It is so interesting how dumb claude in their chat interface feels. Using a local folder with some curated files work wonders. But even without them, so much more intelligent.

Setting up an agentic team template, that can setup a new agentic team for specific needs. And it uses the best practices I gathered. I am anyway using multiple teams for different usecases, and have to create a travel agency team for someone, so it is a great time to start to systemize what I have. Lets see if I can systematically evolve from here.

Relax sometimes, and get bored, so you can utilize your imagination and creativity. great good morning

When you dont have a predefined nym, but your data is connectable together, then you can be assigned an id. (The id doesnt have to mean anything) This I would not consider anonymity, rather pseudonymity.

What is up with this one? Claiming something that isnt actually true sounds bad. And when it is about privacy, it is even worse. great good morning View quoted note →

If nothing goes right, go left. great good morning

I think those who can bear hard and uncomfortable situations, trial and error alot, figure things out on their own (e.g.: doing self-hosting), lets call them hacksters, will have a bright future later too. Everything gets much, much more comfortable and easy, but there will be always things that only hackster can do, and the non-hacksters will either not be able to do, or only much later.

I am tinkering with the thought of hosting my own agent/LLM since a long time. I did not have yet too much success. Maybe its time from the signs, to try again. I think the future is bright to use frontier models for creating new stuff/stuff without personal data/stuff for your local agent teams. Local agent teams to use your data, automate your personal stuff.

My improvement loop now with my agentic team is quite simple. I built an agentic AI five pillars/best practices document. When an agent or the team does something badly, I ask it to audit the current agentic system compared to the five pillars, what caused the issue, and tell me how to fix it with the exact new content and its location. Works great so far.

As much as we want to make everything easy for people, we cannot forget that self-responsibility and learning are not outsourcable.

The next most important skill is system designing. If you want to use AI effectively, you have to create an agentic system that effectively handles the problems you put in front of them.

Agentic AI usage is more of a system design problem than anything else. But this is also true for simple human team work. The quality of the output of the team correlates with the quality of its system design. great good morning

Is there an active group chat on ai team maxxing? Who is doing agentic ai team maxxing?

Maybe I need an innocent user in my agentic team arsenal.

As the result of this best practise idea, I have updated my agentic team at work. Its genial. I have a separate reviewer and a critic agents. Reviewer said, yeah, the code is fine, there are findings, but they are not important. The critic is sending now the dev into the second loop of fixing 😂😂😂 I love him. 😂

Todays learning: Having a best practice document (refered to it in previous post) really helps alot improving an agentic AI team. I can just tell it: "X happened. Please check the current agentic system definitions against the best practice document. Please find out what causes it in the current system and how fix it according to the best practice document." I can even tell it to "create the exact content, and define in which files would you update it", so I can steer before the change. I have 2 five pillars document at the moment, one for a coding oriented team, and one for a knowledge worker oriented business. Testing both in parallel.

After 2 days of small development, but deeper research on the correct ai agent framework, and how to do it. My brainchild was born: "5 pillars of quality agentic system with their underlying principles" Next days are testing it for coding and business automation purposes. Lets go. great good morning

Did anyone have any success with WebLLM? Which model worked for tool calling for you?