OpenTube

5 Best Practices for Building AI Agent Skills

IBM Technology

84,561 views

Learn more about 2026 Guide to AI Agents here → https://ibm.biz/~e7fElYU4A

Building AI agent skills is easier than it looks. Martin Keen and Bri Kopecki explain five best practices for creating reliable, secure, and effective AI agent skills. Learn how to improve triggering, reduce context usage, add deterministic logic, and avoid common pitfalls.

0:00 - Intro
1:29 - Write Skills That Trigger
3:38 - Build from Real Expertise
6:15 - Spend Context Wisely
8:25 - When Agents Shouldn't Guess
11:02 - Can You Trust This Skill?
12:20 - Key Takeaways

AI news moves fast. Sign up for a monthly newsletter for AI updates from IBM → https://ibm.biz/~H9ZIaiRnS

AI was used in the creation of the transcript and metadata for this video.

#ai #aiagents #aiskills #agenticai

20 Comments

@MikeDrapolyuk 1 month ago

Two additions: ship skills as a plugin (versioned + installable beats copy-pasted), and test them like code — smoke tests and a conformance loop.

@SaiSS961 1 month ago

I love this new format. You should do more conversational style videos 🙂

@FriedOkraPickles 3 weeks ago

What a strange world I find myself in. My beer guy is now my AI guy too.

Hey Martin! Loved your homebrewing challenge, love your brulosophy stuff too! Fun to find you here too, what a busy guy

@supersymmetry8964 1 month ago

It sounds like how to handover a job to a new teammate with a well crafted written note.

@castlecodersltd 1 month ago

Great series, thanks 🙂

@AkhlaqurRahman-gf1br 11 days ago

Wonderful analysis as well.

@TheSp_cemanChannel 1 month ago

Great presentation! Will be back for more.

@gabscar1 1 month ago

Great video. You guys are really good communicators. Thanks!

@gustafonseca2 1 month ago

That's gold! It’s all fun and games when we are testing a single agent and skill, but it demands continuous improvement and fine-tuning once we keep triggering agents and skills every single day, especially for the quality and consistency of work. I am in a constant "battle" with Claude and Codex to ensure they are not adding 'rules' that are merely choices for a specific task and to determine where we can write scripts to avoid this kind of "drift" that starts to happen when we keep triggering the same agents and skills for months.

@crab_seeker_42 1 month ago

great advices!

@TheCarribbeanJackSparrow 1 month ago

Good to see both of you in the same video.

@sivarajkumar237 1 month ago (edited)

I love these videos and I cannot believe they are free for everyone! Thanks guys!

@AliEskandari-x2i 2 weeks ago

the "spend context wisely" part hit different, i learned this the hard way building my first agent workflow lol. been doing schole's agent course alongside stuff like this actually, curious how you'd apply the triggering advice to skills meant to run pretty rarely?

@rahlquist-drahcir_rahl 1 month ago

Y'all killed it with this one, very informative.

@shravankarup509 1 month ago

Great content as always. Thanks IBM team ❤

@BrandonLecocq-1 1 month ago

how do you setup your recording to be able to record what you are writing in real time? Awesome video.

@MD21-w9w 1 month ago

Use three architectural patterns depending on the workflow. Decomposition means breaking a large responsibility into focused Skills. Aggregation means independent Skills contribute structured results into common state. Extension means Skills sequentially improve the same artifact, with each step building on what came before.

@csaba-a-tatrai 1 month ago (edited)

6 ) Use proper naming conventions like starting skill's name with gerund in order to understand later what you made the skill for :)

@abigwreck 1 month ago

Good info

@saketthakur1722 1 month ago

wow so much learning in 13 minutes.