AI agents instead of subscriptions
In January 2026, Anthropic released Claude Cowork plugins for sales, finance, legal, marketing and support. According to The Batch, the S&P Software & Services index lost 25% between 12 January and 23 February 2026.
Investors reasoned this way: if an agent works in other tools by itself and assembles tasks from several programs, companies need fewer paid seats in each. The pressure falls on pay-per-seat pricing, not on the systems themselves: an agent still needs a CRM to record the result.
The second signal: on 9 December 2025, MCP, the protocol that connects agents to software, was handed over to the Linux Foundation. Connecting an agent to systems is becoming a standard, and it will be harder and harder to lock a customer into one agent vendor.
For Russia, the conclusion is practical. Russian CRM and accounting systems also sell seats by the number of users, and after an agent pilot some seats can go unrenewed. The analysis with numbers is in the article on agents instead of subscriptions.
Local models
On 24 April 2026, DeepSeek released V4: open-weight models under the permissive MIT license. You can use the weights in commercial products and run the model on your own server without sending your conversations to anyone.
For Russia this matters more than for Europe. If the model runs on a server in Russia, customers’ personal data goes nowhere, and the 152-FZ requirement, Russia’s personal data law, is met by the design of the system, not by a contractor’s promise.
The token price is no longer the main barrier. The barrier has moved to the process, the rules and the data: an open model does not know your prices and stock and does not decide when to hand a dialog to a person.
Your own server does not pay off for everyone: it needs GPUs and an engineer, and for a small business a Russian model through an API is often cheaper. The section on sovereign AI explains when a local model makes sense and what it costs.
Search without clicks
On 9 June 2026, SparkToro used Similarweb data to show that 68% of Google searches in the US from January to April 2026 ended without a single click. The answer on the results page is enough for the searcher.
In Yandex search, AI answers do the same. On 10 June 2026, Kommersant, citing Ashmanov & Partners, reported that the share of sites outside Yandex’s top 10 among the sources of these AI answers fell from 40% to 10%.
General “how to choose” and “what is” articles no longer bring customers: the search engine retells them. Customers come from pages with specific facts, prices, terms and conditions that search quotes with a link.
Every visitor who still reaches the website is worth more, and you cannot afford to lose them on a “we will call you back” form. What a company that lived on traffic should do is covered in the article on zero-click search.
The AI law
Russia now has a framework law on artificial intelligence, Federal Law No. 243-FZ. It came into force on 1 September 2026, and its key rules apply from 1 March 2027.
The law regulates large foundation models and introduces the terms sovereign model and national model: both require a Russian legal entity and data centers in Russia. The main duties fall on model developers, services that give access to models and large platforms.
The direction is clear: the state supports Russian models with data in Russia. For systems that already work when the key rules take effect, the law gives a grace period if their data is processed and stored in Russia.
The law is a framework, and the details will come in government resolutions. What to check in your company now is covered in the article on the AI law.
What business should do
In Russia, 53% of small and medium business owners use neural networks, against 17% a year earlier: PSB and Opora Rossii published these figures on 3 August 2026. Worldwide, 88% of organizations use AI. The gap is closing, and waiting until everyone has adopted AI costs more than starting with one process.
Five steps that follow from the trends:
1. Recount your subscriptions and software seats: which of them an agent will take over after a pilot.
2. Check where your agent’s model runs and whether you can replace it without rewriting the rules.
3. Find out where your conversations with customers are physically processed: in Russia or abroad.
4. Add pages with prices, terms and conditions to your website that search can quote.
5. If your traffic has fallen, put an agent on incoming requests that answers a visitor at once.
All four trends come down to one question: where exactly AI will bring money to your company and which solution will do it. An AI solution map answers it, and you can start with a 30-minute review.