If you were starting from scratch, what AI business would you build today?
AI is moving fast and there still seem to be openings for people willing to learn and build something useful. Starting today, which would you pick for the next two or three years: building agents for businesses, automating repetitive tasks, a niche AI SaaS product, AI services for small businesses, or something else?
Pick the customer before the model. The options on that list are really one question: whose expensive, repetitive problem do you understand well? A service for one industry you know beats a general AI tool, because you can find the buyers and you know what a good result looks like.
Services first is the lower-risk path. Doing the work for a few paying clients, with AI doing much of it behind the scenes, shows you which steps repeat. The parts you do the same way for every client are what is worth turning into a product later.
Check demand with conversations, not a landing page. Ask people how they handle the problem today, what it costs them, and what they already pay for. If nobody has tried to solve it, it may not hurt enough to pay for.
Be careful with thin wrappers. A product that is mostly one prompt on top of a model can be copied quickly, or absorbed when the model makers ship the same feature. Your own data, workflow integrations and relationships with customers last longer than the prompt.
Listings mentioned
- The Mom Test Framework · skill by wondelaiScores your customer-interview questions and rewrites the ones that invite polite but useless answers.
- Lean canvas · skill by phurynPuts problem, segment, channels, costs and revenue for an idea on one page so the weak spots show.
- Opportunity assessment · skill by wondelaiA structured check of whether an opportunity is worth committing time and money to.
Answers by the AgentAlley team, drafted with AI and checked against the listings they link to. Not a real-person reply from the original thread.