BoxGroup's David Tisch on AI M&A, brand moats, and being a seed investor in Cursor and Clay
May 12, 2025 with David Tisch
Key Points
- Brand, not switching costs, will determine winners in AI app-layer companies, with cultural ubiquity functioning as the real moat that survives big tech acquisition.
- BoxGroup's seed investments in Cursor and Clay position the firm to benefit as AI compression of product development timelines makes venture-scale returns feasible on historically slow builds.
- Major tech platforms acquiring AI software teams at scale depends on regulatory permission, but elevated valuations across foundation models and public tech make founder price expectations structurally easier to meet than a decade ago.
Summary
David Tisch, managing partner at BoxGroup and a seed investor in both Cursor and Clay, argues that brand is the most underappreciated moat in AI application-layer companies — and that the coming M&A cycle will test whether big tech can actually buy it.
Brand as the real defensibility
Tisch is skeptical that switching costs will protect AI app-layer winners the way traditional software moats did. What will matter instead is brand momentum — being the fastest mover, not the first. ChatGPT is his clearest example: a technically strange name that became a household verb, the same way Google won search less through pure product superiority and more through cultural ubiquity. A better model alone won't displace that.
At the app layer, brand compounds through user scale and fundraising narrative. Companies that raise at rising valuations every six months signal momentum to potential hires, customers, and acquirers — creating a flywheel where capital itself becomes a brand asset. The endpoint, Tisch says, is becoming the default safe choice: "No one gets fired for picking Salesforce." Disrupting that requires a challenger to be a genuine outlier, not just marginally better.
AI M&A: the real unlock is big tech, not foundation models
Tisch thinks the more interesting M&A question isn't foundation models buying app-layer companies — it's whether the major public tech platforms (he references the "big seven") will acquire at scale, contingent on the regulatory environment allowing it. His read is that today's elevated valuations among both foundation model companies and large-cap tech make it structurally easier to meet founder price expectations without the deal being a meaningful dent to the acquirer.
He draws the parallel to Meta's 2005–2015 run of roughly 50 acquisitions, ranging from aqui-hires like FriendFeed — which brought in engineering talent that built core Facebook infrastructure — to transformative bets like Instagram and WhatsApp. That model of talent and product acquisition largely stopped over the past decade, killed by regulatory pressure and valuation misalignment. Tisch thinks AI could force it back: companies that can't buy modern software capabilities may have to buy the teams building them.
BoxGroup's portfolio positions
BoxGroup was a seed and pre-seed investor in Cursor, and Tisch confirms the math circulating online about FTX's Cursor stake — implying a $200M to $500M range — doesn't add up. He's also a founding investor in Clay, describing it as an eight-year build that is only now executing a go-to-market motion that competes with entrenched software players. His view is that AI-accelerated development will compress the timeline for that kind of deep product construction, which historically required capital and patience that early-stage VC structures don't easily support.
Customer service automation
On Klarna's public walk-back of aggressive AI-driven headcount cuts, Tisch is measured. Around 65% of e-commerce customer service queries — "where is my item" — are clearly automatable. A further tranche is automatable with some configuration. The 3–10% that genuinely requires judgment is typically some version of a retention or save scenario. His conclusion is that most customer service becomes automatable sooner rather than later, but announcing mass layoffs as proof of AI's present-day magic is overstating the case.