The market in numbers
Agents, which are programs that carry out a chain of actions in a company’s work systems on their own, have moved from pilots to daily work. In the KPMG survey (July–August 2026, 314 executives of US companies with revenue above $1 billion), 62% of companies were building or deploying AI agents, against 53% a quarter earlier.
The share of systems with several agents grew in the same quarter from 6% to 25%. Companies are moving from one assistant to a chain of agents that pass work to each other.
Agents also change software buying. According to McKinsey, 32% of organizations gave up buying at least one product because they can build it themselves with agents.
Venture money flows to AI as a whole. In the third quarter of 2026, AI startups received $102 billion, which is 64% of world venture investment of $159 billion in the quarter.
Deals
The largest agent round of the year: in September 2026, Cognition, the maker of the coding agent Devin, raised $2 billion at a valuation of $48 billion, almost twice the May valuation ($26 billion).
Cognition’s annual revenue grew from $492 million to almost $900 million in four months. Investors pay for an agent that companies already buy.
In January 2026, Decagon, a maker of AI agents for customer support, raised $250 million and tripled its valuation to $4.5 billion.
In September 2026, Harvey, an AI platform for lawyers, raised $550 million at a valuation of $15.5 billion. 80% of the 100 largest US law firms use it.
The large purchases of the third quarter were services through which companies reach AI models: Stripe bought OpenRouter for $7.5 billion, and Nvidia offered to buy Hugging Face for $12.9 billion.
All rounds with amounts and sources are collected in the deal tracker.
Russia
Kama Flow estimates the potential of the AI agents and process automation segment in Russia at $940 million by 2030. It is second in their list after industrial AI ($1.1 billion).
By the same estimate, all accessible segments of the AI market give more than $6.8 billion by 2030. An independent AI market needs about $1 billion of private venture money a year.
Actual venture sums are smaller so far. By Venture Guide, AI and machine learning raised $32 million in 33 deals in the first nine months of 2026, first place by volume and by deal count. The data is preliminary, as of 4 October 2026.
Corporate software, the niche closest to agents in the same summary, took $19 million in 20 deals in the same period.
For a Russian company, this means agents for its own processes must be built on its own budget: venture money is scarce, while demand, by Kama Flow’s estimate, is large. The overall picture for the country is on the page about AI investment in Russia.
Where it pays back
Agents have an effect, but it is often not measured. In the CB Insights survey (February 2026), 80% of companies make AI agents a priority, but 40% cannot measure the payback or do not know it.
Those who measure look at three metrics: productivity (63%), cost savings (58%) and time saved (58%). In the KPMG survey, 55% of companies name productivity growth as the effect.
The deals above show where agents already have paying clients and mass use: in programming, Cognition’s revenue is almost $900 million a year, and among lawyers, Harvey works in 80% of the 100 largest US law firms.
The rule for a company: an agent pays back in a process where you can count hours, requests or errors before and after launch. How to estimate the return before you buy is described on the page about calculating the effect of AI.
What a company should do
If you are deciding whether to invest in AI agents, start with the metric, not with the choice of a supplier. According to CB Insights, 40% of companies cannot measure the payback of agents, and an investment without a metric joins that group.
1. List the processes where staff repeat the same actions: handling requests, answering clients, reconciling documents.
2. For each, measure the starting point: hours per month, number of requests, share of errors.
3. Choose one process for a pilot and write the metric into the contract.
4. Check where the agent stores data: under 152-FZ, Russia’s personal data law, personal data of Russian citizens is stored in databases located in Russia.
5. Before you buy a subscription, compare it with your own agent: 32% of organizations in the McKinsey survey already dropped at least one product for this reason.
This order is built into the AI roadmap: the processes where an agent pays back, the price of the solution and the effect in money for each. What an AI agent is and how it differs from a chatbot is explained in a separate article.
Where AI money goes beyond agents is shown on the page about AI investment trends.