The big names are building agents. The money is in keeping them in bounds.
In the first days of October 2026, two established enterprise software companies set out their versions of software that acts on its own. On October 6, SAP demonstrated what it calls the Autonomous Enterprise at SAP Connect. It showed Joule Work and Joule Assistants generating value across every line of business, built on the SAP Business AI Platform.1 The same day, PwC and SAP said they would extend their AI collaboration beyond the Billing Assistant they built together.2
Two days later, at Zeta Live 2026, Zeta Global announced AthenaOS. It describes the product as an enterprise intelligence operating system built on the Zeta Data Cloud, which combines enterprise knowledge with Zeta intelligence and builds itself around a team's objective.3 Zeta also announced Athena MCP, a 'headless' product that brings its intelligence into the applications, agents and workflows teams already use.4 Alongside that came the Athena Inference Model (AIM), developed with Fireworks using NVIDIA Nemotron for specific AthenaOS uses,5 and early work with partners on a specialised AIM chip, described as a capital-light approach.6
Our topic brief also lists Meta's Enterprise Platform, with MongoDB's CEO moving over, as part of this race. The supporting material we hold on that move is thin, so we do not build on it here.
For an ordinary investor the main point is simple. When software is only recommending, a mistake is annoying. When software acts on its own, a mistake can move money or data. Much of the commercial story in this dossier is therefore about the controls around the agents, not the agents themselves.
A small deal with unusually clear numbers
The cleanest evidence of that shift is a modest transaction. On October 6, 2026, Tiny Ltd. (TSX: TINY), a Canadian holding company, said it had completed the purchase of the assets of Oso Cloud, a business specialising in authorization and access control, for approximately $1.9 million in cash.7 Authorization is the layer that decides who, or what, is allowed to do something.
Tiny says the business brings approximately $5.3 million in annualised recurring revenue. That would lift Tiny's total by about 7.6%, from $70.0 million to approximately $75.4 million on a pro forma basis.7 Oso Cloud served approximately 80 customers in security, fintech and developer software, all on recurring subscription contracts, several of them multi-year.7
Put the figures side by side and the price is roughly one-third of a year of the revenue Tiny says it acquired. We draw no valuation conclusion from that, because the dossier does not say how the business was growing or what it earned. It is still a useful marker of how a specialist control layer was priced.
A caution on the source. This is Tiny's own announcement, distributed through NewsEOD. Our measured reliability check found that 55% of 4,960 claims checked from that source held up.7 The deal terms are the company's account of itself, not independent confirmation.
Why the brakes matter: EDB's argument
In a piece presented by EDB, a database company, the central question is what actually stops an agent that tries something it was never authorised to do.8 EDB's answer is blunt: "The enterprise should not rely on a model choosing to follow policy. The policy has to be enforced by the system. That is the difference between hoping an actor stays in bounds and constructing bounds it cannot cross to begin with."8
This is a sponsored article, so it is also a sales pitch for EDB's own approach, and readers should weigh it that way. The logic is nonetheless easy to follow. It is the difference between asking an employee to respect a locked door and simply not giving them the key. It also explains why a business selling access control, like Oso Cloud, drew a buyer.
The brief also records that Harvey acquired Guardrails AI, and that Latitude and Dextr AI raised seed and Series A rounds. We hold no deal terms for Harvey and Guardrails AI or for Latitude, so we say no more about them. For Dextr AI the documents show a $6.7 million raise for hotel AI agents.9 That is an agent aimed at hotels, not at finance or governance, so it shows how widely agents are spreading, not how the control layer is being bought.
Finance: the most promising target, and the least forgiving
Sean Jacobsohn, a Norwest partner, offers the investor's view of the office of the CFO. Our brief records his view that finance is the most promising area to disrupt. He gives a practical reason: "There's actually one less layer of approval when they're buying their own software, so it is a little easier to replace it when they're the direct buyer."10
He is just as direct about the limits: "You don't want AI doing calculations because it is not good at math. There are certain workflows it can handle where it doesn't produce precise numbers. But when you need precision, accuracy and calculations, you can't rely on AI for that."10
He also thinks incumbents are harder to dislodge at their core: "I think it would be very hard to disrupt the core products of Workday, ADP, SAP, UKG and Dayforce. But it's easier to disrupt some of their secondary products, where the category isn't their core business."10
Read together, Jacobsohn's remarks and EDB's argument point the same way. The opening is real and the buyer has the authority to act. Even so, agents that touch the numbers need hard controls around them.
Does being first matter?
Daniel Docter of Dell Technologies Capital says it depends on the type of market: "in cases of category disruption, I think there’s value in first-mover advantage."11 By his account, later entrants into a newly created category can piggyback on the first mover's market education. Whether agent platforms count as disruption or creation is a judgment call, and nothing in the dossier settles it.
Via News analysis: where the evidence is thin
Our reading of the material is that the big platform announcements are claims about direction, not results. The dossier contains no customer outcomes, no revenue figures for SAP's or Zeta's agent products, and no comparison of how well they work. The only transaction with checkable numbers is the small Tiny and Oso deal. That is a pattern, not proof of one.
Source quality also varies. The three Crunchbase News pieces we draw on had a measured reliability of 50% of 38 claims checked, so we rely on them mainly for direct quotes that are attributed to named speakers.9,10,11
What to watch
- Whether SAP or Zeta publish customer results or pricing for their agent products, as opposed to demonstrations.
- Whether Tiny reports how Oso Cloud's roughly $5.3 million of recurring revenue behaves after the deal.
- Whether the other control-layer deals, Harvey with Guardrails AI and the Latitude round, come with disclosed terms.
- Whether the finance-focused startups Jacobsohn describes show they can keep AI away from the calculations while still automating the surrounding work.
