Working with Agents — A Practical Lesson

A narrated lesson (about 7.4 minutes, 1920×1080) on getting useful work from coding agents, adapted from Thorsten Ball's article "How I work with Agents": delegate an outcome, give the agent contact with the real system, and ask for evidence that it worked.

• Source article: https://x.com/thorstenball/status/2114386050638336082 (Thorsten Ball, "How I work with Agents", October 25, 2026)
• Video: working-with-agents.mp4 (H.264 + AAC)
• Subtitles: working-with-agents.srt
• Narration transcript: ../../transcripts/working-with-agents.txt

This is an independent teaching adaptation. The examples and safety notes are instructional additions, and the narration and graphics are AI-generated. The principles are Ball's opinions, not universal laws.

Chapters

1. 00:00 Work with agents. Not autocomplete.
2. 00:27 Start with a capable model.
3. 00:50 Don't assume "slop" wins.
4. 01:13 Prompt the what. Not the how.
5. 01:37 Make the whole lifecycle visible.
6. 02:02 Give it tools for the job.
7. 02:29 Test whether knowledge helps.
8. 02:54 Give each agent room to run.
9. 03:18 Build tools for iteration.
10. 03:42 Ask for proof. Not reassurance.
11. 04:06 Let agents wait on you.
12. 04:28 Make follow-up automatic.
13. 04:50 One task. One thread.
14. 05:12 Threads become team memory.
15. 05:35 Keep your human voice.
16. 05:59 Review by blast radius.
17. 06:26 A better delegation loop.
18. 06:58 Give agents a complete job.