AI-first and AI Agents: Vlad Umnitsyn’s Talk

Vlad Umnitsyn, CEO of Smart Dynamics and owner of Kadam, spoke at MAC Affiliate Conference in Yerevan and shared how the company moved from using AI tools for individual tasks to building a full AI-first approach over the past six months and how this shift has affected the way the business operates.

The recording of the talk is now publicly available on YouTube, where you can watch Vlad’s full presentation in Russian.

What the talk was about

Vlad opened with a simple but revealing question to the audience: who spends more than $1,000 per month on AI? And who spends more than $5,000?

For context, his personal Cursor expenses alone reached $13,000 in April. In the talk, Vlad explained why he sees this not as a cost, but as an investment in speed, experimentation, and team efficiency.

He then broke down the difference between LLMs and AI agents, and explained why agents are becoming the next major step for businesses. A separate part of the talk focused on MCP — Anthropic’s Model Context Protocol. This standard allows AI agents to connect to external services and work with real data, rather than simply answering questions in a chat.

Vlad also drew a parallel with the Mobile First era: in the 2010s, companies that underestimated the mobile trend quickly lost ground. Today, a similar moment is happening with AI. Companies once again face a choice: integrate the new technology into their processes or watch others move faster.

How AI was integrated into work processes

One of the key cases in the talk was the company’s internal AI environment. Using MCP servers in Cursor, Vlad brought together a task tracker, messengers, databases, and other work tools in one place.

What used to require manually collecting information from Jira, chats, and different systems could now be delegated to an agent. This included preparing for a 1:1 with an employee, analyzing tasks, gathering project context, or quickly prototyping a new idea.

Previously, these processes could take hours or even weeks of coordination. Now, the first working prototypes can appear in 10–15 minutes.

A central part of the talk was the story of Woodman, the company’s internal AI bot.

The idea came up over the weekend: what if an agent had access to the source code and production databases, so it could answer technical questions without distracting developers? The first prototype was built in one day.

The bot was given a name and an avatar, added to a work thread, and the team started using it on their own — without instructions or a separate onboarding process.

Later, a role-based security system was added to Woodman: the bot could only see the data that each specific employee was allowed to access. Once it was opened to the entire company, use cases quickly expanded far beyond technical questions.

Within a month, Woodman had already been connected to the core infrastructure. Product teams began using it to formulate business requirements, and feature usage analytics became part of the regular workflow.

The results in numbers

In its first 67 days, the bot processed 11,700 employee requests. According to the team’s calculations, the same amount of manual work would have required around nine person-years — or roughly 100 people when taking the supporting structure into account.

Announcement of Kadam’s MCP server

The final part of the talk focused on the launch of Kadam’s public MCP server for clients. Kadam clients can now manage advertising campaigns, upload creatives in bulk, and work with the network through an AI agent — whether Claude, Cursor, or another tool. Everything is done through a single chat.

Watch the full talk via the link: the recording also includes a demo of the MCP server and a QR code leading to a landing page with setup instructions.