Dillon Rolnick, CEO of Nous Research, gives a 15-minute argument for owning the agent framework around your models. The talk also promotes Hermes Agent.
The business risk sits above the model
- Rolnick argues that model competition squeezes margins and shifts the battle toward agents embedded in business workflows.
- Once a platform holds procedures, organizational knowledge and productive work, replacing it becomes expensive.
- His two scenarios are higher prices after lock-in, or labs using their systems to compete directly with their customers.
- These are strategic forecasts; his spending estimates and agent-workforce projections are not evidence established by the talk.
Treat an agent as an application foundation
- Agents combine routine code, data sources and human judgment with probabilistic model behavior.
- That makes them closer to frameworks or operating systems than a chatbot interface, in Rolnick’s analogy.
- Keep the ability to swap models and modify the framework as workflows become more deeply integrated.
- Build the business-specific layer while inheriting improvements to the underlying agent system.
Open source changes the proposed moat
- Rolnick argues that cheaper software reproduction weakens feature exclusivity.
- He expects network effects, reputation and reliability to matter more than proprietary code alone.
- His discussion of model distillation, learning from another model’s outputs, challenges how labs describe that practice.
- Claims about legal status, vendor motives and software moats remain his argument, not settled findings.
Read the commercial interest alongside the claim
- Hermes Agent is his proposed open foundation; the speaker’s company builds it.
- He positions support, safety and scalability around an open standard as the business opportunity.
- Openness can preserve options, but the talk does not measure migration cost, reliability or security against proprietary alternatives.
“You should be able to swap in and out the models that you want.” — Dillon Rolnick, 12:28