AI Workshop · live

Working with AI agents — live

Alexander Mihalkevich shows on screen how it's really done: setup, connecting and authorizing, steering the agent, prompts and the harness. Unedited — the real pace and real mistakes.

Every Wednesday · 19:00 Moscow time · free · recording and checklist after each stream

Next streamEpisode 1 of 10

Setup from scratch

A clean machine → a working agent setup in one stream: install, environment, CLAUDE.md and AGENTS.md.

Wednesday 7 October at 19:00 Moscow time· 90 min

The stream link arrives by email an hour before the start. A day before — a reminder with the episode plan.

Add to calendarGoogle CalendarEpisode page

What we'll show

  1. Installing Claude Code and Codex CLI on a clean machine, updating, checking the version
  2. Terminal, editor, git and Node — the minimum an agent can't see without
  3. A first CLAUDE.md: stack, build and test commands, prohibitions — and how it differs from AGENTS.md
  4. How the agent reads a project in its first minutes and why repo hygiene matters
  5. A first small task from wording to commit

What the workshop is

Not a course and not a slide webinar. A working screen, a live task and a conversation about why the agent does what it does.

Real work on screen

Terminal, editor and agent — like any working day. You see where the agent goes wrong and how to fix it.

The harness as a through line

Each episode adds a layer of control: CLAUDE.md, permissions, checks, hooks, tests. By the finale you have your own.

You leave with a template

Every episode has a downloadable checklist or template, official docs links and a recording with chapters.

Questions, live

Ask in the chat or reply to the reminder email. The season finale reviews viewer projects.

Season program

Ten episodes in order: from a clean machine to an agent fleet and a review of your project. One Wednesday per episode.

  1. Wednesday 7 October at 19:00 Moscow timeNext

    Episode 1: Setup from scratch

    A clean machine → a working agent setup in one stream: install, environment, CLAUDE.md and AGENTS.md.

  2. Wednesday 14 October at 19:00 Moscow timeUpcoming

    Episode 2: Connecting and authorizing

    Keys, OAuth, MCP servers, secrets and permissions — wiring an agent to services without leaking anything.

  3. Wednesday 21 October at 19:00 Moscow timeUpcoming

    Episode 3: How an agent actually works

    The «think → call a tool → read the result» loop, the context window, and why an agent suddenly «forgets».

  4. Wednesday 28 October at 19:00 Moscow timeUpcoming

    Episode 4: Steering the agent

    Plan → execute → verify. Targeted edits instead of rerolls, and the honest moment when it's time to stop.

  5. Wednesday 4 November at 19:00 Moscow timeUpcoming

    Episode 5: Prompts that work

    Task structure, definitions of done and examples — writing prompts that don't need redoing.

  6. Wednesday 11 November at 19:00 Moscow timeUpcoming

    Episode 6: Rules of intelligent communication with AI

    The workshop manifesto point by point: clarity, context, verifiability, honesty, polite persistence — and no trust without checking.

  7. Wednesday 18 November at 19:00 Moscow timeUpcoming

    Episode 7: Harness: the agent under control

    What a harness is: evals, tests, hooks and a sandbox — building the frame inside which an agent can be trusted.

  8. Wednesday 25 November at 19:00 Moscow timeUpcoming

    Episode 8: A fleet of agents

    Subagents, background agents and orchestration: when one agent isn't enough, and how not to drown in ten.

  9. Wednesday 2 December at 19:00 Moscow timeUpcoming

    Episode 9: The agent on a team: review and CI

    Agents in pull requests and CI: automated review, repo rules and shared team skills.

  10. Wednesday 9 December at 19:00 Moscow timeUpcoming

    Episode 10: Viewer project review

    Season finale: we take real viewer projects and fix them live — setup, prompts, harness.

Workshop manifesto

Rules of intelligent communication with AI

Twelve short rules that all agent work rests on. With do / don't examples.

  1. 01Clarity before speed
  2. 02Context is work, not a luxury
  3. 03Define done before you start
  4. 04Plan first, code second
  5. 05Trust, but run it
  6. 06Targeted edits, not rerolls

Recordings archive

Stream recordings with chapter timecodes. They appear within two days of the stream.

The first recording appears after the stream

Sign up — we'll send the stream link, and then the recording.

Sign up for the stream

Rules of participation

  1. Streams are free. Sign up by email: a reminder a day before, the stream link an hour before.
  2. Ask in the stream chat or by replying to the reminder. Questions sent in advance go first.
  3. Send a project for review by replying to the email. We show it only with your consent and hide keys, tokens and personal data.
  4. The screen shows real, unedited work. Agent mistakes are part of the lesson, not a reason to restart.
  5. If a key shows up on screen, it's revoked and reissued right away — that's part of the harness too.
  6. The chat is for questions on topic. Ads, spam and rudeness are removed without warning.
  7. Recordings go into the archive with chapters. If you'd rather not be recorded, keep your camera and mic off.

The «AI agents» section — terms, tools and how to choose an agent →