Every company in a recent survey says it will use AI agents, software that carries out tasks on its own rather than only answering questions. The same survey suggests the companies may not be able to feed those agents their own information. That gap, between intention and data readiness, is the most important thing to understand about enterprise AI right now.
The headline number, and why it deserves caution
According to a report published by MIT Technology Review, 100% of respondents plan to be using agentic AI within two years, and 69% expect to use it widely.1 In organizations the report labels 'data laggards', AI can reach 30% or less of company data.1 Our topic research summarises the same Google Cloud survey as finding that AI can reach about 45% of company data on average, and that only about half of respondents trust their agents' decisions. We could not match those last two figures to a verbatim claim in the source text we hold, so treat them as reported, not verified.
There is a second reason for caution. Via News measures how often claims from each source hold up when checked. For the MIT Technology Review piece, none of the 11 claims we checked held up.1 That sample is small, and it does not mean the survey is wrong. It does mean you should not treat these percentages as settled. The direction of the finding (enthusiasm far ahead of readiness) is plausible, but the exact numbers are not confirmed.
Where the money and partnerships are going
The deals in our dossier cluster around controlling AI rather than simply buying it. On July 22, 2026, Manulife and Microsoft announced a renewed, expanded five-year agreement. Manulife will adopt Microsoft's Frontier Suite and deploy Microsoft Agent 365, and it will expand Microsoft 365 Copilot to more than 30,000 employees.2 Manulife's Shamus Weiland said: "Our partnership with Microsoft is a critical enabler of Manulife's continued evolution into a truly AI-driven organization."2
A day earlier, on July 21, Box announced controls for AI agents working with enterprise content, including agent guardrails, oversight of third-party agent activity and prompt injection detection.3 Prompt injection is when hidden instructions in a document trick an AI into misbehaving. Tatsutoshi Murata of Nomura Research Institute said that, as the firm advances its use of AI agents, it expects Box "to provide the administrative features needed to safely leverage this new era of AI."3
Both announcements come from newswire sources. Our measurement shows only 57% of 4,956 checked claims from that outlet held up.2,3 These are companies describing their own deals, so read them as announcements, not independent assessments.
Capital is also flowing to new companies. Latitude, a payments platform founded by an ex-Stripe crypto team member, Cyril Mathew, raised a $35M Series A on September 10, 2026, for stablecoin-to-local-currency payments.4 The dossier does not say how Latitude uses AI, so we do not claim a direct link to the agent trend. Siemens and NVIDIA are also named in our topic summary as partners, but the dossier holds no detail on that deal.
What the startups say they are selling
CB Insights has been publishing a steady run of interviews with enterprise-AI executives. Their pitches show where companies think the money is. Penguin AI's Glenn Herzberg frames the market as labor, not software: "US Healthcare Administration runs about a trillion dollars a year, about a quarter of the total health spend, and the published estimates put around $570 billion of that in work that has no effect on health outcomes."5 That is a company's market pitch, not an audited fact. It does show the ambition: replacing administrative work, not just selling software.
Covecta's Ben Thomas makes the same move in banking: "for them we are not just disrupting their software budget but their labor budget as well."6 He says Covecta serves banks, non-bank lenders, building societies, credit unions and private credit organizations, currently in the US and UK.6 David Villalon of Maisa AI says his market is process automation at regulated industries, for tasks that must be "auditable, reproducible, and hallucination resistant."7 Hallucination here means an AI stating something false with confidence. The common thread is that buyers in regulated fields want agents they can check.
Investors are looking for the same pedigree. Emily Man of Primary said that colleagues' feedback on Casap's founders was "resoundingly clear that these were two exceptional builders."8 CB Insights also published CEO interviews on September 22 and 24, 2026, covering Veridox, Binary World, Shepherd, Arlo and LARX.9 The volume suggests the market is being mapped and sold as a category almost as fast as it is being built.
The chip supplier: what the checked numbers show
The one company in our dossier with fidelity-checked financials is Nvidia, whose chips and software underpin much of this AI buildout. Its connections in our data include products such as NeMo Guardrails and NeMo Agent Toolkit, customers such as Mount Sinai Health System and Yum! Brands, and Advanced Micro Devices as a competitor.10
Checked against SEC filings, Nvidia's cost of revenue (what it spends to deliver what it sells) was $16.62 billion in fiscal 2024, $32.64 billion in fiscal 2025 and $62.48 billion in fiscal 2026.11 That is nearly four times the fiscal 2024 level in two years. For scale, fiscal 2026 spending is more than the entire fiscal 2024 and fiscal 2025 totals combined. Cost of revenue is not sales or profit, so this shows the business growing, not how profitable it is.
Its cash balance rose from $7.28 billion to $8.59 billion to $10.61 billion over the same three fiscal years.12 Quarterly cash was $15.23 billion in Q1 2026 and $13.24 billion in Q1 2027, so it moves around within the year.12 Earnings per share were $11.93 in fiscal 2024, $2.94 in fiscal 2025 and $4.90 in fiscal 2026.13 That series does not rise smoothly, and the dossier does not explain why, so we do not draw a conclusion from it. If you hold a broad index fund you may own Nvidia indirectly, but our dossier contains no index weighting, so we cannot say how much.
What to watch
- Whether data access improves. If the 30% figure for laggards and the reported 45% average rise in the next survey, the readiness gap is closing. Look for sources whose figures hold up better under checking.
- Trust in agents' decisions. The reported finding that only about half of respondents trust their agents is the number most likely to slow adoption.
- Governance spending. Manulife's five-year commitment and Box's new controls suggest buyers pay first for control. Watch for more deals in that shape.
- Nvidia's next filings. Cost of revenue and cash on the balance sheet are the checked indicators of whether demand for AI hardware keeps growing.
Sources:
1 MIT Technology Review, "Scaling AI agents with trustworthy data" (undated)
2 NewsEOD, "Manulife Expands Partnership with Microsoft to Accelerate Enterprise AI Governance and Innovation" (2026-07-22)
3 NewsEOD, "Box Unveils New Controls to Secure AI Agents Operating Across Enterprise Content" (2026-07-21)
4 Via News event record, "Latitude raises $35M Series A" (September 10, 2026)
5 CB Insights, "Executive Interview: Penguin AI" (undated)
6 CB Insights, "Executive Interview: Covecta" (undated)
7 CB Insights, "CEO Interview: David Villalon, Maisa" (undated)
8 CB Insights, "Investor Interview: Primary on Casap" (undated)
9 Via News event records, CB Insights interviews published September 22 and 24, 2026
10 Via News entity graph (Nvidia relationships)
11 Via News verified data — Nvidia cost of revenue, FY 2024-FY 2026, SEC filing
12 Via News verified data — Nvidia cash, FY 2024-FY 2026 and Q1 2026/Q1 2027, SEC filing
13 Via News verified data — Nvidia EPS, FY 2024-FY 2026, SEC filing
