Yandex has just given its AI assistant a pair of hands. The Russian search giant opened early access to Deystvuy! — the name translates roughly to "Take action!" — an agent that lives inside Alisa AI and doesn't stop at giving advice. Instead of telling you which salon has a free slot on Saturday, it goes ahead and books it.

That's a bigger shift than it sounds. For the past couple of years, most AI assistants have been very good at talking and fairly bad at doing. Yandex's bet is that people will notice the difference when an assistant actually finishes the job.

What happened

Yandex announced the agent in early access, which means it's being handed out gradually rather than switched on for everyone at once. According to the company's press service and coverage from iXBT, RB.ru, and CNews, it handles errands like booking a beauty salon appointment, hunting down plane tickets, calling a doctor and ordering flowers.

The detail that matters most is transparency. Every step the agent takes shows up in the chat as it happens, so you're not handing over your card details and hoping for the best. You can watch it work, stop it halfway, or point it in a different direction. Before anything consequential — a purchase, a confirmed booking — it checks with you first.

What the agent actually does, in plain terms

Think of the difference between a friend who texts you a restaurant's phone number and a friend who calls the restaurant and gets you a table. The agent works with websites and files, so it can fill in forms, compare options and follow a task through to the end — the kind of chore that quietly eats twenty minutes of your evening.

Plugins and MCP, explained without the jargon

Yandex says the agent's abilities can be extended with instructions and various plugins, including MCP.

MCP stands for Model Context Protocol: an open standard for plugging outside tools into an AI model. Picture it as USB-C for AI. Instead of every assistant needing a bespoke cable for every service, one connector fits all — so the agent can be wired up to more tools over time without Yandex rebuilding it from scratch.

"Instructions" are the simpler half. You write plain-language rules for how you want it to behave, like "always pick the cheapest direct flight" or "never book anything over my budget without asking me first."

What it means for you

You don't need to be a developer to care about this. Here's where it lands in ordinary life.

At home

The flower-ordering example is the giveaway. These are the errands that pile up precisely because each one is too small to schedule: the birthday bouquet, the salon appointment, the doctor's call you keep putting off. An agent that handles three of those a week gives you back an evening. If you'd like to see what else is free before you commit to anything, our free AI tools collection is a reasonable place to start browsing.

At work

Booking travel, chasing schedules, pulling details off websites into a file — that's the shape of a lot of office admin. Because the agent works with sites and files, the "open this, copy that, paste it here" loop is exactly what it's built for. You stay in charge of the decisions; it does the clicking.

For your business

Small teams burn hours on admin that never shows up in a strategy meeting. An agent that can research, compare and confirm — with a human approval gate on anything that costs money — is a bit like a junior assistant who doesn't need a desk. For solo founders and one-person shops, that's the most interesting part of the launch.

For studying

Comparing course options across several sites, gathering material into one folder, keeping track of what you've already checked — the tedious half of research. The agent won't write your essay, and it isn't pretending to. It shortens the scavenger hunt that comes before the thinking.

For creative work

Nobody's claiming this thing designs a logo. The win is offloading the unglamorous half of creative work: sourcing references, chasing suppliers, booking the studio, ordering the props. That's time that goes back into the actual making.

For extra income

If you bill by the hour, unbilled admin is the enemy. Freelancers who cut their research and booking time can take on more work without adding hours to the week — a small edge, but an edge.

How to try it right now

Here's the shortest path, free option first.

1. Open Alisa AI — Yandex's assistant, in its app or on the web. The agent is built into Alisa AI rather than sitting in a separate product.

2. Look for Deystvuy! and join early access. If you don't see it, you're in the queue — access is being rolled out in waves, not all at once.

3. Start with something small and low-stakes. Ask it to find three flights for a specific date and show you the prices, rather than "plan my whole holiday." You'll learn its habits fast on a task where a mistake costs nothing.

4. Watch the chat. Approve the steps that matter, and stop it when it drifts. That's the whole safety model — it's designed to be supervised, not trusted blindly.

5. Add an instruction. Once you've seen it work, write one plain rule ("cheapest direct option first") and watch how the behavior changes.

6. Explore plugins later. That's the MCP layer — useful once you know what you actually want it to do.

On cost: Yandex hasn't published pricing for the agent, and nothing in the announcement suggests a separate fee for early-access users. Treat it as a free trial of something that will probably get a price tag later.

Upsides and what changes

Three things stand out. First, the move from advice to action — an assistant that books, orders and confirms is a different category of product from one that writes you a to-do list. Second, the visibility: showing every action in the chat is the right call, because trust in autonomous software is earned in small increments, and people need to see the machine think. Third, the extensibility — instructions plus MCP plugins mean the agent gets more capable over time without you switching tools.

The bigger picture: this raises the bar for every assistant. Once one of them reliably books your table, "here are three links to try yourself" starts to feel like a shrug.

Limitations

Be clear-eyed about this one. It's early access, so availability is patchy — you may not get in today, and features can change under you. The agent still asks for approval on important steps, which is sensible but also means it can't fully run your errands while you're asleep. MCP plugins require setup, and anything that leans on external websites can break when those sites change. There are no independent, long-run tests yet; the capabilities described so far come from Yandex's own announcement and press coverage, not from months of user experience. The product is Russian-language and aimed first at Yandex's home market, so if you don't read Russian you'll hit friction. And you are, ultimately, letting a piece of software act on your behalf with your accounts and your bookings — start small for a reason.

Conclusion

Yandex's Deystvuy! agent is a signpost more than a finished product: the assistant era is moving from "here's what you could do" to "I did it, check the chat." The recipe — action, visible steps, human approval, pluggable tools — is one the rest of the industry will almost certainly copy. One action for today: open Alisa AI, ask it to handle one real, small errand you'd normally do yourself, and watch how it works before you decide how much you trust it.