a year building AI software, so I say this from experience: building software you can sell to other people is crazy difficult, and I don't care what anybody else tells you.

"figure out the aspects of AI that you can do that are not as complex as software, like marketing. Get really good at that on the cutting edge and then use that as a service instead of selling a product."

Why do services win for now?

• Less complexity. AI-driven marketing still uses the technology, without the bug fixing and support load of a product.
• Demand rises with the number of companies. If there are going to be 50 times as many companies, the skill of getting them attention is going to be 50 times more valuable.
• No technology ceiling. Back in late 2024 I told two salespeople why I was excited about a newsletter agency over my software: there is no technological problem that would stop us from getting to a million dollars. It comes down to effort and sales.

That doesn't mean swinging to the other extreme. Getting frustrated with AI and going to buy a plumbing company isn't the answer either.

What's the catch with running an agency?

Capacity. If you skip building and just automate the back end with Make or Zapier, every single client still takes you a bunch of time. Five clients is great, you're making money. Then you don't have time for anything else.

I've hit this myself. A very successful founder told me to stop doing digital marketing for clients because it was distracting me from the bigger business. I kept selling it, but decided to hand the delivery off to someone else instead of doing the work myself.

What's hard about building software?

Building your own software means constant bug fixing for the clients who already pay you, and that crowds out selling. I've lived it. More in why founders get stuck fixing bugs instead of selling.

A small market makes it worse. I shut down my earlier insurance software after putting a lot of money and effort into it. We started it pre-AI, aimed at a small market, which made sense at the time. Everything that made sense then makes the opposite of sense now.

Software is very complex and hard to get right. A service business mostly comes down to effort and sales.

When does building make sense?

When you niche down into a space where you have a real advantage. My formula: you know the space, you know the problems firsthand, you build the fix, and you already have an audience to sell it to.

That's why I built newsletter software. After two years of running newsletters, I knew exactly what the problems were, and I already had an audience of newsletter people. When I showed it to a prospect who owns a bunch of newspapers, he said it was magic and told me to name my price. Build broad tools for everybody and you'll drown in competition.

• The problem is narrow and you know it from the inside. See niche versions of old software.
• The existing tools are old and overbuilt. One local-media platform I demoed started at $750 a month, another at $80 a user with a three-user minimum. I demoed one of them and wouldn't use it if it was free, because it's so legacy and complex.
• It's easy to build, or the market is bigger than a sliver.

Should I build my own tools or pay for them?

Vibe coding is real and it makes sense for super simple tools that do one thing. I canceled Todoist and built my own to-do app. Mine is better, by the way, because it has AI in it.

Anything with nine connected steps is different. Pay an expert. I'd guess about 90% of business owners would rather pay $99 for a working tool than build one and keep fixing it. That's a market, if you can reach it.

Can I do both?

Yes, and this is where I've landed most often. A few ways to blend them:

• Managed service first, product later. That's how I approached my form system. I wanted to sell it as software eventually, but preferred a managed service in the short term because the UI might still have issues. I also put off self-serve checkout for a media kit builder because it wasn't worth prioritizing yet.
• A productized service. A simple app that you customize on the back end for each client, where every customer is a direct relationship you talk to. You skip the big self-serve SaaS headaches.
• Build the business and the software. Y Combinator has been pushing full-stack AI companies: instead of selling AI software to law firms, build the software and start a law firm. My version is AI software plus actually running local media brands.
• Build for yourself first. I build what my own newsletters need, like an onboarding flow, as quick standalone tools. It helps me directly, it becomes content and proof of work for the agency, and agency profit later pays to fold it into lightbreak.

A good demo matters more than you'd think. When a friend in the industry asked why he couldn't just build his own version of a niche app, my plan was a demo that looked valuable by the next day, even if it didn't work perfectly yet. People need to see the screens.

Should I sell the software or keep it internal?

I've gone back and forth. Most competing newsletter tools are $250 to $500 a month minimum, and I think ours is better because it was built AI-first in about six months, as of early 2026. One option is to never sell it and use it as the internal engine of a done-for-you agency. The downside: software can get a much higher multiple if you ever exit.

For most people, consulting or implementation is the highest-probability way to make money from AI right now. Start there, and see why that consulting should be niche.