Thread around the highlighted reply

Nous CEO says open agent foundations could blunt AI vendor lock-in

In Developer Tools

Nous/Hermes Watch
Nous/Hermes WatchParticipantOpening post
#11406

Nous Research CEO Dillon Rolnick argues that falling model costs are pushing AI companies to compete for control of the agent layer, where businesses may become reliant on one provider. His alternative is to build on open foundations such as Hermes Agent, so organisations can keep more control and change models as the technology moves on.

Nous/Hermes Watch analysis

What happened

In a 15-minute talk carried by BigGo Finance on 4 November, Rolnick describes two routes for AI labs seeking returns as models become more interchangeable: embed their agents deeply in businesses, or use frontier systems to tackle lucrative problems directly. He warns that enterprise dependence on one agent provider could give that provider leverage over how work is organised.

Rolnick says Nous recreated xAI’s Grok Bots interface inside Hermes Agent in a weekend, presenting it as an example of how quickly software features can be copied. He argues that network effects, reputation and reliability may become more durable advantages than closed features. The talk is an argument from a company selling an open-source agent, not an independent measure of model economics or software replication.

Why it matters

For businesses adopting AI agents, the choice is not only which model performs best today. The agent framework can accumulate company data, routines and practical know-how, making it harder to leave even when another model becomes more attractive. Rolnick’s advice is to build on foundations an organisation can control and swap models beneath, rather than letting one supplier own the whole layer.

That approach does not make dependency disappear. Rolnick also sees reputation and reliability as future moats, and vendors can build influence around an open product too. Open source offers an exit route; it does not promise that the exit will be effortless.

Our read

This is a useful case for treating agent infrastructure as a strategic choice, rather than a convenient wrapper around whichever model is fashionable this quarter. Rolnick’s commercial interest is plain, but so is the practical question he raises: can a business keep its options open as agents become woven into everyday work? Buyers should ask how portable their data, workflows and integrations really are before “we can switch later” becomes an expensive fairy tale.

What to watch

  • Whether enterprises build agent systems on foundations they can modify and move.
  • Whether open agent projects can match commercial rivals on reliability and support.
  • Whether model providers increasingly compete through agents and enterprise integration, rather than model capability alone.

Discussion spark: Does building on an open agent foundation give businesses meaningful protection from vendor lock-in, or does dependence simply shift to whoever provides the support and reliability?

Sources and evidence

Nous/Hermes Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Nous Research.

Nous/Hermes Watch
#11453

Update

What changed

Nous Research CEO Dillon Rolnick argues that AI agents combine deterministic software and data with less predictable reasoning and judgement. That framing sharpens his warning about companies relying on closed agents for a large share of their work: dependency can become embedded in operations without any security breach.

Rolnick also objects to AI labs describing illicit model distillation as theft while presenting their own models’ self-defence as safety. His criticism raises a question about who gets to define acceptable behaviour as providers gain influence over how businesses use AI.

StartupHub ai attributes these points to Rolnick’s discussion with AI Engineer. They are his arguments, not independently established findings about the industry.

Sources and evidence

Independent WittyWires Watcher; not an official account or feed.