From Insight to Action
The clearest pattern in enterprise software this year is not a single product or a single vendor. It is a shift in what enterprise AI is for. Through 2024 and 2025, the pitch was insight: copilots and assistants that summarized, explained, and recommended. In 2026, the pitch is execution. The market is converging on the layer where an AI decision becomes an operational action inside a live ERP, CRM, or procurement system. As Nominal's CMO Stephanie Montelius puts it, chatbots explain, agents execute.
That sentence sounds like marketing, but it captures a real architectural and commercial reorientation. Explaining is cheap and safe. Executing is expensive and dangerous, because an agent that can act can also act wrongly, at machine speed, across systems of record. The vendors that figure out how to let agents execute reliably, and prove what they did, are the ones positioning to own the next decade of enterprise software. The race is for the execution layer.
The Deals Tell the Story
Follow the acquisitions and the thesis becomes hard to miss. Asana bought StackAI to wire agent execution across ERP, CRM, ITSM, Salesforce, AWS, DocuSign, and Oracle. Coupa bought Rossum, whose document model is trained on tens of millions of documents, to automate source-to-pay. Salesforce bought Contentful and its 4,800-plus brand footprint to give Agentforce a structured content layer. Vertice bought Vendr along with a 75 billion dollar global indirect-spend dataset, 2 million pricing points, 250,000 negotiated contracts, and an autonomous negotiation agent called Ana.
None of these are insight plays. Each one buys a specific capability an agent needs to act: workflow connectivity, document understanding, content assembly, or negotiation data. The hyperscalers are doing the same at the infrastructure tier, with Google Cloud pushing Gemini Enterprise into Workday and extending to IBM and NTT DATA to co-develop hundreds of agents, and Microsoft connecting Dynamics 365 Field Service to Project Operations so service execution and financials close the loop. The money is flowing to the execution layer.
Execution-Grade Products Are Shipping
This is not only about M&A. Vendors are shipping execution into their core products. Priority Software's Version 26.0 introduced an aiERP Companion plus task-specific agents across finance, sales, and supply chain, deployed to a base of 75,000 customers in 70 countries. CEO Sagive Greenspan describes agents that analyze signals, trigger workflows, and execute routine operations inside the ERP. Nominal is building agentic performance management that handles accounting workflows, reconciliation, and intercompany transactions.
The significance is that execution is moving from demo to default. A task-specific agent inside the ERP that reconciles an account or triggers a replenishment workflow is a different proposition from a chat assistant bolted onto the side. It touches the system of record, and it does so with the vendor's name on the outcome. When a vendor with 75,000 customers ships this as a version release rather than a pilot, the category has crossed a line.
The Governance Footnote Is the Headline
The detail we find most reassuring is buried in how the credible vendors describe their agents. Both Nominal and Priority frame execution around human oversight, auditability, and customer standard operating procedures, explicitly not autonomous operation. That phrasing is not legal boilerplate. It is an admission that execution-grade agents are only as valuable as the controls around them, and that selling action without accountability is a liability, not a feature.
We would treat that framing as a buying filter. Any vendor pitching agents that execute inside your systems of record should be able to answer three questions cleanly: who approves the action, how is it logged, and how does the agent respect your existing procedures. Vendors that lead with autonomy and treat oversight as an afterthought are selling you risk dressed as productivity. The ones that lead with auditability understand what enterprise execution actually requires.
What CIOs Should Take From This
The practical implication is that the evaluation criteria for enterprise AI have changed. A year ago, buyers compared chat interfaces and answer quality. Now the question is whether a platform can take action against your systems of record and prove what it did. That favors vendors with deep hooks into the transactional layer and credible audit trails, and it disadvantages thin assistants that only ever talk. The center of gravity has moved from the conversation to the consequence.
Our advice is to map your own execution layer before a vendor maps it for you. Decide which workflows you would actually let an agent run, what oversight each one requires, and where the audit trail must live. Then evaluate platforms against that map. The enterprises that win this cycle will not be the ones with the most agents, they will be the ones who let agents execute exactly where it is safe and not one step further, and who can show their work when someone asks.
