Maria shared the technical details and background of creating the assistant with dev.ua.
Lioncom event agency marketing director Maria Kalitsun created an AI assistant that took over the full cycle of content management on LinkedIn and became a strict deadline controller for management.
Maria shared the technical details and background of creating the assistant with dev.ua.
From others' experience to own development
The idea to automate the routine emerged after observing colleagues and attending a relevant workshop.
"The idea to create an agent came when I saw that a marketing colleague from the Select Form brand doesn't collect Instagram statistics manually — a bot does it and provides everything to her as a dashboard. I wanted that too. That same day I had a workshop from Marketinium about AI tools, where I learned a lot of cool tricks about how artificial intelligence takes over all the routine. All I have to do is teach it and sometimes manage it," says Maria Kalitsun.
The bot's tasks: from analytics to "denunciations"
The AI agent processes a large array of data and participates in content planning for LinkedIn at all stages.
The agent tracks which posts become viral on the network, and also studies the company's most successful and weakest publications.
Based on the collected analytics, the bot generates a list of topics in the form of an interactive dashboard. The CEO only needs to click the appropriate button under each idea: "Take," "Later," or "Taboo." Posts marked "Take" are automatically formed by the bot into a content plan (CP) and arranged by dates.
A day before publication, the assistant checks for ready material. If there is no text or visual — it sends a reminder. Maria uploads the finished files to the bot, and it independently saves them to disk and attaches them to the plan. On the posting day, the bot awaits confirmation and after publication provides a report in Telegram.
It is precisely because of the categorical notifications, when the CEO fails to deliver a post on time, that the team jokingly dubbed the bot "the agent who snitches on management."
Technical details: Claude, TOV, and 24 hours to launch
The agent is implemented on the basis of Claude by Anthropic. In order for the assistant to adhere to the corporate style and properly format materials, Maria uploaded the company's key documents into it.
"All skills are programmed to work in accordance with the brand's Tone of Voice — I threw it the TOV file at the very beginning. I also threw in the brandbook and obligated it to produce all visual materials strictly according to it," notes the marketing director.
According to the specialist, the basic design and setup of the bot from idea to launch took only one day. However, the work did not stop there: the agent continues to learn every day in the process of real tasks.
How it works in practice
Today the AI agent fully controls the publication schedule and promptly reports results in the Telegram work chat.
As an example of effectiveness, the bot successfully published the CEO's latest post (including a 10-page carousel document) and issued a clear update: confirmed the publication, reminded the executive about the need to respond to comments within the first hour, and recorded the status of the content plan for the day.