Claude Code like a pro: the field tricks the docs don't tell you
The documentation shows you the features. This session shows how to get real speed out of them: keep the context clean, make the agent check its own work, run sessions in parallel and keep the bill down. Three hours, hands-on, from daily production use.
- Thu 19 November 2026
- 09:00–12:00 CET
- Live online
- Hands-on, you write code
- Worth EUR 500 · free for TechEx winners
Most teams get a fraction of what Claude Code can do
Installing a coding agent takes five minutes. Using it well takes months of trial and error: bloated sessions that forget the rules, tests the agent quietly weakens, usage limits hit by lunchtime. We made those mistakes on client projects so you don't have to. You leave with the habits, the settings and the files we use every day.
- Context: why long sessions get worse, and the reset discipline that fixes it
- Verification: loops and hooks that make the agent prove its work
- Speed: parallel sessions, headless runs and commands for repeat work
- Cost: what really burns tokens and how to stay inside your plan
- Date
- Thursday 19 November 2026
- Time
- 09:00–12:00 CET
- Format
- Live online, hands-on
- Trainer
- Miklós Tóth
- Level
- Developers, data and ML engineers
- Value
- EUR 500 · free for TechEx winners
- Register by
- Tuesday 17 November
Three hours, four hands-on blocks
Kick-off: where your tokens and your time really go
What happens inside one Claude Code turn, why the agent gets worse as a session grows, and the three mistakes we see in almost every team. Setup check on a shared practice repository.
1. Context is your budget
A CLAUDE.md that the agent actually follows: short, with the exact commands, split per folder. When to clear and when to compact. Plan files that survive a reset. Subagents that do the research so your main session stays clean. Why fewer MCP servers make a smarter agent.
2. Make the agent prove its work
Hands-on: build a feature test-first. Commit the failing tests before the agent touches the code, so it cannot "fix" them by weakening them. Hooks that run the formatter, linter and type checker on every edit, without asking. Screenshots and logs as feedback. Interrupt early, rewind, retry.
Break
3. Go faster: parallel and headless
Several sessions at once with git worktrees. Headless runs in scripts and CI for bulk changes. Your own slash commands for the work you repeat every week. A permission allowlist instead of clicking "yes" two hundred times, and when skipping permissions is safe (in a sandbox only).
4. Cost, limits and safety
Which model for which task, what burns tokens, how to stay inside your plan's limits, and when not to use the agent at all. Prompt injection through repository files and web pages, and the settings that protect you.
Q&A and take-home
Our CLAUDE.md template, a hook and command pack ready to copy, and a one-page cheat sheet.
Who it is for and what you need
Who it is for
Software, data and ML engineers and tech leads who already code with an AI assistant, or are about to, and want to get much more out of it. Beginners to Claude Code are welcome, the setup check gets everyone started.
What you need
A laptop with Python 3.11+, Git and a terminal. A Claude subscription or an Anthropic API key. The setup guide comes with the calendar invite, a week before the session.
What you take home
A CLAUDE.md template, hooks and slash commands you can copy into your own projects, the practice repository and a one-page cheat sheet of everything we covered.
Reserve your seat
Free seats go to the winners of our stand games and to guests we invited at the stand. Seats are limited. Can't make it? Pass the seat to a colleague.
You're in. Check your inbox.
We've sent a confirmation to . The calendar invite, the joining link and the setup guide follow before Thursday 19 November.
Can't make it? Reply to that e-mail and we'll move the seat to a colleague.
Want this workshop, or the full-day Develop with Claude Code course, for your own team? Book a 30-minute call with Miklós.