The startup anti-pattern
Back when I was young and naive, like in 2024, I assumed that everything I’d learned a founder would apply to the current moment. After all, we did so much good early AI at Textio. While most software was stuck on workflows, Textio promised outcomes! Our customer success engineers were essentially FDEs! We built generative writing features in 2019!
But while we broke ground in many respects, in most ways, we were SaaS 101. I literally ran a bootcamp for Textios that was called “SaaS 101.” 🤣
In all seriousness, we pretty much followed the B2B playbook. Started with an MVP and expanded from there. Used content to get attention. Had a free trial, but mostly sold to enterprise executives. Promised quantitative outcomes, but stopped short of pricing our product based on those outcomes. All the metrics we cared about were SaaS metrics.
Over the last couple of years, I have looked at 1000+ AI-native companies and spent time with a couple hundred. Diving back into AI product development myself recently, I am increasingly struck by how much of what worked for us in the SaaS era is an anti-pattern now.
Companies don't buy software anymore
Back in the day, companies bought software to solve workflow problems. If you were trying to organize a bunch of customer conversations, you used a CRM where you could store all your engagement with all your customers. To help you close the books every month, you used expense reporting software where employees could upload their charges for reimbursement. If you had an operational problem, you found a piece of software to solve it.
Companies do still buy some software. But increasingly, they buy two things instead of traditional software: Tokens and outcomes.
If you want to use someone else's AI model to build your own tools, you're going to pay by the token. If you use the model a lot, you going to pay a lot. Tokens are the new unit of consumption, closer to electricity than software licenses. Rather than buying a workflow, you are buying the right to generate your own workflows on demand. The meter is always running in the background.
Most business leaders aren't building their own AI tools, though; just like before, they buy tech to solve operational problems. But instead of buying traditional tools, these days they're paying for end results, like "Cut your sales cycle in half" or "Respond to customers within 5 minutes."
By the way, outcomes are what executives wanted all along. In the past, they had to settle for tools as the best available proxy. Now they expect AI to close the deal, resolve the ticket, or flag the risk without needing a human to stitch the workflow together.
If you're bringing a B2B product to market today and you want to stay viable, you're charging for outcomes or tokens. People pay for results. And wow, do they pay; even if the unit economics are questionable, they pay.
MVPs are no longer very M
Conventional startup wisdom is that you never overbuild. You build the Minimum Viable Product: build only what you need to be able to validate your idea. As you add new functionality, you add as little as possible until you prove product-market fit for each piece.
That made sense in a pre-AI world. But these days, building is no longer the bottleneck. You can create something that delivers end-to-end outcomes much faster than before. Since customers are mainly buying from you based on those outcomes (see above!), they expect things that deliver out of the gate.
MVP thinking is still useful. But now, minimum viability is purely about the scenarios you light up and not about the features you build.
Consumer is the new B2B
In the old world, decisions about buying business software were mostly made in top-down, centralized ways. Yes, some products offered freemium options and credit card billing. But for business-critical software, companies generally bought site licenses from the top.
But now, it is the standard for AI products to offer free or cheap options to consumers. Many people side-load powerful AI capabilities to their personal work stack. Token-based pricing means that AI companies never need to sell site licenses to make a whole lot of money from business products. The more people use AI products, the more they pay as individuals. It adds up fast.
The bottom line: The structures from five years ago are gone and they aren't coming back. That said, AI GTM hasn't settled into durable patterns yet. We still have a lot to figure out. It is a learner's moment.
If you grew up professionally in high-growth SaaS companies, the most useful thing you learned is still the most important: learning how to learn.
Kieran
If you liked this story, why not get the back issues?
My latest data stories | Build like a founder | nerdprocessor.com