B-TIER technique
Self-improving skills vs AI teammates — log issues, rewrite skills
100k ft — What is this?
Self-improving skills vs AI teammates — log issues, rewrite skills
50k ft — Why care?
Argues personified 'AI teammates' are static while skills are adaptable: keep every skill in a database table, have each skill log issues about itself when used into a second table, then periodically run a prompt that reads the accumulated issues and rewrites the skill. Specific workflows with clear end states beat persona prompts.
10k ft — Key takeaways
• Batch-apply those notes to rewrite the skill regularly.
• Have skills log their own failure notes as they run.
• Skip 'act as an expert' personas; define the workflow and end state.
5k ft — Tools & techniques
Grokbot
issue logging, skill self-correction
1k ft — Deep-dive / try it
“But angus I can achieve the same production pipeline in Grokbot, isn’t the difference between skills and these teammates mostly semantics?”
Yes, but that in itself is the problem. Personifying ai like this is great for marketing to beginners. But it creates a fundamental misrepresentation of how you should use it imo.
It feels like we are going back to the GPT days where everybody had their fancy custom GPTs that they would put on a pedestal even though it was feeding them terrible advice.
If you want to be excellent at whatever you do, you have to treat AI like the tool it is: an engine that’s really good at executing specific workflows with clear instructions and a clear end state.
Semantics absolutely matter my dudes.
#ai #buildwithai #vibecoding #grokbot
Transcript excerpt
Grokbot is going to make itself extinct. Remember when we were all like, bro, you just gotta tell AI to act as an expert? And then we realized that was totally pointless. AI is only really good at fulfilling specific workflows. So that means the correct way to use Grokbot is to have very ultra specific bots that do very specific things. What does that remind you of? Try this, write down every AI skill that you've ever created. Put them all in a database table like this and have another table of things that you wanna change about these skills. When you use the skill, have it correct itself and leave issues of things it wants to change in that table. And then you can run a prompt like this. It will look at all of the things that you wanna change for that specific skill and then update the skill to be better. So I think the framework of skills is far superior than the framework of AI teammates for capturing and automating workflows. Teammates are static, skills are adaptable. If you aren't like constantly updating your skills, that is bad. That means you aren't learning or improving. I know Grokbot has a bunch of other fucking sick ass features. I'm in agreement. Don't fall for their teammate propaganda. Stop pretending it's a fucking human, bro. It's not.
Appears in
- Guide: Skills That Earn Their Lines — benchmark-gated self-editing plus subtraction
- Deeper dive: Skills That Earn Their Lines — benchmark-gated self-editing plus subtraction
- Guide: Benchmark Your Brain — the skill-optimizer loop, pointed at the human
- Deeper dive: Benchmark Your Brain — the skill-optimizer loop, pointed at the human
More from @angus.sewell
- B-TIER AI networking for job referrals — archetype-based LinkedIn outreach
- B-TIER Chat-history mining — derive better custom instructions from your AI logs
- A-TIER Fish Audio phone agent — AI receptionist that gathers context then hands off
- B-TIER Clipper outreach agent — Roy Lee's pay-per-view clipping army, automated
- A-TIER Anthropic job index audit — map your tasks to automation vs augmentation