Why 2027’s AI acquisitions will target companies run by agents

In 2026, acquirers are buying the builders.

Most AI acquisitions today buy a capability. BCG reviewed 11,372 AI-linked acquisitions made between January 2020 and August 2026. Among the deals made by AI companies, 95% either added a capability or extended a product. Consolidation barely registers: 23 deals over the whole period.

Capability deals form the largest group. BCG describes their targets as small, often pre-revenue and “valued for what they can do rather than what they earn.” The buyer is paying for technology or talent it cannot build fast enough on its own.

This year’s deals follow the same pattern. In January, Aptean acquired OpsVeda to bring agentic orchestration to its Logility supply chain platform.

Buyers have also become more selective. Telegraph Hill Advisors, a San Francisco investment bank that advises founder-led technology companies, sets out what strategic acquirers are paying for in 2026. In short, acquirers want agents already running in production for paying customers who stay. AI branding without that evidence, and systems that only work in their founder’s hands, are discounted.

In 2026, the premium goes to the companies that build agents, provided the agents already work.

The next buyers are already visible.

A second kind of buyer has appeared. It does not buy the company that makes agents. It buys a traditional business and runs it with them.

General Catalyst calls this the AI-enabled roll-up. The firm backs founders who build AI for one services industry, then funds them to acquire and operate companies in that industry. Its case rests on scale: US services industries generate more than $6 trillion a year, against about $370 billion for the entire US software market. By August 2025, one of these companies, Long Lake, had acquired 18 businesses.

In May 2026, Long Lake agreed to acquire American Express Global Business Travel for about $6.3 billion in cash, with support from General Catalyst and Alpha Wave. The $9.50 per share offer was a 60.2% premium to the previous close. The plan is to pair Long Lake’s applied AI with Amex GBT’s customers and marketplace.

These buyers are not paying for agents. They are paying for businesses they believe agents can run better. For now, they take on the hardest part themselves: turning a company that runs on people into one that runs on agents.

Anyone can buy agents. Few companies run on them yet.

Part of the builder premium comes from scarcity: few teams can make agents that work. That scarcity is fading. Gartner predicts that 33% of enterprise software applications will include agentic AI by 2028, up from less than 1% in 2024. In McKinsey’s 2026 State of AI survey, 32% of respondents say their organization decided against buying at least one software product or feature because it could build it with agentic coding tools.

Running a company on agents is a different matter. In the same survey, even scaling agents in a single function remains a minority practice: 40% of respondents at organizations with more than $1 billion in revenue report doing so, and the share at smaller organizations has stayed flat at 22%. Just 37% of respondents attribute any EBIT impact to their use of AI, about the same share as a year earlier. Gartner, for its part, expects more than 40% of agentic AI projects to be canceled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls.

That gap is the core of our thesis. Buying agents is becoming easy. Making them carry a company’s daily operations, profitably, is not, and the companies that manage it will stay rare.

Buyers pay for what is scarce. By 2027, that will be a business that already runs on agents.

In 2027, buyers will ask for proof of autonomy.

The discipline of 2026 will carry over. Gartner already warns about agent washing, where existing chatbots or automation tools are rebranded as agents. Expect the same skepticism toward businesses that call themselves autonomous. A claim of autonomy will be worth little without evidence a buyer can test.

This is how we expect the two waves to differ:

The evidence a buyer can test is mostly operational:

• Owner hours: how much time the owner and any staff spend each week, and on what.
• Interventions: a log of when a human had to step in, why, and how often.
• Margins: earnings after model, hosting and software costs, not before them.
• Track record: how many months the system has run without a rescue.
• Dependencies: the providers, accounts and people the operation relies on.

Autonomy also addresses a discount buyers already apply to founder-dependent companies. We covered what makes an autonomous business transferable in a previous article.

If you own a business, 2027 starts now.

Evidence of autonomy takes time to build. A buyer will want months of history, not a demo recorded last week. If you might sell in 2027 or 2028, the record starts now.

Begin with measurement. Track your hours and every intervention before you automate further, so the improvement shows. Record margins before and after each workflow moves to agents, net of the new costs. Document what each agent does, what it is allowed to do and who takes over when it fails. Where a single model provider or a single person holds the operation together, add a fallback.

Our guide to automating before you sell covers where to start, and the valuation calculator shows how earnings translate into a price. When the record is ready, list your business with the evidence attached.

If you are buying, look for the same evidence. On our listings, owner involvement and autonomy sit alongside revenue, so you can compare businesses before you request private details.

This is a forecast, not a certainty.

We could be wrong about the timing. Agents still make mistakes, model prices change and regulation may slow some uses. The shift from builders to operators could take longer than a year, or reach some industries well before others.

Nor is a premium guaranteed. An autonomous business still needs demand, customers who stay and a price its cash flow can justify. Automation that saves hours but loses customers destroys value.

The direction is what we are confident about. In 2026, acquirers are paying for the ability to build agents. As that ability spreads, we expect the premium to move to companies that have turned agents into dependable operations and can prove it.

The next wave of AI acquisitions will be less about who makes agents, and more about who runs on them.