1Mby1M Virtual Accelerator AI Investor Forum: Marina and Nick Davidov, DVC (Part 4)

Sramana Mitra: I actually have a much more interesting question to explore with you. We talked about Perplexity as a wrapper. Give me examples of other interesting wrapper companies. I think these wrapper companies are very easy to bootstrap and very desirable for people to build revenue on top of quickly. So, give me more examples of what you’re seeing.

Nick Davidov: We have 125 portfolio companies, out of which 72 are agentic businesses — what you might call wrappers. We can name all of them, but the challenge is, it’s not easy to build them. It’s a different business from building foundational models from scratch.

Sramana Mitra: And I don’t think people need to build more foundational models. I think that’s done.

Nick Davidov: Yes, it’s like building a car and deciding to make your own oil. But with great niche products that aren’t expensive to kickstart — we just announced the round of Key, which is software for due diligence. Private equity funds do a lot of due diligence when they buy companies. They usually have a standard set of questions. The team created a beautifully designed agentic system, plus a bit of data room software. It took them about three months to get to $3M ARR. They went through Y Combinator and raised a $5M seed round.

Solve Intelligence is another one — an agentic solution for patent attorneys. Agents write patents from technical documentation.

FleetWorks uses agents to schedule freight dispatchers — it calls drivers, negotiates, gets them to take payloads from point A to B.

Avoca AI is another breakout success from our portfolio. It’s an AI salesperson/CRM for contractors like electricians and HVAC installers who don’t have time to talk to customers. The AI handles communication and scheduling.

Kick.co is an AI bookkeeper we’ve used for the last two years to file our own taxes. They’re growing incredibly well.

Sramana Mitra: Let me ask you a question. In the companies you’re seeing take off, can you bucket them into two categories. One is PLG — Product-Led Growth, like Lovable and Base 44, where growth is driven by user adoption. I suspect some of your pool of 72 companies including Perplexity are probably following that.

Then there is also Enterprise Applications. I suspect Private Equity pne is not exactly product-led growth. There’s a sales cycle involved.

Nick Davidov: There’s always an element of both, but yes in our fund it’s 50% is product-led growth. 50% is you take it and sell it. But a lot of times, once you start selling something that people love, word spreads. If you sell something to HR people, they’ll talk to each other. There is some element of product led-growth.  So even sales-driven products can benefit from network effects.

Sramana Mitra: There’s some element of word-of-mouth.

To the extent that you have adoption inside enterprises or businesses, what are you seeing? Are you seeing more “human in the loop” or “human replacement”?

Marina Davidov: We’ve seen both.

In terms of enterprises, you have to understand that you see a lot of growth in startups because enterprises are under pressure to implement AI. But it’s often done from pilot or experimental budgets — not from long-term budgets.

So while enterprises are using AI agents and tools, it might not be long-lasting. And that can mislead entrepreneurs who count pilot revenue as ARR and show it as growth.

Sramana Mitra: How does that affect your investment thesis?

Nick Davidov: We try to disregard pilot revenue if we don’t see how it will stick.

For example, a company might get $300K in revenue from Netflix. But for Netflix, that’s a rounding error — just a trial.

We look deeper into the product. Often, one of our LPs is a user of the product. That’s how we hear about startups. For example, someone using something for data compliance in a bank will tell us, and we reach out to the founders.

Companies often run parallel experiments — building with open source while testing startup solutions. Management struggles to know which approach is right.

That’s why we invest in both startups and open-source tech.

Regarding human-in-the-loop, we see four categories: Human doing, AI doing, Human managing, and AI managing. These four intersect in interesting ways.

One of our companies, Copilot Kit, builds open-source frameworks for human-in-the-loop systems. When evaluating a process, we ask three key questions:

1. Are there rules that define this process? If there are no rules or it’s beneficial to break them, AI won’t work.

2. Do we have a single source of truth for this process? If data is scattered across emails, spreadsheets, and texts, AI won’t work well.

3. Is there enough data showing input and output? Without data, AI can’t learn.

For example, we can’t automate investment decisions — we’ve made 125, but each company is unique. Not enough data to generalize.

Sramana Mitra: Yes, it’s very hard to automate.

This segment is part 4 in the series : 1Mby1M Virtual Accelerator AI Investor Forum: Marina and Nick Davidov, DVC
1 2 3 4