Hermes Puts On a Suit and Asks for Your Enterprise Budget

A $1.5 billion valuation tells you what investors believe about Nous Research; it tells you nothing about whether Hermes for Businesses will survive contact with your actual workflow.

That is the whole review in one sentence, and I want to be upfront that it is not a knock on the company. Nous has built real credibility with the open-source Hermes agent over three years, and credibility is harder to buy than a funding round. But the thing being sold now is different from the thing that earned the goodwill. An open-weights model you can poke at on your own hardware and a managed enterprise agent platform are two different products with two different failure modes, and the second one has a much longer list of ways to disappoint you.

What was actually confirmed

Nous Research confirmed a $1.5 billion valuation and the launch of AI agents aimed at business users. The round was led by Robot Ventures, with Nvidia, Union Square Ventures, Menlo Ventures, Samsung, and 1789 Capital participating. The company’s total funding now sits at $158 million after three years. The money is earmarked for the enterprise push, branded “Hermes for Businesses,” which lets companies deploy customized agents that handle multi-step workflows while keeping their data private and secure.

Worth flagging for anyone tracking the numbers: coverage has been inconsistent on the round size. Reporting from July 2026 described a Series B of at least $75 million at the same $1.5 billion valuation, while the confirmed announcement puts the round at $90 million. Rounds grow between first report and final close all the time, so this is probably just the gap between a leak and a signed term sheet. Still, if you see both figures floating around, that is why.

The claim I would test first

“Multi-step workflows” is the phrase doing the heaviest lifting in this launch, and it is the one I trust least in any agent product until I have watched it fail a few times. Single-step agents are a solved problem. Call a model, get text, done. The difficulty compounds the moment step three depends on step two having been correct, because errors do not stay small. They inherit. An agent that is 90% reliable per step is roughly 59% reliable across five steps, and most real business processes are longer than five steps.

So the questions I would put to Nous before signing anything:

• What happens when step four fails? Does the agent retry, halt, or hallucinate a result and keep moving?
• Can you see the intermediate steps, or only the final output? Opacity in the middle of a workflow is where trust goes to die.
• How much of “customized” means prompt configuration versus actual fine-tuning on your data?
• What does “private and secure” mean architecturally? Self-hosted, VPC-isolated, or a contractual promise about a shared cloud?

That last one matters more than anything else in the announcement. Data privacy is the pitch, and Nous has an open-weights heritage that makes the pitch believable in a way it would not be coming from a closed lab. If Hermes for Businesses can genuinely run inside your own perimeter, that is a legitimate differentiator for regulated industries and the one reason I would actually put this on a shortlist. If it turns out to mean a tenancy boundary on someone else’s infrastructure, it is the same promise everyone else makes.

Why the investor list is interesting

Nvidia’s participation is the least surprising part of this; they have backed a long list of model companies and it is strategically obvious. Samsung is more curious, and Union Square Ventures and Menlo Ventures are standard-issue for this stage. The presence of 1789 Capital, where Donald Trump Jr. is a partner, is the detail most writeups are leading with. I do not think it tells you anything about the product. It does tell you something about how AI funding has become entangled with politics, and that is a story for a different blog than this one.

My read for buyers

Nous has earned the right to be taken seriously, and $158 million across three years with this investor mix is a real mandate to build enterprise infrastructure. But a valuation is a bet on the future, not a product review, and nobody outside the design partners has run Hermes for Businesses against a messy internal process yet.

If you are evaluating agent platforms right now, treat this as a credible new option worth a pilot, not a default choice. Pick one workflow you understand end to end, one where you can audit every step and where a failure costs you time rather than money. Run it. Count the errors. Then decide.

I will be doing exactly that, and I will report back on what breaks.

🕒 Published: October 8, 2026