Right now almost every founder gets the same question from investors, customers and their own team: "Where's the AI?" So AI features end up in MVPs that don't need them, adding cost and complexity to a product that hasn't yet proved anyone wants it.
Sometimes AI absolutely belongs in version one. Often it doesn't. Here's how I help founders decide.
The one-question test
An MVP exists to prove one thing: that the core workflow is valuable enough for someone to use it, and ideally pay for it. (I wrote more about that in how to build a SaaS MVP without overengineering it.)
So the test is simple:
Does the core workflow work without AI?
- If the answer is no, AI is the product. A tool that turns receipts into categorized expenses, a data analyst that writes reports, a voiceover generator: remove the model and there's nothing left. AI goes in version one, and most of the MVP's engineering effort goes into making it reliable.
- If the answer is yes, AI is an enhancement. Ship the workflow first, learn how people use it, then add AI where it removes a real, observed pain.
Most products are in the second group, even when the pitch deck says otherwise.
When AI belongs in version one
AI earns its place in an MVP when:
- It's the reason the product exists. The value proposition is literally something a model does.
- The manual alternative is the problem you're solving. If customers today spend hours reading documents or entering data, the AI that removes that work is the product.
- You can measure "good enough". You know what a correct output looks like and can test the model against real examples before launch.
In these cases, don't treat AI as a feature to wire up at the end. Prototype the core AI loop first, on real data, before building anything around it. If the model can't do the job well enough, you want to find that out in week two, not week ten.
When AI should wait
Hold it back from version one when:
- It's a "nice to have" layer, like an AI summary of a dashboard that already shows the numbers.
- You don't know yet what users will ask for. A chat assistant over your product is far better designed once you've seen real usage.
- It adds a cost you can't price yet. Every model call costs money per use. Before you know your usage patterns and pricing, an always-on AI feature can quietly eat your margin.
Waiting isn't a weakness. A clean, fast core product with AI added in month three usually beats a slower, buggier one that had AI on day one.
If AI goes in, build it properly from the start
Even a small AI feature needs a few things that are painful to retrofit later:
- A dedicated backend module for model calls, with keys never exposed to the browser.
- Per-user usage tracking, so you understand cost before it surprises you.
- Timeouts, retries and a fallback, so a failed model call shows a sensible message instead of an error page.
- Background jobs for anything slow, instead of making users wait on one long request.
This is a few days of work in an MVP, not weeks, and it's the difference between an AI feature you can grow and one you'll have to rewrite. The details are in how to integrate OpenAI into a SaaS product properly.
Design for AI later, even if you skip it now
If AI waits, you can still make it cheap to add. Keep your data clean and structured, store the events and documents a future model would need, and keep business logic in a backend you control. Then adding AI in month three is a feature, not a rebuild.
The short version
If AI is the product, put it in version one and make it the first thing you prove. If AI is an enhancement, ship the workflow, watch real users, and add AI where it clearly saves them time.
Not sure which group your product is in? That's exactly the kind of question I sort out in a first call. Here's how I approach SaaS MVP development and AI product development.
