Skip to content
Auf Deutsch ansehen

Learning pathLevel 3 Expert

Running agents

The pack is not unlocked yet. Sign up below and click the link in the confirmation email – that link opens it. Jump to the practice pack ↓

You are subscribed, but this browser does not carry the access. Open the link from your welcome email here once more. Jump to the practice pack ↓

Getting an agent to run takes an afternoon. Keeping it running for six months is a different job. This path covers what lies in between: evals that show whether a change improved anything, skills as reusable procedures, the question of when several agents pay off, the cost calculation and a runbook somebody can find in an emergency. The duties from the AI Act come last because they bite at the same points.

Stations
7
min
71
Who it is for
For anyone running agents for others – in a team, a store or a company.

What comes up

0 of 7 stations · 0 of 71 min

Your progress lives in this browser. No account, no data with us.

  1. 01Lesson10 minStart here

    Evals: how you know it works

    You build a small test set against which you measure every change to prompt, model or tool.

  2. 02Lesson9 minStart here

    Skills: procedures an agent loads on demand

    You package a recurring procedure as a skill an agent loads on demand – and vet third-party skills before you use them.

  3. 03Lesson10 minStart here

    Multi-agent systems: when several are worth it

    You make a reasoned decision on whether a task needs one agent or several, and know the three patterns that hold up.

  4. 04Article11 minStart here

    Understanding AI Costs: Tokens, Context, Caching

    You work out what an agent run costs and know the levers in order of effect.

  5. 05Lesson9 minStart here

    A runbook for agents

    You write a one-page runbook for an agent: ownership, limits, log, kill switch.

  6. 06Article10 minStart here

    The EU AI Act in November 2026: What changes on 2 December

    You map an AI use to the duties of the AI Act and know who in the business supplies which evidence.

  7. 07Article12 minStart here

    Adopting AI at work: From a pilot to a service you can run

    You set up a pilot so that its result supports a decision – even if that decision is “stop”.

Work through it, don't just read it

The practice pack holds one exercise per station, the answers to every self-test and the closing overview. It is free for newsletter subscribers – confirm the newsletter once and it opens.

Double opt-in: you first receive a confirmation email – without your click you are on no list. Unsubscribe using the link in any issue. Details in the privacy policy.

Already subscribed on another device? Open the pack

← All learning paths