Great take and agree with your assessment. I think the longer term play for the model owners (realizing they are absolutely going to be fully commoditized) is using these armies of FDE's (that they own) and build vertical specific skills/products/agents. Anthropic did the v1 of this for the Finance Agents they just launched.
It's not just vendor lock from models themselves, customers don't want vendor lock-in from AI-Native companies who are currently trying to "own the workflow". These armies of FDE's directly from OpenAI/Anthropic are going to be purpose built to reinforce all their learnings on how to weed through the messy enterprise noise, in order to own the workflow as well.
I still can't decide if Anthropic/OpenAI will go out and buy vertical-specific agentic companies. Maybe for customer logos, pre-built FDE teams, and maybe domain knowledge...but not from a technology perspective.
Big portions of workflows can be productised, but not all of them, and whoever owns that last bit owns the customer. The analogy I keep coming back to is ERP. The integration partners ended up owning the customer relationship more than the ERP vendors did, because the customer-specific configurations, the customisations, and the institutional muscle memory all lived with the partner rather than with SAP or Oracle. Tomoro is the same play in reverse. The labs are buying into the integration layer because they’ve worked out that’s where the commoditisation stops.
On acquisitions, I’d push further than you. I think the labs will end up buying vertical-specific agentic companies, for exactly the reasons you laid out. Logos, pre-built FDE teams, and domain knowledge are what’s actually scarce now, and that’s the moat.
Great take and agree with your assessment. I think the longer term play for the model owners (realizing they are absolutely going to be fully commoditized) is using these armies of FDE's (that they own) and build vertical specific skills/products/agents. Anthropic did the v1 of this for the Finance Agents they just launched.
It's not just vendor lock from models themselves, customers don't want vendor lock-in from AI-Native companies who are currently trying to "own the workflow". These armies of FDE's directly from OpenAI/Anthropic are going to be purpose built to reinforce all their learnings on how to weed through the messy enterprise noise, in order to own the workflow as well.
I still can't decide if Anthropic/OpenAI will go out and buy vertical-specific agentic companies. Maybe for customer logos, pre-built FDE teams, and maybe domain knowledge...but not from a technology perspective.
Two things to add:
Big portions of workflows can be productised, but not all of them, and whoever owns that last bit owns the customer. The analogy I keep coming back to is ERP. The integration partners ended up owning the customer relationship more than the ERP vendors did, because the customer-specific configurations, the customisations, and the institutional muscle memory all lived with the partner rather than with SAP or Oracle. Tomoro is the same play in reverse. The labs are buying into the integration layer because they’ve worked out that’s where the commoditisation stops.
On acquisitions, I’d push further than you. I think the labs will end up buying vertical-specific agentic companies, for exactly the reasons you laid out. Logos, pre-built FDE teams, and domain knowledge are what’s actually scarce now, and that’s the moat.