Who’s buying AI agent startups?
In our agentic AI market deck, you will find everything you need to understand the market
SUMMARY
Enterprise software companies are buying AI agent startups most aggressively today, led by Salesforce and Workday, while cybersecurity vendors, AI infrastructure providers and well-funded agent companies are quickly joining the market.
This is already a real acquisition wave, not a handful of isolated deals. CB Insights counted roughly 80 agent or agent-infrastructure acquisitions in 2025 alone, and the 2026 transactions now cover customer service, security, marketing, coding, work management and infrastructure.
The repeat buyers are assembling product stacks rather than collecting random AI assets. Salesforce has filled specific Agentforce gaps one by one, while Workday has bought the tools to build agents, give them enterprise knowledge, connect them to outside software and deploy them in valuable workflows.
SaaS incumbents have a structural advantage because they already control the data, permissions, workflows and customer relationships an enterprise agent needs. The agent may become the new interface, but the systems underneath remain valuable and, in many cases, become even harder to displace.
The largest checks are going to products that already work in production. Moveworks had five million employee users, Fin had an installed base of more than 30,000 companies, and Sana had served more than one million users before their respective deals.
Customer service is leading because the return is easy to measure and the deployment rails already exist. That buying pattern is now spreading into marketing and sales, where agents can be tied directly to conversion, pipeline and revenue.
Big Tech’s appetite is partly hidden by deal structure. Licensing-and-hiring arrangements can transfer research leaders, technology rights and intellectual property without a conventional acquisition, so standard M&A statistics understate how much agent talent is actually being absorbed.
Agent security has become a secondary acquisition market of its own. Once agents hold credentials, call tools and take actions, identity, governance, runtime protection and behavior monitoring stop being optional add-ons.
Well-funded agent startups are becoming consolidators too. Sierra and Cognition are using acquisitions to add geography, voice, workflow expertise and interaction design faster than internal teams could build every piece.
The targets commanding attention usually own something difficult to copy: distribution, deep workflow knowledge, live infrastructure, trusted integrations or an unusually fast team. A generic horizontal agent with little proprietary data and few customers is becoming the weakest acquisition profile in the market.
This market map, featured in our agentic AI market deck, highlights top companies and startups in the agentic AI market
Is there really an AI agent acquisition wave right now?
AI agent M&A is already a real market, and the pace remains high.
CB Insights counted 782 acquisitions of private AI companies in 2025, more than 1.5 times the 2024 level. Roughly 10% involved AI agents or the infrastructure around them. That puts the agent-related total at around 80 deals in a single year, before adding the new transactions announced in 2026.
Fresh 2026 deals now span customer service, security, marketing, work management, coding and agent infrastructure. Buyers are purchasing both finished agent products and the software needed to build, connect and control them.
CB Insights also recorded 181 private-AI M&A deals in Q2 2025 and 172 in Q3, its two strongest quarters on record at the time. Agent companies kept appearing among the largest exits rather than only among small acquihires.
Around 80 agent-related acquisitions in one year is enough for us to call this a genuine consolidation market, even with a fuzzy category boundary around what counts as an “AI agent.”
| Buyer | Target | Deal value | Status | What the buyer gets |
|---|---|---|---|---|
| Salesforce | Fin | ~$3.6B | Agreed | Customer-service agents |
| ServiceNow | Moveworks | $2.85B | Completed | Employee search and workflow agents |
| Workday | Sana | ~$1.1B | Completed | Enterprise search, learning and agents |
| NiCE | Cognigy | $955M | Completed | Conversational customer-service agents |
| Okta | Permiso | ~$200M reported | Agreed | Identity security for agents and machine identities |
If you want more recent data on this point, please see our latest agentic AI market report.
Who is buying AI agent startups most aggressively today?
Enterprise software companies are currently the most consistent buyers of AI agent startups.
CB Insights named Salesforce the most active AI acquirer of 2025 with 10 AI acquisitions overall, while Workday, Meta and CoreWeave each made four. Inside that broader count, the agent deals are concentrated around a few repeat buyers. Salesforce repeatedly bought technology for Agentforce, while Workday assembled an agent builder, a recruiting agent, an AI work interface and an integration layer.
The buyer list widened in 2026. Zendesk, Asana, Genesys, MoEngage and Klaviyo all bought agent-related technology for customer service, work management or marketing. Figma also bought the team behind Bud as it pushed beyond design into AI-assisted building and prototyping.
Big Tech still makes very large bets, and infrastructure and cybersecurity vendors are becoming more active. But when we look for companies buying agents again and again, enterprise software has the deepest bench right now.
As this chart shows, and as featured in our agentic AI market deck, search interest in AI agents has been rising rapidly
Why are SaaS companies buying AI agents so quickly?
SaaS companies are buying AI agents fast because they can drop them into workflows and customer accounts they already control.
An enterprise agent becomes much more valuable once it can read company data, respect permissions and take action inside systems employees already use. Incumbents start with a real advantage. A work-management company already knows the projects and owners. A CRM already holds customer history. An HR platform already has employee records, roles and approvals.
Asana’s $75 million purchase of StackAI is a clean example. StackAI built agents that can connect to systems such as Salesforce, Slack, ERP and IT-service tools. Asana had already been pitching itself as a place where humans and agents coordinate work; buying StackAI gave those agents more ability to execute across the rest of a company’s software.
Figma’s acquisition of the Bud team shows the same logic from another angle. Bud had moved from vibe coding into a broader agent platform that could browse the web, use services and write code. Figma already owned the place where many product teams design. Bringing an agent-building team inside lets Figma move further from static design toward producing working software.
This is why the “agents will kill SaaS” story looks too simple. The interface may change dramatically, but the underlying systems, permissions and customer relationships remain valuable. Software incumbents are trying to make the agent the new way customers use their products before an outside agent takes that role.
If you want more recent data on this point, please see our latest agentic AI market report.
Is Salesforce the biggest repeat buyer of AI agent startups?
Salesforce is the clearest repeat buyer of AI-agent capabilities right now.
We can identify at least six acquisitions or agreed acquisitions since 2025 that directly strengthen Agentforce: Convergence.ai, Spindle AI, Qualified, Cimulate, Momentum and Fin. CB Insights separately ranked Salesforce as the most active AI acquirer of 2025 with 10 AI deals overall, so the agent push sits inside a much broader acquisition program.
Those six purchases cover very different jobs. Convergence brought agents that can navigate changing websites and multi-step digital tasks. Spindle added multi-agent analytics and observability. Qualified added autonomous B2B marketing. Cimulate added intent-aware commerce search. Momentum turns conversations from tools such as Zoom and Google Meet into structured context that agents can use. Fin adds a mature customer-service agent with more than 30,000 companies in its installed base.
Salesforce has been buying the gaps around Agentforce one by one. Action, measurement, marketing, commerce, context and customer service have each been added through separate deals.
The Fin deal also shows how high Salesforce will go for a proven product. Salesforce agreed to pay about $3.6 billion even though Agentforce already had service-agent capabilities. At the time, Salesforce said Agentforce had reached $1.2 billion in annual recurring revenue, up 205% year over year. Buying Fin looks like a decision to accelerate deployment and product maturity rather than spend several more years trying to reproduce what Fin had already learned in production.
This chart, included in our agentic AI market deck, illustrates yearly VC funding for agentic AI startups
Why did Workday buy so many AI agent companies in a few months?
Workday bought four AI companies in less than a year to cover the main pieces an enterprise agent needs.
Flowise gave Workday a low-code agent builder. The open-source project had already passed 42,000 GitHub stars and was processing millions of chats and workflows when Workday bought it. Paradox added a recruiting agent built for high-volume hiring; Workday said Paradox had powered more than 189 million AI-assisted candidate conversations.
Sana then added the employee-facing layer. Workday paid roughly $1.1 billion for the enterprise search, learning and agent company, whose products had served more than one million users across hundreds of enterprises. Finally, Pipedream added more than 3,000 pre-built connectors so agents could pull data and execute tasks in outside applications. Workday later confirmed that the Pipedream acquisition had closed.
Put together, those purchases cover building an agent, giving it enterprise knowledge, connecting it to other software and deploying it into a specific high-value workflow. Few buyers have assembled the pieces this deliberately.
Workday also has a distribution base of more than 11,500 customers, including more than 7,000 core HR and financial-management customers. That combination explains the strategy better than any generic claim that Workday “wants more AI.” The company already owns sensitive systems of record; the acquisitions help turn that data into actions employees can trigger through agents.
Why did ServiceNow spend $2.85 billion on Moveworks?
ServiceNow bought Moveworks because it already had real enterprise adoption at a scale that would take years to reproduce.
When ServiceNow announced the deal, Moveworks was used by more than 350 large enterprises and over five million employees, including roughly 10% of the Fortune 500. Nearly 90% of customers had deployed it across their full workforce. Those deployment numbers make Moveworks much more valuable than a promising agent demo with a few design partners.
ServiceNow already had AI, workflow automation and its own agent ambitions. What Moveworks brought was a familiar employee interface sitting on top of many corporate systems. Around 250 customers already used both companies, which also lowered the integration risk.
Enterprise agents need permissions, integrations, trusted data and habitual use. A buyer can build models and interfaces internally, but reproducing five million deployed users and hundreds of large-company rollouts takes much longer.
Moveworks gives one of the clearest valuation lessons in the market: once an agent becomes the place where employees routinely ask for help and start work, the interface itself can become valuable enough to buy.
This chart, included in our agentic AI market deck, shows how Cognition is positioned in agentic AI
Why do buyers keep targeting AI agents that talk to customers?
Customer-facing AI agents are getting bought first because companies can measure the ROI quickly and push them to thousands of existing customers.
Support teams track ticket volumes, resolution rates, handling time and cost per contact, so buyers can quickly see whether an agent is saving money or improving service. The category also had years of chatbot and contact-center infrastructure in place before the current agent wave.
Zendesk’s acquisition of Forethought is a good example. Forethought had won TechCrunch Battlefield back in 2018, years before the current agent boom, and by 2025 said it was supporting more than one billion customer interactions per month. NiCE paid $955 million for Cognigy, whose conversational AI was already used by large companies including Mercedes-Benz, Nestlé and Lufthansa Group. Genesys later acquired Pinkfish, which had built more than 500 integrations and roughly 25,000 Model Context Protocol server tools for connecting customer requests to enterprise actions.
The buying has spread from support into marketing and sales. Salesforce completed its acquisition of Qualified, whose AI worker engages website visitors and qualifies inbound buyers. MoEngage bought Aampe in an all-cash deal worth tens of millions of dollars according to TechCrunch; Aampe assigns an autonomous decisioning agent to individual customers and had grown annual recurring revenue by 150% over the prior year. MoEngage already serves more than 1,350 consumer brands across 75 countries, giving it an immediate route to distribute that technology.
Klaviyo’s latest move adds another piece of evidence. It agreed to acquire Agency and put founder Elias Torres in charge of product, with the goal of accelerating Klaviyo’s Composer and Customer Agent products across roughly 200,000 businesses. Salesforce also bought Momentum to turn sales conversations from Zoom, Google Meet and similar tools into context that agents can act on.
Customer-facing agents have a simple advantage in M&A: their output can be tied to costs, conversion, pipeline or revenue. That makes it much easier for a buyer to decide whether the product is worth integrating and scaling.
Is Big Tech actually leading AI agent acquisitions?
Enterprise software companies currently buy whole AI-agent companies more often than Big Tech. The tech giants lean heavily on licensing-and-hiring deals.
Meta looked like the strongest counterexample when it agreed to buy general-purpose agent company Manus for roughly $2 billion to $3 billion after Manus said it had crossed $100 million in annual recurring revenue. The deal later became a geopolitical problem. Chinese regulators ordered the acquisition unwound, and by June 2026 Meta had reportedly begun separating its systems from Manus and moving toward a full divestiture.
Windsurf shows why Big Tech can be hard to count. OpenAI spent months discussing a roughly $3 billion acquisition of the AI coding company. After that deal fell apart, Google paid about $2.4 billion for non-exclusive technology rights while hiring Windsurf’s CEO, co-founder and selected researchers into DeepMind. Three days later, Cognition agreed to acquire the remaining Windsurf product, IP, brand and employees.
One startup effectively produced value for two different buyers, with Google taking the research leadership and Cognition taking the operating business. Regulators have since been paying closer attention to licensing-plus-hiring arrangements because they can transfer much of a startup’s value without a conventional change of control.
Meta has also bought smaller AI teams such as PlayAI and Moltbook, while the large labs keep recruiting aggressively from startups. Big Tech clearly pushes up the price of scarce agent talent. But on repeat purchases of whole enterprise-agent products, Salesforce, Workday, ServiceNow, Zendesk, NiCE, Asana and Genesys are doing more of the straightforward buying.
If you want more recent data on this point, please see our latest agentic AI market report.
This chart, included in our agentic AI market deck, illustrates yearly funding for agentic AI startups
Are AI agent startups starting to buy each other?
Yes. Well-funded AI agent startups are now becoming buyers themselves, and Sierra is the clearest example.
Sierra made three public acquisitions by April 2026: Japanese enterprise AI company Opera Tech, voice-agent startup Receptive AI and French workflow startup Fragment. It then acquired Takeoff in July. Takeoff said it had gone from zero revenue to nearly an eight-figure annual run rate in 2026 with only three people.
Takeoff is useful here because it was hardly a distressed acquihire. The startup said it had several seven-figure contracts and enough capital to keep operating independently. Joining Sierra gave it a larger deployment machine and access to an agent company that had already reported more than $150 million in ARR earlier in the year.
Cognition is following a similar route in coding. It acquired consumer assistant Poke in July 2026 in a deal valuing the startup in the low nine figures. Poke had handled more than 100 million user messages in its first three months, and Cognition wants to bring its interaction style into Devin.
We now have agent startups acting like mini consolidators. The best-capitalized ones can use M&A to add geography, voice, workflow expertise, user experience or specialized runtimes faster than building every capability from scratch.
Why are cybersecurity companies buying AI-agent startups now?
Cybersecurity companies are buying agent-security startups fast because AI agents now hold credentials, call tools and take actions inside company systems.
The latest example is Fortinet’s acquisition of Virtue AI. Virtue AI tests autonomous agents across more than 50 sandboxed environments and 14 high-stakes domains, scans MCP tools and source code, and can block malicious tool calls at runtime. Gartner expects spending on tools for securing AI ecosystems and agents to rise from $2.8 billion in 2026 to $16.4 billion by 2030, according to the figure cited by Fortinet.
Fortinet joins a crowded buyer list. Okta agreed to acquire Permiso for just under $200 million according to TechCrunch, with a focus on AI agents and other machine identities. SailPoint agreed to buy Entro for non-human identity and credential security. Palo Alto Networks completed its purchase of Portkey, an AI gateway already processing trillions of tokens per month. Proofpoint bought Acuvity for AI governance and runtime protection. CrowdStrike bought Pangea for about $212 million in total cash consideration according to its annual filing.
Six large security vendors have converged on a very young software problem. They sell different pieces of cybersecurity, yet they are all buying around identity, governance, runtime control and agent behavior. Agent security has already become one of the clearest secondary M&A markets created by the agent boom.
| Security buyer | Target | What the acquisition adds |
|---|---|---|
| Fortinet | Virtue AI | Agent validation, runtime protection and MCP scanning |
| Okta | Permiso | Identity security for AI agents and machine identities |
| SailPoint | Entro | Non-human identity and credential discovery |
| Palo Alto Networks | Portkey | AI gateway, routing and governance |
| Proofpoint | Acuvity | AI visibility, governance and runtime controls |
| CrowdStrike | Pangea | AI detection and response across agents and interactions |
This chart, included in our agentic AI market deck, compares the main business model options for autonomous AI agent platforms
Why are AI infrastructure companies buying agent tools?
AI infrastructure companies are buying agent tools because raw compute alone leaves too much value in the software above it.
CoreWeave bought OpenPipe, a platform for training agents with reinforcement learning. A few weeks later, CoreWeave launched a serverless reinforcement-learning product combining OpenPipe with its earlier Weights & Biases acquisition. Better agent-training tools can feed more training and inference work back into CoreWeave’s cloud.
Nebius bought Tavily for another missing piece: real-time web search. Tavily gives agents current information rather than leaving them dependent on static model knowledge. Nebius has already integrated Tavily into its cloud as part of a broader stack for building and running autonomous agents.
DigitalOcean acquired Katanemo Labs to add orchestration, observability and safety primitives above raw inference. TrueFoundry bought Seldon AI to strengthen the control plane used to connect, observe and govern agentic applications.
All four buyers are moving upward from raw infrastructure into the software that agents need. Owning that software gives them a better chance of making money above the increasingly commoditized compute layer.
What makes an AI agent startup worth buying?
An AI agent startup becomes worth buying when it has something the buyer cannot quickly copy: distribution, workflow knowledge, infrastructure or a team that ships unusually fast.
Scale helps. Sana’s products had served more than one million users before Workday bought the company. Forethought was handling more than a billion customer interactions per month. Poke passed 100 million user messages within three months. Those numbers show that people are actually using the product, which removes a large part of the risk for an acquirer.
Specialized knowledge can be just as valuable. Paradox had handled 189 million candidate conversations and understood high-volume recruiting. Pinkfish brought hundreds of enterprise integrations. Tavily had built search infrastructure specifically for agents that need live web information. These capabilities take more than a model API and a polished interface.
Then there is speed. Takeoff reached nearly an eight-figure annual run rate with three people before Sierra bought it. In a market where underlying models can change every few months, acquiring a tiny team that has already found a working architecture can be cheaper than giving an internal product group another year.
The weakest acquisition profile is increasingly the generic horizontal agent with little proprietary data, few customers and no difficult technical layer. Buyers can reproduce that product more easily. The bigger checks are going to companies that already control something the acquirer would struggle to build quickly.
If you want more recent data on this point, please see our latest agentic AI market report.
This chart, featured in our agentic AI market deck, shows the share of revenue generated by each customer segment in the agentic AI market
Where will the next AI agent acquisitions happen?
The next AI-agent deals will probably cluster in security, customer service, sales and marketing, coding infrastructure and specialized vertical agents.
Security has accelerated the most lately because large vendors are racing to cover machine identities, MCP connections and autonomous actions. Customer service has the clearest proof that agents can be sold and measured at scale. Marketing is catching up as vendors connect agent decisions directly to conversion and revenue.
Coding will probably keep producing unusual transactions because a handful of researchers or product leaders can represent a large share of a startup’s value. Infrastructure should keep consolidating around search, memory, evaluation, reinforcement learning, observability and orchestration as cloud companies try to own more of the production stack.
The next interesting area is vertical agents. Recruiting has already produced acquisitions, and legal, finance, healthcare, procurement and industrial operations all contain workflows where domain knowledge, permissions and integrations matter more than having the newest general-purpose model. Those characteristics make a specialized startup easier for an incumbent to defend after acquisition.
| Category | What we’re seeing now | What buyers are trying to own |
|---|---|---|
| Customer service | High activity | Autonomous resolution and contact-center distribution |
| Agent security | Rising very fast | Identity, governance and runtime control |
| Sales and marketing | Rising | Revenue-generating decisions and customer context |
| Coding | High but structurally unusual | Talent, IP and developer workflows |
| Agent infrastructure | High | Search, training, orchestration and observability |
| Vertical agents | Earlier | Deep domain workflows and proprietary context |
So who’s buying AI agent startups today?
Today, enterprise software companies are the main buyers of AI agent startups; cybersecurity, AI infrastructure and well-funded agent companies are catching up fast.
Salesforce is the clearest repeat acquirer, while Workday has one of the easiest agent-buying strategies to read: build, connect and deploy agents around HR and finance. ServiceNow, Zendesk, NiCE, Asana, Genesys, Klaviyo and MoEngage show that the behavior has spread well beyond the two largest enterprise-software names. Lately, the security side has accelerated even faster, with Okta, Fortinet, Palo Alto Networks, Proofpoint, SailPoint and CrowdStrike all buying technology designed for a world where software agents hold credentials and take actions on their own.
Big Tech still sets eye-catching prices, especially around scarce research talent and coding. The attempted Manus acquisition and other licensing-and-hiring deals also show why conventional M&A statistics miss part of the story. Large technology companies can absorb the people and IP without buying the whole corporate entity.
Our final judgment is straightforward. The current AI-agent acquisition market is being driven by companies that already own something agents need: customers, enterprise data, workflows, compute, security control points or distribution. The valuable targets increasingly fill a specific missing layer and have evidence that the layer works in production. These days, a generic “AI agent” label is cheap. A deployed workflow, a trusted control point or a product people already rely on is what buyers are actually paying for.
If you want more recent data on this point, please see our latest agentic AI market report.
This chart, included in our agentic AI market deck, shows how autonomous AI agent platform technology has evolved over time
OUR METHODOLOGY
There is no single clean dataset that answers who is really buying AI agent startups today. The category is still evolving, buyers use different deal structures, and a few headline transactions can easily distort the picture. We therefore broke the question into recent acquisition activity, repeat buying behavior, buyer type, deal structure, strategic fit, target traction and the parts of the agent stack where activity is concentrating.
For each dimension, we prioritized the freshest available evidence: announced and completed transactions, company disclosures, operating metrics and market-level M&A data. Older examples were used mainly when they helped establish a pattern that was still visible in 2026.
We included acquisitions where an AI agent or the infrastructure needed to build, connect, train, observe, govern or secure agents was central to the buyer’s rationale. We separated conventional acquisitions from licensing, hiring and technology-transfer structures because those deals reveal demand for talent and intellectual property but do not represent the same kind of buying behavior as acquiring an operating company.
No single metric was treated as the answer. Deal count shows activity, deal value shows conviction, repeat acquisitions reveal strategy, and product or customer traction helps explain what buyers are paying for. We assessed those measures together, buyer by buyer and deal by deal.
Transaction status and pricing were kept distinct. Announced deals are labeled as agreed, closed deals as completed, and values described as approximate or reported remain marked that way. We did not infer prices where the parties did not disclose one.
The forward-looking conclusions are based on where the evidence is converging now: repeated purchases by multiple buyers, clear strategic gaps incumbents are trying to fill, accelerating activity in specific parts of the stack, and proof that acquired products already work in real enterprise environments. The aim is to identify where the acquisition logic is strongest, not to predict individual deals.
Key sources used for this analysis include CB Insights’ State of AI 2025, CB Insights’ State of AI Q3 2025, Salesforce’s announcements for Convergence.ai, Spindle AI, Qualified and Fin; Workday’s announcements for Flowise, Paradox, Sana and Pipedream; and primary acquisition announcements from ServiceNow on Moveworks, NiCE on Cognigy, Asana on StackAI, Zendesk on Forethought, Genesys on Pinkfish, Klaviyo on Agency, Okta on Permiso, Proofpoint on Acuvity, Palo Alto Networks on Portkey and Nebius on Tavily.
In our agentic AI market deck, we identify pain points entrepreneurs should prioritize
Related blog posts
- Agentic AI M&A: what is happening now?
- What are the top startups in the agentic AI market?
- Is AI agent traffic klling the internet?
- The startups that have raised the most funding in the agentic AI market
