AI agents for business
An agent is a teammate that never gets tired of the routine: every day it gathers data, sorts it, answers and prepares materials, and it leaves the decisions to people. We don’t sell agents off the shelf. First we find the process where an agent saves hours or makes money, then we build and launch it. Below are agents already at work for us and for clients.
01
Routine to the agent, decisions to people
First we find the process where an agent pays for itself, then we build it on your data and launch it. All our agents follow one rule: anything about customers, money or deadlines goes through a person.
01
Find the process
Where hours go or money leaks: requests, chats, data, reports.
02
The agent does the routine
Reads, sorts, answers and prepares materials around the clock.
03
A person decides
Customers, money and deadlines only with a person’s sign-off.
02
A concierge that plans your night on request
The problem: a city has thousands of events, and people get tired of choosing and end up staying home. On ZDES Events you describe the night you want in your own words, and the concierge builds a plan from real events and venues, explains each pick and takes you through registration to a pass at the door.
Behind it, agents work through Telegram channels, venue websites and listings every night, check the city and the organizer, and drop ads and duplicates. The database holds more than 10,000 venues. The same approach works on any company’s catalog: ticketing, retail and marketplaces, travel, loyalty programs.
03
A producer agent that won back half the budget
The problem: on a production for a Swiss watch brand, revisions and requests landed in the chats every day, and work beyond the agreed scope got lost in the thread. The agent ran the client chat in English and the team chat in Russian, logged revisions, deadlines and tasks, and checked every new request against the approval stages and the price list.
The result: a 22-line register of extra work worth €5,444, about half the project budget, which went back into the estimate. The agent handled routine like status updates and file requests on its own, while anything about the client, scope, deadlines or money went through a person.
01
Two chats, two languages
The client writes in English, the team in Russian, and the agent runs both.
02
An extra-work log
Every request beyond the scope is checked against the stages and price list, then logged.
03
Money goes through a person
Status updates and file requests go out on their own; billing the client needs a sign-off.
04
A newsroom without a staff of editors
The problem: publish tech and AI news every day without hiring a newsroom. ZDES Media’s agents read sources several times a day, pick the stories worth a post, write a full article for the site and a teaser for Telegram and Dzen, and make the cover. That’s about a dozen posts a day plus scheduled issues.
The same setup can run a company blog, internal digests for your team, or market and competitor monitoring.
01
Reads the sources
Telegram channels, RSS feeds and websites, several times a day.
02
Writes and designs
An article for the site, a teaser for Telegram and Dzen, a cover.
03
Publishes on schedule
About a dozen posts a day and an evening recap.
05
An agent that takes a spot from brief to video
Our storyboard generator writes the script, builds the moodboard and storyboard, and marks which shots to film live, which to composite and which to make with AI. On our test film the whole generation took 40 minutes: 7 scenes and 19 shots. The producer gets a shooting plan, and the client sees the picture before the budget is spent.
06
An agent that illustrates posts by itself
For TON’s crypto media we trained a model on a meme character and built an agent that reads a Telegram post and draws an image for it with the same character. A steady stream of content with no hand-drawing for each post.
One-off or on a membership
One-off
One task from the catalog. Final price, two rounds of revisions.
- Scripted AI assistant · up to 5 business daysAn assistant for one flow, with a knowledge base₽165K
- AI agent or internal tool · 2–3 weeks, in stagesScope, prototype, build and rolloutfrom ₽500K
- Working product prototype · 2–3 weeks, in stagesFlows, interface, build and launchfrom ₽550K
On a membership
A month of work across the whole track for one price. A producer runs your tasks, and an art director reviews each one before delivery.
from ₽210K a month
What a month can look like
- 5 quick tasks: screen reviews, UI fixes
- 2 week-long tasks: a flow prototype and a live landing page
Ordered one by one, this would cost ₽585K