Enterprise software is consolidating around a single question: who controls the layer where an AI agent stops advising and starts acting. In a four-week stretch of 2026, Asana, Coupa, Salesforce, and Vertice each bought a company that supplies exactly that capability. The execution layer, not the model, has become the battleground.
Why the Execution Layer Became the Prize
For two years, the AI conversation centered on model quality. That race has largely converged, and the harder problem is now visible: a model that can write a perfect summary of an invoice still cannot pay it, post it to the ledger, or route an exception to the right approver. Doing those things requires domain data, document context, system connectors, and workflow history that a general-purpose model does not have.
That gap is what vendors are racing to close. According to ERP.today's analysis, a cluster of 2026 acquisitions reflects a single pattern: enterprise software companies are buying the specific capabilities agents need to act, not just advise. The model is increasingly a commodity input. The execution layer, where actions actually happen inside business systems, is where differentiation and switching costs live.
The buying appetite is backed by budget. Info-Tech Research Group's enterprise survey, reported in June 2026, found that 91 percent of IT executives are bullish on AI and 96 percent expect their AI budgets to increase over the next twelve months. Vendors are acquiring execution capabilities because they know buyers are about to spend on agents that do real work.
The Four Deals That Define the Trend
Four acquisitions, all announced within a few weeks, map the shape of the execution layer.
Asana bought StackAI. Asana acquired the no-code agent builder StackAI in a deal reported at roughly 75 million dollars. StackAI builds agents that pull data from systems like Salesforce, Slack, and Google Workspace and act across them. Asana framed the purchase as a step toward becoming "the operating system for human-agent teams." The capability bought here is cross-system orchestration.
Coupa bought Rossum. Spend-management platform Coupa acquired Rossum to push intelligent document processing deeper into source-to-pay. Rossum runs a specialized transactional model trained on tens of millions of documents that learns each customer's specific paperwork. The capability bought here is turning unstructured documents into actions an agent can execute.
Salesforce bought Contentful. Salesforce signed a definitive agreement to acquire Contentful, a content platform used by more than 4,800 brands, to give Agentforce a native content layer for personalized experiences. The capability bought here is structured content that agents can assemble and publish.
Vertice bought Vendr. Vertice acquired Vendr on June 1, 2026, combining datasets that span more than 75 billion dollars of indirect spend across 32,000 vendors to power autonomous AI negotiation. The capability bought here is the proprietary data that lets an agent negotiate a contract rather than describe one.
Different markets, same logic. Each acquirer bought the missing piece that lets its agents do something, not just say something.
What "Act, Not Advise" Actually Means
The distinction sounds subtle and is not. An advising agent drafts a reply, summarizes a thread, or flags an anomaly, and a human does the rest. An executing agent updates the CRM record, submits the purchase order, reconciles the invoice, or sends the negotiation counteroffer. The first saves minutes. The second removes the handoff entirely.
That is also where the risk profile changes. An advising agent that is wrong produces a bad suggestion a human can ignore. An executing agent that is wrong produces a bad outcome a human has to unwind. This is the same lesson companies learned moving from pilot to production: the closer an agent gets to real actions, the more the work shifts from model selection to guardrails, permissions, and integration quality. If you want a grounding on what agents are and where they fit, our explainer on AI agents covers the fundamentals.
What This Means for Your Build vs. Buy Decision
What this means for businesses: the consolidation quietly raises the bar for building agents from scratch. A year ago, custom development was often the only way to get an agent to act inside your specific stack. Now the major platforms are absorbing that capability, which means an off-the-shelf agent can increasingly execute common workflows out of the box. The classic build versus buy framework still holds, but the line has moved: build when the process is genuinely unique, the data is too sensitive to hand a third party, or the agent's actions are a competitive differentiator. Otherwise, a platform-native agent is getting harder to beat on speed and cost.
There is a catch the marketing skips. A platform-native agent acts well inside its own walls and poorly outside them. Most real businesses run on a dozen systems, not one. The value is not in any single vendor's execution layer but in connecting agents to the actual systems where work gets done, which is where investment in workflow automation across your tools tends to pay off more reliably than betting everything on one suite's agent.
Our take: treat these acquisitions as a signal, not a directive. The fact that Salesforce bought a content platform does not mean you should standardize on Agentforce, and the fact that Coupa bought a document engine does not settle your procurement stack. It does mean that "we will build all of this ourselves" is a weaker default than it was six months ago, and "we will wait until the dust settles" is a slower one than your competitors can afford.
The New Risk: Lock-In at the Execution Layer
Concentration is the cost of convenience. When one platform owns your data, your workflows, and the agents that act on them, the switching cost is no longer just data migration. It is rebuilding every automated action the agent performed. This is the same vendor concentration dynamic we flagged in the AI vendor landscape shakeup, now pushed one layer deeper into operations.
Buyers can protect themselves without sitting out the trend. Three contractual and architectural requirements matter most:
- Data portability. Insist on clean export of both your data and the agent configurations that act on it. If the agent's logic is trapped in the vendor's format, you do not own your automation.
- Action audit logs. Every action an agent takes should be logged with what it did, when, on whose authority, and with what inputs. Without this, you cannot debug, audit, or defend an agent's decisions.
- Human approval gates. High-impact actions, anything touching money, contracts, or customer commitments, should require human sign-off until the agent has earned trust on lower-stakes work.
These are not exotic asks. They are the procurement floor for any system that will act on your behalf.
How to Prepare for Agents That Execute
- Map tasks by reversibility. List the actions an agent might take and sort them by how hard they are to undo. Reversible, low-stakes tasks are where execution belongs first. Irreversible, high-stakes tasks stay advisory until proven.
- Fix your data and APIs. An agent can only act on systems it can read and write cleanly. If your records are messy or your tools lack APIs, no acquisition will save you. Integration readiness is the real prerequisite.
- Define permission boundaries. Decide, in writing, what each agent is allowed to do, in which systems, up to what threshold, and who approves exceptions.
- Instrument everything. Log every agent action from day one. You cannot govern what you cannot see, and you cannot scale what you cannot audit.
Key Takeaways
- In mid-2026, Asana, Coupa, Salesforce, and Vertice each acquired a company supplying execution-layer capability, signaling that agents that act, not models that advise, are the new competitive ground.
- The execution layer is the data, connectors, and workflow logic that let an agent complete a task end to end inside business systems.
- The consolidation raises the bar for building agents from scratch; platform-native agents can increasingly act inside common workflows out of the box.
- The new risk is lock-in one layer deeper: rebuilding automated actions, not just migrating data. Require portability, audit logs, and human approval gates.
- Readiness is mostly about clean data, APIs, and governance, not picking the single smartest model.
The businesses that move early on agents that execute will have a meaningful advantage over those still treating AI as a suggestion engine. If you want to be one of them, let's start with a conversation.