About AI agents, without the hype
Deployment practice, honest numbers and the things that never make it into pitch decks.
A Spec for an AI Agent: Five Points Without Which the Project Falls Apart
One task, a metric before the start, data, handover to a human, code ownership, and a one-page template.
AI Agents and Data Law (152-FZ): When You Need a Closed Perimeter
Anonymisation, a local cloud model or your own server: who each one suits and what the strictest option costs.
What You Walk Away With After the Project: Code, Models, Documentation
Code is the smaller part of an agent. The full list of what should stay with you, and a one-day check.
A Legal AI Agent: What It Does With a Contract in Three Minutes
A first read of the contract against company rules, with a quote for every remark. And what not to trust it with.
Weather AI Agent on DeepSeek: OpenAI Agents SDK + Pydantic
Live Gismeteo data through a tool, a Pydantic response contract and honest json_schema pitfalls.
Weather AI Agent on DeepSeek and Pydantic AI: Same Agent, Different SDK
The same weather agent as in the previous article — now on the framework by the Pydantic team: a native provider and a contract the framework validates itself.
RAG or Fine-Tuning: What to Choose for Your Knowledge Base
Why RAG wins in 9 out of 10 enterprise cases — and where it still loses.
How to Calculate the ROI of a Support Agent
A simple model on three numbers: request volume, the routine share, the cost of an operator minute.
Once a month — an email about what worked
No "every Tuesday" newsletters. Only when there is something to show.