The most telling hire in enterprise AI this autumn was a database executive, not a model researcher. On September 28, 2026, Meta Platforms announced a new Meta Enterprise Platform, to be led by Chirantan Desai.1 Desai had been CEO of MongoDB for about a year. He announced he was leaving to head the new unit, where he will not be CEO.2 Via News's reading of the dossier is that a company known for consumer reach and open models is staffing its enterprise push with someone who has run a data-infrastructure business. We offer that as analysis, not as a stated Meta position. The announcements themselves do not describe the product.
Three groups are moving at once
The narrative we track frames this as three simultaneous moves. Big Tech is entering, with Meta as the example. Capital is flowing to AI-native vertical startups, named as Dextr AI, Latitude, Freehand and Groq. Incumbents are defending through a guidance raise at UiPath, security M&A in the form of Fortinet's deal for Virtue AI, and cloud growth at Alphabet of 82%.3 That narrative is our own aggregation, and the dossier holds no underlying filings for UiPath, Fortinet or Alphabet. We therefore repeat those items as part of the narrative and do not independently verify them.
What the money is buying: narrow agents, not general ones
The best-documented funding example is small and specific. Dextr AI emerged from stealth on September 24, 2026 with a $6.7 million seed round to build AI agents for hotel reservations, guest requests and staff coordination. Elevation Capital led the round and Foundation Capital participated.4 Dextr is headquartered in San Francisco, was founded by Sajid Shariff and Scott Arnold, and has built two products, Alfred and Daisy.5
The origin story matters for practitioners. Crunchbase News reports that Shariff returned to the Bay Area in 2024 and reconnected with people from his years in hospitality. He heard that hotels still faced rising labor costs, staffing gaps and pressure to raise revenue, and then offered to build an AI agent for one of them.6 That is a problem-first path to an agent product, and it is not a model-first one.
Latitude, a global payments platform founded by Cyril Mathew, a former member of Stripe's crypto team, raised a $35 million Series A on September 10, 2026. It builds a stablecoin-to-local-currency payments platform.7 Our entity graph also links Lightspeed Venture Partners, Wellington Management, Goldman Sachs Alternatives and Tiger Global as investors in Temporal Technologies, headquartered in Bellevue, Washington.8 The dossier does not say what Temporal builds, so we draw no conclusion from that. The point is only that large institutional capital is also present in this layer.
Vertical focus is a strategy
Two CB Insights interviews show the same pattern. Dentira's CEO, Vikas Gupta, says the company provides procure-to-pay services for multi-site healthcare. Its customers include management services organizations (MSOs) and dental service organizations (DSOs), which support hundreds, sometimes thousands, of offices or clinics.9 Gupta says: “Dentira is already the biggest player in a few verticals, and we're planning to become the largest vertical software provider in procurement technology overall.”9 That is the CEO describing his own company, and the dossier offers no independent check of the first half of the claim.
Light, a finance product, describes its target as fast-moving multi-entity businesses. Luke Richardson, its VP Marketing, says it is headquartered in Europe but operates across the world.10 Both interviews come from vendors describing themselves and should be read that way.
The investor's map: where incumbents are exposed
Sean Jacobsohn of Norwest offers the sharpest view of where agentic software can win. Asked about the big HR and payroll suites, he says: “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.”11 He also argues that buyer structure matters. “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,” he says of CFOs.11 For agent builders, the practical reading is to enter at the edges of an incumbent's portfolio and to sell to the person who holds the budget.
Jacobsohn also describes how he judges founders: “Before I invest, I'll go on a lot of sales calls I set up with the CEO to see how good they are at selling. To me, that's a big way of assessing the potential of a company.”11 He says he has built a Failure Museum of more than 1,500 items from failed companies and products.11 The signal is that in this market, distribution is being weighed alongside model quality.
The limits: governance and arithmetic
The brakes are as visible as the accelerator. In a sponsored piece on VentureBeat, EDB argues that governance cannot rest on the agent's good behaviour: “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.”12 The article was presented by EDB, a database vendor with a product to sell, so it is a position and not a neutral finding. The technical point is still a sound one. The article frames the central review question as what actually stops an agent from completing an action it was never authorized to do.12
The second limit is calculation. Jacobsohn says: “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.”11 This matters most in regulated finance and ERP, which Norwest sees as the next area of disruption. The narrative also names data sovereignty as a constraint.3 The likely shape is agents that handle judgment, routing and coordination, with deterministic systems left to do the arithmetic. That is our inference from these two quotes, not something either speaker says.
How much to trust the sources
Reliability varies across the dossier. Our measurements show that 50% of 38 checked claims from Crunchbase News held up, so its interview-based reporting here (Dextr AI, Jacobsohn) deserves a second look before it is relied on for anything consequential.4 The CB Insights interviews and the VentureBeat article have no measured reliability score in our dossier. The funding-round facts come from event records rather than from filings, which is another reason to treat the figures as reported and not audited.
What to watch
- Whether Meta publishes a product description or customer list for the Enterprise Platform. Today only its leader is public.1
- Whether Dextr's agents expand beyond hospitality, and what follow-on funding looks like after its seed round.4
- Whether vendors move policy enforcement into the data layer, as EDB argues, and whether buyers in regulated finance require it.12
- Whether the incumbents' defensive moves (guidance raises, security acquisitions) show up in filings that confirm the narrative.3