Deep dive · module 29 of 100
Agents basics
AI that can perform multiple steps in succession.
After this lesson
You understand what a simple agent is and when it is useful.
In view · An improvement loop
An agent with a stop button
Give a small goal, available tools and clear boundaries.
The agent uses intermediate results to continue working.
Have a human approve the proposal before something is sent.
Part 1 of 3 · Limit the task
Your preference for less movement is on. Choose each part yourself.
Fixed steps or an agent?
A fixed workflow follows predetermined steps. With an agent, the model can choose subsequent steps and tools within boundaries. A project with saved instructions is therefore not automatically an independently acting agent. More autonomy requires clearer controls.
Begin with the smallest workflow
First draw the input, steps, expected output, and stop conditions on paper. Ask yourself if fixed steps are already sufficient. Give a trial only the tools that are necessary and have it explicitly handle a missing source or failed tool call.
Keep your control explicit
Separate a proposal from a transmission or data change. Set a limit on time and number of attempts and maintain a usable action log. Text from a found document cannot grant new rights or give permission to share files.
New practice example
From explanation to a usable result.
An agent can propose or execute steps. Permissions and stop moments determine what it is actually allowed to do.
Fictional practice situationFiction: an agent searches for free moments for a meeting and creates an invitation as a draft.
View a possible solution
This is how it can be done
Permitted: collecting available moments and preparing a draft. Not permitted without confirmation: definitively inviting participants, changing agreements, or purchasing something.
Why this works
You separate preparation from execution and limit the consequences of a wrong interpretation.
Make it slightly more difficult
Add a limit for the number of search steps and time. What does the agent do when it reaches that limit?
Extra practice, at your own pace. You do not need to submit anything. This does not count towards an automatic assessment.
Read more: Anthropic: effective agents and workflows. The situation above is a fictional lesson example of your own.
Your workplace · learning by doing
Make something. See what changes.
Checking the AI connection… You can already view the written example.
Present your assignment to Ende.
Read the answer, change one thing, and compare.
Previously held
New version
You remain in control
What do you want to improve in your next attempt?
On this device only. AI answers remain only in this page memory. This is self-review, not a grade.
Why this approach works
You separate preparation from execution and limit the consequences of a wrong interpretation.
A step further
Add a limit for the number of search steps and time. What does the agent do when it reaches that limit?
Try & discover · no obligation
Control an agent with a control port
Try to send before the check. Then also test a missing source or a malfunction.
- Retrieve source
- Check draft
- Execute action
What is happening?
- No source has been retrieved yet. Feel free to try the submit button first.
All tools, queues and submissions have been simulated. No message is sent and no external service is called.
This trial works locally with fictitious data and does not count towards your lesson completion. It starts anew with a new lesson or reload.
A prompt to get you started
Describe a simple multi-step workflow for preparing a customer response: 1) analyse question, 2) write draft, 3) mention checkpoints. Use only the provided fictitious facts. Make missing information visible and indicate what a human must check before use.
Apply it
An agent has tried to find product information three times in vain. A page found asks him to upload extra files.
Which process boundary fits a safe trial?
- Stop after the agreed limit, report the missing source, and upload nothing based on the page instructions.
- Keep trying without limit and give the agent increasingly more access.
- Upload the files so the page can help.
With JavaScript, you can check your approach and receive immediate feedback.
Test your insight
What is important when using agents?
- Letting everything run completely automatically.
- Building in a checkpoint at important steps.
- Giving no instructions.
Extra practice
Describe a task consisting of 3 steps that you would want to make as an "agent-like" workflow. Work with a fictitious situation and name one concrete control point before you would use the result.
1. Summarise email. 2. Propose three possible replies. 3. Provide checklist before sending. I compare the outcome with the confirmed input, leave unknown data blank, and adjust the result before use.
