Series · 8 posts

AI agents: from the first run to a fleet

Eight posts on working with agents like Claude Code, Codex and Cursor: setup, permissions, checks, orchestration. Each post is a Telegram carousel plus a full version with commands and links to the docs.

Series

Eight steps to an agent of your own

Post 1 of 8

How to take the right first steps toward AI agents

Where to start if you want an agent rather than another chat: pick a task with a verifiable result, start with the simplest solution, give the agent a way to check itself, and only then widen its autonomy.

9 slides · 4 min read
Post 2 of 8

First-time setup of an AI agent

Step by step: install Claude Code (or Codex, or Cursor), write a CLAUDE.md or AGENTS.md instructions file, set permissions, give the first task in plan mode and verify the result.

8 slides · 5 min read
Post 3 of 8

Advanced AI agent setup: skills, hooks, subagents, MCP

What to add after the first setup and why: path-scoped rules, skills for repeatable procedures, hooks for hard rules, subagents for context isolation, MCP for external tools, the sandbox and running in CI.

9 slides · 6 min read
Post 4 of 8

Types of AI agents: where they run and how they're built

Agents differ along three axes: where they run (chat, IDE, terminal, cloud, CI), how they're built (workflow, single agent, orchestrator with workers) and how much they may do without a human. We go through each axis with Claude Code, Codex, Cursor and Copilot.

7 slides · 5 min read
Post 5 of 8Live

AI agents: news from primary sources

A live digest: releases of Claude Code, Codex, Cursor, the MCP specification and agent frameworks. Collected from official blogs, changelogs and GitHub releases, every note links to the primary source.

9 slides
Post 6 of 8

What you need to have an AI agent of your own

The seven parts of a working agent: model, harness, tools, context, checks, permissions and budget. What you can take off the shelf, what you'll have to build and what it costs in tokens.

8 slides · 5 min read
Post 7 of 8

An AI assistant with a fully connected stack

How to connect an agent to your repo, tracker, database, browser and monitoring via MCP without handing it excess permissions. How the protocol works, where configs live, how auth works and why prompt injection matters.

8 slides · 5 min read
Post 8 of 8

The fleet: several agents, orchestration and background work

When one agent isn't enough: subagents for context isolation, parallel sessions in git worktrees, background and cloud agents, agent teams and orchestration. How not to drown in parallel work, burn the budget or accept unchecked results.

8 slides · 5 min read

How the series works

Carousel

6–10 slides: the gist in a minute. Published in the Telegram channel.

Full version

Steps, commands and config files, with links to official documentation.

Glossary

Every term from the posts explained on its own. A regular channel rubric.

Glossary

A dictionary of agentic development

70 terms in plain words, from harness to fleet. Each has an example and a link to the primary source.

AI agentagentHarnessagent harnessContext engineeringMCPModel Context ProtocolSubagentsub-agentHookPreToolUseSkillSKILL.mdEvalevalsOrchestratororchestrator-workersAgent fleetfleetPermission rulespermissionsCLAUDE.mdCLAUDE.local.md
All terms

Join the community on Telegram

New chapters and simulators, breakdowns of course tasks and discussion with other learners. Free.

Open the channel