▶ Watch ↗15:12
Conference affiliation: Nous Research
Dillon Rolnick is CEO of Nous Research, where he is building a business around AI software that users can run, modify and use with different model providers. His concern is what happens as a company builds its procedures around an agent: changing models may remain easy while replacing the software that coordinates the company’s work becomes increasingly expensive.
Before Nous, Dillon led research at a venture capital fund. He later served as Nous’s chief operating officer, the role listed in September 2025 SEC meeting materials, before becoming chief executive.
His work includes collaborative technical research. Dillon is one of six co-authors of the 2024 preliminary DisTrO report, alongside Bowen Peng, Jeffrey Quesnelle, Ari Lotter, Umer H. Adil and Esteban La Rocca. DisTrO addresses a physical constraint on who can train large models: conventional distributed training exchanges large quantities of information between accelerators, favoring machines connected by expensive, high-bandwidth networks.
The team explored optimizers that reduce that communication. Its preliminary experiment trained a 1.2-billion-parameter language model with convergence comparable to a conventional AdamW setup while requiring much less bandwidth. The project’s ambition was to make training across ordinary internet connections practical, widening participation beyond organizations able to assemble a tightly connected supercomputer. The report establishes Dillon’s participation in that shared work; it does not assign him a separate technical invention.
Nous’s early model work emphasized post-training: adapting existing open models through data and fine-tuning. In his October 2026 fundraising letter, Dillon described how the team subsequently developed an agent stack for its own work and released Hermes Agent under the MIT license in February 2026. Hermes retains context across sessions and turns experience into reusable skills, allowing users to build on previous work.
That accumulation is both the attraction and the risk. In his AI Engineer NYC talk, Dillon argued that falling model prices need not mean falling costs for a business dependent on a proprietary agent. Once employees have built years of procedures around it, the provider may be able to raise prices because leaving would require reconstructing those processes. He also warned financial firms that a lab seeking their market expertise as a partner could eventually compete for the same investment opportunities. These are his arguments about providers’ incentives, rather than established outcomes.
Dillon compares agents to operating systems because they coordinate more than conversation. Drawing on another speaker’s investment-decision example, he described an application that combines data sources, executable routines, analysts’ opinions and model judgment. The firm’s expertise lies partly in how it arranges those inputs. Owning and modifying the agent software lets it preserve that arrangement while replacing the reasoning model underneath.
He also draws a limit around what customers should have to maintain. A financial firm should concentrate on its investment work while benefiting from improvements to the agent framework. His proposition is that an open agent can provide those improvements without requiring the firm to surrender control over where its software runs or which models it uses.
Dillon applies the same argument to Nous’s own competitive position. He said the team admired an interface feature in xAI’s Grok bots, used Hermes Agent to reproduce it over a weekend, and released it as an open-source feature. For him, the episode illustrates why proprietary features may provide only a short-lived advantage.
He expects network effects, reputation and reliability to matter more as software becomes easier to reproduce. His fundraising letter puts that position into a commercial plan: bring Hermes to businesses while preserving their control over the intelligence stack. In his talk, he described support, safety and scalability as services Nous could provide around the open agent. Customers’ freedom to leave makes continued adoption something the company must earn.
▶ Watch ↗15:12
Loading more talks and highlights…