M&A in Agentic AI: Which Incumbents Are Actually Shopping for Agent Companies

Table of Contents

• Why Incumbents Are Buying at All
• The Four Acquisition Theses
• The Legacy SaaS Giants: Defending the Seat
• Cloud and Infrastructure Players
• Security Vendors and the Agent Attack Surface
• Vertical Incumbents Nobody Is Watching
• How to Read an Incumbent's Shopping List
• What Founders Should Take From This
• Insights Most People Overlook
• References

Why Incumbents Are Buying at All

Start with the uncomfortable math that every incumbent boardroom is now staring at. The dominant software pricing model of the last fifteen years was the seat, you pay per user, per month, forever. Agentic AI-as-a-service breaks that model on purpose. When an agent does the work of a sales development rep or a junior analyst, the customer no longer needs the seat, and increasingly wants to pay for the outcome, a booked meeting, a resolved ticket, a reconciled invoice, rather than for access.

That is an existential pricing transition, not a feature gap. An incumbent that sells 50 seats of a CRM to a mid-market company has every reason to fear a GaaS vendor that sells the same company one agent at a fraction of the headcount cost. So the buying spree is, at root, defensive. Incumbents are acquiring agent capabilities to make sure they are the ones cannibalizing their own seat revenue rather than ceding it to a startup. Microsoft's own framing of its Copilot strategy, selling agents as additive consumption on top of seats, is the polite version of this. The blunt version is that everyone in enterprise software now agrees with the a16z thesis that AI will compress the price of services into software, and nobody wants to be on the wrong side of that compression.

There is a second driver that gets less airtime: talent scarcity. Building a reliable production agent is genuinely hard. The teams that have shipped agents which survive contact with real enterprise data, handling tool failures, retries, evaluation, guardrails, are rare, and they are expensive to assemble from scratch. Buying one is often faster than hiring one, which is why so much of the actual deal volume is small acqui-hires rather than revenue acquisitions. I'll come back to that, because it is the single most under-reported pattern in the space.

The Four Acquisition Theses

Every agent acquisition I've watched fits one of four theses. Naming them makes the noise legible.

The product-hole fill. The incumbent has a roadmap gap an agent neatly plugs, say, an autonomous data-engineering agent or a coding agent, and buying is faster than building. These deals carry real revenue and real product, and they price like normal software M&A, just at frothier multiples.

The acqui-hire. The incumbent doesn't want the product; it wants the eight people who know how to make agents not fall over in production. Revenue is incidental, sometimes zero. The deal is structured around retention packages, and the acquired product usually gets sunset within a year. This is the pattern of big tech absorbing AI teams that has accelerated since 2023, and it overlaps heavily with the acqui-hire dynamics covered elsewhere in this beat.

The defensive moat. The incumbent buys to deny a capability to a rival or to a category of insurgents, locking up a workflow, a dataset, or a distribution channel before someone else does. These deals can look overpriced in isolation and only make sense as board-level insurance.

The attach play. The incumbent buys an agent specifically to bolt onto an existing product and lift its consumption or expansion revenue, the "agent attach" thesis. The agent is a revenue multiplier on the installed base, not a standalone line.

Most real deals are blends. But if you can't articulate which thesis dominates a given transaction, you don't actually understand the deal.

The Legacy SaaS Giants: Defending the Seat

This is where the money and the anxiety are concentrated.

Salesforce

Salesforce has been the loudest, repositioning the entire company around Agentforce and explicitly pricing agent conversations as consumption. Its M&A behavior follows the defensive-moat and attach theses: data infrastructure (the Informatica acquisition is the obvious tell, agents are worthless without governed data) plus smaller tuck-ins for vertical workflow and conversational capability. Watch Salesforce for data and unstructured-content deals, because its bottleneck isn't reasoning, it's feeding the agent clean enterprise context.

ServiceNow

ServiceNow sits on the workflow layer that agents most naturally automate, IT service management, HR, customer ops. It has been acquiring aggressively around AI reasoning, data, and workplace agents. Of all the incumbents, ServiceNow has the cleanest strategic logic: it already owns the system of action, so an agent that executes within its workflows is pure attach revenue with very little cannibalization risk. Expect it to keep buying reasoning and orchestration layers.

Intuit, Workday, SAP

The back-office incumbents are quieter shoppers but arguably more exposed. Intuit's domain, bookkeeping, tax prep, payroll, is almost a textbook target for outcome-priced agents, and Intuit has responded by building heavily and acquiring selectively around financial-data and assistant capabilities. Workday and SAP face the same dynamic in HR and ERP: an agent that closes the books or processes a requisition end-to-end is a direct threat to module-based licensing, so both are buying to internalize that automation. SAP's positioning of Joule as an agent orchestration layer is the defensive version of this move.

The through-line: these companies are not buying agents because agents are cool. They are buying because the alternative is watching a startup unbundle a profitable module one outcome at a time. Bain's analysis of generative AI's disruption to software economics lays out why the incumbents with the most seat revenue have the most to defend, and the deepest balance sheets to defend it with.

Cloud and Infrastructure Players

The hyperscalers and model labs play a different game. Microsoft, Google, and Amazon mostly don't need to buy agent applications, they want the platform layer underneath: orchestration frameworks, evaluation tooling, memory and retrieval infrastructure, and the talent that builds it. Their acquisitions skew toward the "picks and shovels" of agent infrastructure rather than vertical agents, and a meaningful share are acqui-hires dressed as licensing deals to dodge antitrust scrutiny.

The model labs themselves, OpenAI, Anthropic, are an emerging and unusual class of acquirer. When a foundation-model lab buys an application-layer agent company, it's a signal about where the lab thinks the value will accrue, and it puts the lab into direct competition with its own API customers. That tension is a recurring theme across this funding-and-market beat, and it shapes how application-layer founders should think about who their eventual buyer might be.

Security Vendors and the Agent Attack Surface

Here is a buyer category most coverage misses entirely. Autonomous agents create a brand-new attack surface, prompt injection, tool-permission escalation, data exfiltration through agent actions, identity and authorization for non-human actors. Security incumbents (Palo Alto Networks, CrowdStrike, Okta, and the identity vendors in particular) have started acquiring and partnering around agent security, non-human identity, and runtime guardrails.

This matters strategically because agent reliability and agent security are fast becoming the gating factor for enterprise adoption, and therefore for which GaaS startups become acquirable at all. A vertical agent with weak security posture isn't just a worse product; it's a smaller acquisition target, because the incumbent will have to rebuild the trust layer anyway. Founders building in regulated verticals should read the security M&A wave as a direct signal about what enterprise buyers will demand before they sign.

Vertical Incumbents Nobody Is Watching

The most under-covered shoppers aren't software companies at all. They're the vertical incumbents whose core business is the workflow an agent automates: the large staffing firms eyeing recruiting agents, the legal-tech and information giants (think Thomson Reuters, RELX/LexisNexis) buying legal and research agents, the healthcare-IT players acquiring clinical-documentation and prior-authorization agents, and the BPO and contact-center operators buying customer-service agents to avoid being disintermediated by them.

These buyers are strategically motivated in a way pure-software incumbents aren't: for a BPO, an agent isn't an attach play, it's a replacement for the labor that is their business model. That makes them willing to pay strategic premiums and to move fast. If you're a founder in a vertical, your most likely acquirer may not be the obvious SaaS name, it may be the incumbent that currently sells the human-delivered version of what your agent does. The build-versus-buy calculus these enterprises run is shifting toward buy precisely because the in-house teams that could build agents are the scarcest resource of all.

How to Read an Incumbent's Shopping List

A few practical heuristics for anyone tracking this market, investors, operators, or founders planning an exit.

Follow data and orchestration deals, not just agent deals. When an incumbent buys a data-governance or retrieval company, it's clearing the runway to deploy agents at scale. Those deals predict the agent acquisitions that follow.

Distinguish announcements from acquisitions. Every incumbent has an "AI agent strategy" slide. Far fewer are writing checks. The check-writing is the signal; the slide is noise.

Size tells you the thesis. Sub-$100M deals are almost always acqui-hires or capability tuck-ins. Mid-size deals are product-hole fills. Only the rare nine- and ten-figure deals are genuine revenue acquisitions, and those are where the market is testing whether agent revenue is durable enough to underwrite at a real multiple.

Watch the org chart after close. If the acquired founders get a platform-level title, it was a talent and strategy bet. If the product gets folded into an existing SKU within months, it was an attach or defensive play. The post-deal structure reveals the true thesis more honestly than the press release.

What Founders Should Take From This

Building to be acquired is usually bad advice, but understanding who buys and why is just good market literacy. Three takeaways.

First, the durability of your usage revenue determines whether you're an acqui-hire or a revenue acquisition, and the multiple gap between those outcomes is enormous. Second, your security and reliability posture is increasingly a precondition for being acquirable at all, not a nice-to-have. Third, your most motivated buyer may be the incumbent whose labor your agent replaces, not the software vendor whose product it resembles. Map that buyer early, because they'll move when they feel the threat, and right now, a lot of incumbents are starting to feel it.

Insights Most People Overlook

1. The biggest acquirers of agent companies may end up being labor-heavy services firms and BPOs, not software vendors. Everyone watches Salesforce. Far fewer watch the staffing giants and contact-center operators whose entire business model is the human labor an agent replaces. Those firms are existentially motivated and cash-rich, and a disproportionate share of strategically-priced deals will come from them.

2. Acqui-hires are deliberately mispriced to look like product deals. Because acqui-hire valuations are awkward to disclose and can spook the acquired team, incumbents (and especially hyperscalers) structure them as licensing or "talent" arrangements. The reported deal value often has almost nothing to do with the product's revenue, which means anyone valuing the GaaS market off announced deal sizes is reading a fogged mirror.

3. Data and identity acquisitions are leading indicators of agent acquisitions. Agents fail without governed data and without a way to authorize non-human actors. When an incumbent buys a data-governance or non-human-identity company, it is quietly building the foundation to acquire and deploy agents next. Track the plumbing deals to predict the headline ones.

4. The model labs becoming acquirers is the most destabilizing development in the space. When OpenAI or Anthropic buys an application-layer agent, every startup building on their API has to ask whether they're a partner or a future competitor. This chills the application layer in ways that don't show up in deal counts but absolutely show up in fundraising conversations.

5. An incumbent's defensive M&A often signals which categories are about to get commoditized. When a giant overpays to lock up a workflow, it's tacitly admitting that workflow is now automatable enough to be dangerous. Defensive buys are a map of where seat-based software is about to lose pricing power, useful intelligence whether you're building, investing, or just trying to time the market.

References

• a16z, The Economic Case for Generative AI and Foundation Models
• CB Insights, Generative AI Acquisitions Research
• Bain & Company, Insights on Generative AI and Software
• Harvard Business Review, Build vs. Buy and Technology Strategy