I actually have a different view on this topic, as I think it’s becoming increasingly difficult for startups to build strong vertically integrated products.
To frame the discussion, here are my thoughts:
I don’t see agents as the new apps. Apps aren’t going away, while the agent layer itself will likely be dominated by large companies, as it feels like a natural progression of foundation model capabilities rather than a layer startups can sustainably own. Over time, horizontal agents will outperform vertical, domain-specific ones as foundation models become better at understanding context across domains. Because of that, owning the agent layer within a vertical won’t be a long-term advantage.
The real defensible moats for AI-first vertical products are either deep integration with physical operations or operating in heavily regulated spaces (e.g. fintech).
As for AI-first products more broadly, I believe the more sustainable path is FDE-style distribution rather than trying to build standalone agent-driven products.
Just to clarify, I’m not arguing that vertical startups win by owning the agent layer. I agree that this layer likely commoditizes and consolidates over time.
When I say “vertical AI,” I’m referring less to a domain-specific agent and more to a system that owns execution of a workflow — the rules, data, integrations, and decision loops that actually run the work. Agents are just one interface into that system, not the moat itself.
In that sense, FDE-style distribution and regulated or physical domains feel like enablers of vertical AI, rather than alternatives to it.
Thanks for the great writeup, Nouf!
I actually have a different view on this topic, as I think it’s becoming increasingly difficult for startups to build strong vertically integrated products.
To frame the discussion, here are my thoughts:
I don’t see agents as the new apps. Apps aren’t going away, while the agent layer itself will likely be dominated by large companies, as it feels like a natural progression of foundation model capabilities rather than a layer startups can sustainably own. Over time, horizontal agents will outperform vertical, domain-specific ones as foundation models become better at understanding context across domains. Because of that, owning the agent layer within a vertical won’t be a long-term advantage.
The real defensible moats for AI-first vertical products are either deep integration with physical operations or operating in heavily regulated spaces (e.g. fintech).
As for AI-first products more broadly, I believe the more sustainable path is FDE-style distribution rather than trying to build standalone agent-driven products.
Thanks for the thoughtful perspective.
Just to clarify, I’m not arguing that vertical startups win by owning the agent layer. I agree that this layer likely commoditizes and consolidates over time.
When I say “vertical AI,” I’m referring less to a domain-specific agent and more to a system that owns execution of a workflow — the rules, data, integrations, and decision loops that actually run the work. Agents are just one interface into that system, not the moat itself.
In that sense, FDE-style distribution and regulated or physical domains feel like enablers of vertical AI, rather than alternatives to it.