VCs Are In a Love-Hate Relationship With the Model Layer
Everything is up for grabs again! Or is it?
If you’re willing to take the bet with me that LLMs are the new OS, then it’s fair to assume that nearly 100% of our digital interactions will be reinvented or entirely replatformed over the next 5 to 10 years.
The exciting thing is that AI creates such massive efficiency unlocks - either by 100x-ing what software could previously do, or by entirely replacing labor - that every market suddenly has a newfound ‘Why Now?’. *Cue every VC analyst realizing their GP hasn’t asked for that slide in the diligence deck in a while.* What that means for tech investors like me is that every market is up for grabs again. That’s great news.
The other big consequence (spoiler: it’s not as great) is that the world has suddenly been split into three categories for us:
1/ The stuff model layer companies do
2/ The stuff that model layer companies don’t do yet (or the Mirage Arena)
3/ The stuff that model layer companies will never do (the Safe Arena)
Category 1 is hopefully straightforward. Cat 2 and 3 – much less so.
Category 2, what model layer companies (or MLC) don’t do yet, is what I call the Mirage Arena. And it is not as obvious as you’d think. It’d be easy to lump in video generation, code generation, or horizontal voice companies in here, for example. But Cursor ($500M ARR), ElevenLabs ($100M+ ARR), and Synthesia ($100M+ ARR) have proven that you can reach escape velocity with these standalone value propositions, even with MLC shipping updates every week. Does that mean it’ll be true for every other horizontal code/video/voice generation company? Certainly not. But these players have shown that it is possible to earn real user adoption alongside MLC, by channeling the models’ latest advances into very useful offerings.
There’s both a pessimistic and an optimistic read on where Cursor, ElevenLabs and Synthesia really sit:
Pessimistic: “Yet” is a tricky but beautiful temporal adverb, and “Don’t Do Yet” could mean one day or five years. Understand: it’s just a matter of time before they realize they’re doomed.
Optimistic: These companies will keep innovating at the edges, always going the extra mile, and that extra mile will keep earning them enough goodwill with users to stick around.
As an investor, it means waking up every morning and scanning X and tech news to see what the MLC have just released, and what it might break. That’s the love-hate relationship at play. You’re rooting for them to push the frontier, but dreading the day they cross into your territory.
Only time will tell if you were actually in the Mirage Arena all along; and just convinced yourself you were safe.
The Safe Arena: What VCs Should Invest in Now
That’s a clickbait title if I’ve ever seen one.
More seriously, investors are currently rushing to build a framework for what a defensible company in this category – the Safe Arena – should actually look like.
Some elements are more obvious than others:
The solution should require unique integrations into idiosyncratic systems (billing platforms, internal databases, EHRs, etc). MCPs may chip away at this over time, but for now it is still a bottleneck.
The product should include robust evaluation and observability layers: indeed, once the tech interacts with internal data, employees, and customers, it should benefit from ongoing tuning and monitoring.
The target buyer should have complex regulatory, security, or even on-prem needs, requiring a partner who can offer control, audit trails, and enterprise-grade support.
But the truth is, it’s still early days, and no one has a clear blueprint yet. Think about it, just 40 years ago, we were semiconductor investors. Then we had to learn how to be software investors. Then we had to learn SaaS. When SaaS came around, it took years to understand it and establish the benchmarks that eventually became industry standards (CAC, LTV, Rule of 40, MDR, GDR).
In reality, and as Ramy Adeeb told me in a recent conversation, “We’re going to have no idea how to assess these companies for the next four or five years”. As much as VCs love to emphasize how long they’ve been investing in AI (“I was investing in AI before it was cool!”), the reality is, we’re all learning how to be AI investors today.
Hooray! You’re in the Safe Arena! Now Don’t Get Knocked Out by The One-Two Punch
Let’s assume you’re a company, or investor, in the Safe Arena. You don’t know you are, but you are. Congrats. Now what you have to worry about are your market’s incumbents.
Take AI startup Glean, for example. Time will tell whether they belong in the Mirage Arena or the Safe Arena, but they’re a striking example of the one-two punch that plays out at the edge of those two categories:
First comes the Jab from MLC: every week brings a new “ChatGPT just killed Glean” article, the most recent one pointing to Deep Research’s new ability to search across private documents in Drive, Dropbox, and more with ‘data connectors’.
Then comes the Cross from Incumbents: As The Information recently reported, Salesforce unexpectedly blocked Glean from accessing Slack data, and Atlassian and Notion have been throttling them. A reminder that data access can vanish overnight when you're building on someone else's platform.
Indeed, most AI companies (though not all) need to draw from or push data into a customer’s incumbent system of record. And that system won’t go down without a fight.1
By now, it should be clear there’s nothing particularly safe about the Safe Arena.
That said, I do think we’re starting to see early signs of how AI companies can, over time and with enough work, displace their legacy system of record incumbents. It’s not a quick win. It requires patience and a strategy that slowly wraps around the incumbent with the goal of eventually replacing it. I call it the Strangler Fig Playbook. There’s a lot more to explore on that topic, and I plan to dig into it in a future post.
In the meantime, stay safe out there!
It might not go down at all if it manages to hire the right AI talent to build its own solution or acquihire a strong team before it’s too late.