Fifteen years in SaaS, and for the first time I would hesitate before hitting send on a renewal email.
SaaS isn’t dead. But the kind that made sense to build five years ago mostly is. Back then the thin layer between a customer and an API was worth $29 a month because building that layer was the hard part. Now a customer can rebuild it with a prompt in an afternoon, and they know it.
Andrej Karpathy put it more bluntly than I would have. He built a small app that turned a photo of a menu into images of every dish, then watched a single prompt do the same thing and make his version pointless. His test for anyone building today is one question: could one multimodal prompt replace what you’re building? If the honest answer is yes, you don’t have a product. You have a head start that’s already shrinking.
The model you wrapped can turn on you
Look at Jasper. Around $120M in revenue in 2023, down to roughly $55M the next year. A peak valuation near $1.5B, cut by about a fifth by the time the founders stepped down. No competitor took them out. The model they wrapped did. The thing that made the company possible is the same thing that made it disposable.
That’s the part founders underrate. When your product is a polite interface over someone else’s capability, your roadmap is set in a lab you don’t work at. They ship a feature on a Tuesday and your moat is gone by Thursday.
Once your product is one good prompt away from being copied, the subscription stops feeling like something you pay for and starts feeling like a tax.
People don’t churn loudly about it, either. There’s no angry email. They sit with the renewal notice a little longer than they did last year, decide they can probably do this themselves now, and quietly cancel. By the time it shows up in your numbers it’s already a trend, not a blip.
The floor, and then the goal
What’s left standing is the SaaS a customer can’t walk away from: years of their own data sitting inside it, a workflow the team has slowly bent itself around, integrations nobody wants to spend a quarter ripping out. That’s real, and it’s worth building toward.
But that’s the floor, not the goal. Lock-in buys you time. It doesn’t earn you love, and a customer who stays because leaving is annoying is a customer who leaves the moment it isn’t.
The product that actually wins does more than hold the data. It takes months of inputs and hands the customer something they would never get from a blank prompt: the pattern in their own numbers they had missed, the next move pointed out before they thought to ask for it. Past a certain point the software stops being a tool they open and starts running a piece of the business for them. That is a thing you cannot rebuild in an afternoon, because most of its value is the accumulated context, not the code.
I think about this with the apps my own teams have built inside the Shopify ecosystem. The ones that feel safe are never the ones with the cleverest feature. They’re the ones where a merchant’s whole operation has quietly grown around the thing.
A $29 wrapper has a shelf life, and the labs set the date. The one that survives won’t be whoever hoarded the most data. It’ll be the one a customer genuinely can’t picture working without.