In the first two weeks of June 2026, five enterprise software vendors closed or announced acquisitions targeting the same architectural gap: the execution layer that lets AI agents do real work inside the enterprise. Asana bought StackAI for $75 million. Salesforce signed a definitive agreement to acquire Contentful for $1 billion to $1.5 billion. Coupa acquired Rossum. Vertice acquired Vendr. And Palo Alto Networks closed the acquisition of Portkey to add an AI gateway to its Prisma AIRS security platform.
Five deals. Four different software categories — work management, CRM, spend management, procurement, cybersecurity. One pattern: every vendor is buying the layer that converts AI recommendations into autonomous execution against live enterprise systems, governed data, and real business processes. None of these companies are buying models. None are buying chatbots. They are buying the connective tissue between an AI agent and the systems it needs to act on.
The timing is not accidental. Worldwide AI spending is forecast to reach $2.52 trillion in 2026, a 44% year-over-year increase according to Gartner's January 2026 forecast. Gartner projects that 33% of enterprise software applications will include agentic AI by 2028, up from less than 1% in 2024. And yet Forrester's State of Agentic AI, 2026 report found that while three-quarters of enterprise leaders are adopting agentic AI, fewer than 10% have scaled it beyond pilot chatbots. The gap between intention and execution is where these acquisitions land — and where the next round of enterprise AI competition will be decided.
This article unpacks what each deal targets, maps the emerging enterprise AI agent stack by acquisition, and provides two frameworks every CIO should run before the end of Q3: the Enterprise AI Agent Stack Layer Map to see where your vendor portfolio has gaps, and the Build vs. Buy Decision Matrix to decide whether to close those gaps with acquisitions, partnerships, or internal builds.
The Five Deals and What They Actually Target
Each acquisition fills a specific layer of the enterprise AI agent stack. None overlap. That is the signal.
Asana + StackAI: Cross-System Agent Execution ($75M)
Asana acquired StackAI on May 28 to add no-code AI workflow capabilities that can execute across enterprise systems. StackAI's platform connects workflows, data, and actions across ERP, CRM, ITSM, document systems, and industry applications, with integrations spanning Salesforce, AWS, DocuSign, and Oracle.
The strategic logic is specific: Asana had the project context, ownership, and work history. StackAI gives it the ability to act across systems, not just track what happened after. Dan Rogers, Asana's CEO, framed it around "human-agent teams" — the idea that agents are no longer utilities bolted onto existing workflows but first-class participants in how work gets done. The price — $75 million for a no-code agent builder with enterprise-grade integrations — reflects what vendors are willing to pay for execution capability they cannot build fast enough on their own timelines.
Layer targeted: Cross-system workflow execution
Salesforce + Contentful: Agentic Content Assembly ($1–1.5B)
Salesforce signed a definitive agreement to acquire Contentful on June 1. Contentful is a composable content platform used by more than 4,800 brands. The Information reported the price at between $1 billion and $1.5 billion — significantly below Contentful's $3 billion valuation from its 2021 Series F.
The deal addresses a gap most enterprises have not yet noticed: AI agents need content, not just data. Salesforce already owned the customer record in Data Cloud and the agent runtime in Agentforce. What it lacked was a structured, governed content layer that agents could query, assemble, and deliver dynamically without manual publishing steps. Jujhar Singh, president of C360 applications at Salesforce, said Contentful adds a "native, headless, composable content layer that lets Agentforce dynamically assemble and deliver personalized experiences across channels."
For enterprise buyers, the signal is clear. Agents that answer questions need data. Agents that create customer-facing experiences need content — approved, versioned, permissioned, and composable. Salesforce is the first major vendor to acquire that layer explicitly for agentic use.
Layer targeted: Structured content for agent-assembled experiences
Coupa + Rossum: Transactional Document Intelligence
Coupa acquired Rossum to bring intelligent document processing into the core of its source-to-pay platform. Rossum's technology is powered by a specialized transactional LLM trained on tens of millions of documents and capable of learning from each customer's document set. The two companies had already partnered on complex invoicing for accounts payable.
The strategic logic: spend management agents cannot operate autonomously if they cannot read and interpret the documents that trigger spend — invoices, purchase orders, contracts, bills of lading. Coupa CEO Leagh Turner said the combined value has been proven in accounts payable and invoicing, and that Coupa sees future value in applying Rossum's technology across the entire platform. Rossum CEO Tomáš Gogár described it as combining "transactional intelligence with Coupa's large spend dataset."
The Coupa-MIT Data Science Lab Business Spend Index, published the same week, applied machine learning to an anonymized subset of $877.9 billion in 2025 procurement spend — the kind of dataset that makes document intelligence actionable at enterprise scale.
Layer targeted: Document understanding for autonomous financial workflows
Vertice + Vendr: Procurement Intelligence and Autonomous Negotiation
Vertice acquired Vendr to create what it describes as the world's largest procurement intelligence dataset: more than $75 billion in global indirect spend across 32,000 vendors, more than 2 million pricing data points, and more than 250,000 negotiated contracts covering software and services.
The combined platform feeds Vertice's autonomous negotiation agent, Ana, which negotiates with vendors within buyer-defined priorities, thresholds, and guardrails. The company said the acquisition strengthens more than 60 AI agents operating across benchmarking, vendor consolidation, third-party risk, renewal management, and procurement orchestration.
This is the most aggressive agentic play in the batch. Ana does not recommend a negotiation strategy and wait for a human to execute. It negotiates. The data is the moat: a negotiation agent is only as good as the pricing intelligence it can reference. Vendr's dataset — real-world pricing from real-world negotiations — is the kind of proprietary advantage that cannot be replicated from public data.
Layer targeted: Domain-specific intelligence for autonomous agent action
Palo Alto Networks + Portkey: AI Gateway Security
Palo Alto Networks completed its acquisition of Portkey on June 2, adding an AI gateway to the Prisma AIRS platform. Portkey's technology inspects AI traffic in real time while enforcing governance policies — providing runtime security, identity control for agents, and observability for enterprise AI operations.
Lee Klarich, Palo Alto Networks' chief product and technology officer, framed the acquisition around a specific failure mode: organizations "forced to choose between scrambling to integrate a patchwork of point products or falling behind while waiting for legacy platforms to catch up." The Portkey acquisition positions AIRS as the security control plane that sits between every agent and every model, inspecting every interaction.
This is the deal the other four depend on. None of the execution, content, document, or negotiation capabilities above can reach production at scale without a security layer that can govern autonomous agent traffic in real time. Forrester's Security Survey, 2026 found that 49% of security decision-makers named agentic AI as a concern. The attack surface is not theoretical: agents can impersonate each other, escalate privileges, and operate beyond real-time human oversight.
Layer targeted: Runtime security and governance for the AI agent fleet
Why Vendors Are Buying, Not Building
The urgency behind these acquisitions is a market timing problem. Forrester's data tells the story in two numbers: 75% of enterprise leaders are adopting agentic AI. Fewer than 10% have scaled it. That gap is the window. The vendor that closes the execution gap first — that makes the agent actually do the thing, not just recommend the thing — captures the switching cost before competitors can respond.
Building execution capabilities organically takes 12 to 18 months minimum. Acquiring takes 60 to 90 days. When Gartner forecasts that 40% of enterprise applications will integrate task-specific AI agents by end of 2026, the math favors buying.
The global agentic AI market is growing at 45.8% CAGR, projected from $7.6 billion in 2025 to over $50 billion by 2030. AI M&A has already exceeded $35 billion. And the acquisition targets are getting more specific: in 2024, vendors bought models and talent. In 2025, they bought infrastructure and platforms. In 2026, they are buying the domain-specific execution layers — the document intelligence, the content assembly, the negotiation data, the cross-system connectors — that make agents production-grade in specific enterprise workflows.
The Forrester report identified three structural barriers that explain why enterprises can't scale agents on their own:
- ROI uncertainty traps ambition in pilot mode. Most companies can't justify production beyond narrow efficiency gains.
- Governance gaps drive agentic sprawl. More than half of enterprises report sprawl even after adopting the NIST AI RMF.
- Platform confusion freezes commitment. Teams argue over whether to bet on a SaaS agent, an SI-built system, or a custom build while competitors ship.
The acquisition wave is the vendor market's answer to all three: prebuilt execution layers with domain-specific data, delivered inside platforms enterprises already own, with governance baked in at the architecture level.
Framework #1: The Enterprise AI Agent Stack Layer Map
The five acquisitions map to five distinct layers of what is emerging as the enterprise AI agent stack. Use this map to audit your current vendor portfolio and identify where you have gaps. Gaps in any layer block your agents from reaching production at scale.
| Layer | Function | Who Just Acquired It | What You Need |
|---|---|---|---|
| 5. Security & Governance | Runtime inspection, identity control, policy enforcement for AI traffic | Palo Alto Networks (Portkey) | AI gateway that sits between every agent and every model, enforcing permissions and audit in real time |
| 4. Cross-System Execution | Agents act across ERP, CRM, ITSM, document systems | Asana (StackAI) | Workflow engine with enterprise connectors that lets agents trigger actions in systems they don't own |
| 3. Structured Content | Agents assemble approved, versioned content for customer-facing delivery | Salesforce (Contentful) | Composable content platform with API-first architecture and permission inheritance |
| 2. Document Intelligence | Agents read, interpret, and act on transactional documents | Coupa (Rossum) | Specialized IDP trained on domain-specific document types (invoices, contracts, POs) |
| 1. Domain Intelligence | Proprietary data that gives agents competitive advantage in specific workflows | Vertice (Vendr) | Dataset that cannot be replicated from public sources — pricing, negotiation history, benchmark data |
How to use it: Map your current vendor stack against these five layers. If you have a gap at Layer 5 (security), nothing below it reaches production safely. If you have a gap at Layer 1 (domain intelligence), your agents will be generic — they'll work, but they won't outperform a competitor's agent that has proprietary data. The layer where you have the weakest coverage is your highest-priority Q3 investment.
Three patterns emerging in the field:
-
Platform-first enterprises (heavy Salesforce, ServiceNow, or Microsoft tenancy) will get Layers 2–4 from their incumbent vendor's acquisition strategy. They need to independently solve Layer 5 (security) and Layer 1 (domain data).
-
Best-of-breed enterprises (heterogeneous SaaS sprawl) will need to assemble across vendors. The risk is integration complexity: five different execution layers from five different vendors do not automatically compose into a governed agent stack.
-
Regulated enterprises (financial services, healthcare, government) must solve Layer 5 first or nothing else matters. The Palo Alto/Portkey acquisition is the most strategically important deal in the batch for this segment.
Framework #2: The Build vs. Buy Decision Matrix for AI Agent Capabilities
If you're a CIO, CTO, or VP of Engineering deciding how to close gaps in your AI agent stack, use this matrix. Score each capability gap on five criteria, weight by strategic priority, and let the weighted total guide your approach.
| # | Criterion | Build (Score 1–5) | Buy/Acquire (Score 1–5) | Partner (Score 1–5) |
|---|---|---|---|---|
| 1 | Time to production (months to first production workload) | 1 if >12mo | 5 if <3mo | 3 if 6–9mo |
| 2 | Data moat (does the capability require proprietary data you don't have?) | 2 (must accumulate) | 5 (comes with acquisition) | 3 (shared access) |
| 3 | Integration depth (how tightly must this layer couple to your existing platform?) | 5 (native from day one) | 3 (integration work required) | 2 (API surface only) |
| 4 | Competitive differentiation (does owning this layer create switching cost for your customers?) | 4 (IP stays internal) | 5 (absorb competitor's IP) | 2 (commoditized access) |
| 5 | Governance and audit (can you certify the capability meets regulatory requirements?) | 5 (full control) | 3 (must audit acquired code) | 2 (dependent on partner's posture) |
How to use it: Weight criteria 1–5 by your strategic priority on a 1–5 scale. Multiply each weight by the approach score. Sum across all five criteria. The approach with the highest weighted total is your default.
Decision rules we're seeing in the field:
-
If time to production is your binding constraint (agents need to be in production by Q4 2026), buy or partner. Building from scratch misses the window. Asana, Salesforce, Coupa, and Vertice all made this calculation.
-
If data moat is your binding constraint (the capability is only valuable with proprietary data), buy. Vertice's acquisition of Vendr is the template: $75 billion in negotiation data cannot be assembled organically in any reasonable timeframe.
-
If integration depth is your binding constraint (the agent layer must operate natively inside your platform), build. No acquisition integrates itself. Salesforce will spend 12+ months integrating Contentful into Customer 360. If you need that level of integration without the acquisition overhead, build it.
-
If governance is your binding constraint (regulated industry, audit-ready requirement), build or buy with extensive due diligence. Partner integrations in regulated environments introduce third-party risk that auditors will flag.
The Architectural Question Behind Every Deal
These five acquisitions answer a question that has been circulating in enterprise architecture circles for the past six months: is the value in the model, the platform, or the execution layer?
The model layer is increasingly commoditized. Microsoft Foundry now catalogs more than 11,000 models. Switching between Claude, GPT-5.5, Gemini, or an open-source alternative is an API call, not a replatforming project. The platform layer — the context substrates, knowledge graphs, and orchestration engines — is where Microsoft IQ, Glean, ServiceNow, and Salesforce are competing.
But execution is where the agent actually does the thing. An agent with the best model and the best context layer still cannot negotiate a vendor contract without pricing data, cannot process an invoice without document intelligence, cannot deliver a personalized customer experience without structured content, and cannot operate at scale without runtime security.
The execution layer is where domain expertise becomes a defensible moat. Vertice didn't buy a model. It bought 250,000 negotiated contracts. Coupa didn't buy a chatbot. It bought a transactional LLM trained on tens of millions of financial documents. Palo Alto didn't buy an AI assistant. It bought the control plane that sits between every agent and every enterprise system.
That is the pattern. That is where the enterprise AI market is moving. And it will accelerate: expect at least a dozen more execution-layer acquisitions before the end of 2026 as every major enterprise software vendor races to fill the same five-layer stack.
What CIOs Should Do This Quarter
The acquisition wave crystallizes three actions for Q3 2026:
1. Audit your agent stack against the five-layer map. Every enterprise already has some agent capability — a Copilot deployment, an Agentforce pilot, an internal build on LangChain or CrewAI. Map what you have against the five layers. The gap is your risk surface and your investment priority.
2. Run the build vs. buy matrix for your top three gaps. The five acquisitions prove that buying is faster than building when time to production and data moats are the binding constraints. But integration depth and governance requirements may push you toward building — especially in regulated industries. The matrix gives you a structured way to make that call instead of defaulting to whatever your largest vendor is selling.
3. Solve security first. Layer 5 is not optional. The Forrester data on agentic AI security concerns — 49% of security decision-makers flagging it, governance gaps persisting even with NIST AI RMF adoption — means that no agent stack reaches production at scale without runtime security. If you don't have an AI gateway strategy, nothing else in the stack matters.
The window is 90 days. The vendors are moving. The question is whether your enterprise is building the stack, buying it, or watching while your competitors do both.
Continue Reading
-
Microsoft IQ Is GA. The Enterprise Agent Context War Just Reset. — How Microsoft's four-layer unified context system reshapes the grounding substrate beneath every enterprise agent.
-
88% Have AI Agent Incidents. 14% Have Approval. The Gap Cognizant Just Productized. — Why the governance gap in agentic AI is the largest unpriced risk in enterprise software.
-
IBM Bets the Stack: An AI Operating Model for the Multi-Agent Era — IBM's vision for how enterprises govern, orchestrate, and scale multi-agent systems.
Rajesh Beri is Head of AI Engineering at Zscaler.