We have gathered talks from the IT Party AI meetup, held on 20 December 2025 at the GrandDaddy.House bar, in one place (all the recordings are in Russian). These are not round-ups of new models or “ten prompts that will change your life”: the speakers talked about work they had actually done, from integrations and architecture to economics and failures.
Below is what each talk covers, so you know where to start.
MCP: how to give a model access to real services
Ilya Pluzhnikov broke down the Model Context Protocol, the open standard that lets a model connect to external services.
- how to give a model access to real services: Notion, GitHub, analytics, files;
- why context is the biggest problem with today’s models;
- how MCP kills answer quality if you use it the wrong way;
- how Docker’s Dynamic MCP and Claude Code subagents solve this problem;
- real-world cases: analytics, a planner, finance, content generation;
- hands-on experience with Claude, Claude Code and his own MCP servers.
Generative video: an amplifier for ideas, not a magic button
Yevgeny Lyubimov on how AI doesn’t rescue a bad concept but scales whatever you put in at the start.
- GIGO in reverse: gold in, gold out — the value of what goes in decides everything;
- a breakdown of the high-profile McDonald’s case: a big budget does not equal success;
- why the market can reject a perfectly made video;
- the contrasting Kalshi case: speed, context and an honest idea;
- a new production pattern: from concept to generation in hours, not months;
- cases with avatars, digital twins and AI agents;
- the role of customer development and an expert before you generate anything.
Why building an AI product is harder than it looks
Fyodor gives an honest, tough breakdown of where expectations of an AI startup fall apart.
- why an AI product is first and foremost a business, not a technology;
- competing with big tech and the infrastructure gap;
- customer development as the essential foundation;
- why data matters more than the model, and why an idea doesn’t exist until it’s been tested;
- venture capital, money, and why everything works differently outside the US;
- CAC, unit economics and financial models that don’t add up;
- why custom models are usually a dead end;
- the personal price of a startup: burnout, stress, motivation.
The whole meetup in one playlist
If you’d like to watch them back to back, there is a playlist with all the talks. The talks are aimed at founders, product managers, tech leads and developers who already work with AI rather than just trying it on for size.
We now meet to talk about AI on Tuesdays: the schedule and topics are on the AI Tuesday page (in Russian). We discuss the talks and suggest topics in the community chat (in Russian).
Topics from the talks are also covered in writing: context engineering and the engineering process of working with an AI agent.
