Product-market fit in the AI age

It seems every week, an AI startup sets a new record for speed to revenue. Mercor (AI recruitment) and ElevenLabs (AI voice) hit annualised revenue of $75m and $100m in 24 months. Cursor (AI IDE) hit $100m in 21 months.

In the SaaS age, growth like this was far less common with the bulk of unicorns achieving $100m revenue in a 5-15 year timeframe, bar a few exceptions like Wiz and Deel (see Figure 1).

When we look at revenue growth, we are using a crude proxy for product-market fit (“PMF”). Factors like churn and growth efficiency are also important - First Round provides a great framework on the four levels of PMF and the metrics you would expect to see at each level (see Figure 2). 

The question is whether our understanding of PMF needs to change in the AI age. 

We believe the levels themselves still apply, but two things change: 

1. The speed at which startups move through these levels will be much faster than before. For example, Bolt, Lovable, HeyGen, Mercor and Cursor’s product-led growth (“PLG”) motion saw them racing through the levels of PMF in under two years (see Figure 3).
2. Movement back and forth between these levels will also increase. Startups can lose PMF for many reasons. For example, the AI copywriting startups of yesteryear saw scintillating growth before OpenAI’s products rendered their use case obsolete. We expect that it will be much more common to see startups gain and lose PMF.

These changes are equally exciting and daunting for founders. It’s never been a better time to build a business that can leapfrog competitors, including incumbents that seemed unassailable before the AI age. 

However, it’s also never been more true that “only the paranoid survive” (to quote Andy Grove). With thousands of AI-native teams chasing the same opportunities, someone out there might be working harder and faster than you. That’s the risk every team faces, but especially so in the AI age. The founders of Cursor articulated well why the loss of PMF is such a real threat in their Lex Fridman interview. Right now, the model layer and below are not stable - with every model release, new capabilities are unlocked, enabling new features, interfaces and form factors at the application layer. If you’re not on top of these changes, your product will feel significantly worse than your competitor who is bringing the latest capabilities to their users.

This begs the second question we’ve been asking ourselves: what type of founding team can not only find PMF but hold onto it in the AI age? 

We believe founding teams will need these three qualities:

1. They ship relentlessly. 

• HeyGen (AI video app) operate on a weekly release schedule vs. the fortnightly standard at other fast-growing companies
• Mercor (AI recruitment) run the whole company on a 9-9-6 schedule (9 am to 9 pm, 6 days a week)

2. They are paranoid about fake PMF and the loss of PMF.

• If your users aren’t sticking around, it’s fake PMF. Demo magic can only get you so far. Many startups in the AI SDR space are facing this issue - overselling, high churn, disappointed customers.
• If you’re running an AI-enabled services playbook, revenue growth can feel easy, but cost containment is much harder. If your costs are growing linearly with your revenue, it’s fake PMF, no matter how fast that revenue is growing. A founder running this strategy recently said to me, “Our success in selling services was slowing down our ability to automate”. Startups have finite resources, so founders will constantly be forced to choose between having their engineers serve the customer’s needs right now, even if it’s in a custom or unscalable way, or spending that time to automate and build product. Unfortunately, the pull of the former can be too strong‍

3. They are strategic about product design, business model and go-to-market motion - meeting their users where they’re at. 

• AI adoption will happen at different speeds across the economy. Enterprise will necessarily be slower, bogged down by compliance, cybersecurity and the sheer scale of the change management required. Glean targets enterprise customers where credibility and trust are of paramount importance. Their go-to-market motion for scaling from $10m to $40m and beyond was partnerships - partner with the likes of Google and Amazon and borrow their halo. 
• For startups running PLG motions, rethinking design is key - we are still in the early innings of AI enabling new interfaces. To date, there has been a lot of anchoring to the chat interface, but we’re starting to see this evolve. AI reliability isn’t solved yet, but that doesn’t mean users will reject it outright - apps like Bolt and Cursor find ways to make AI feel less ‘black box’ and allow users to make reversible choices. 

If you’re a founder building at the AI application layer and the above resonates with you, I’D love to hear from you. You can reach me at lucy@squarepeg.vc.