Just about to sign-off for the day and saw this awesome comparison of open-source models for group chat speaker selection in AutoGen. It's a global hustle. github.com/marklysze/Auto…

1. Dogfood; 2. Talk to users everyday. 3. Clarity >> fuzzy vision.

Bring your own customized and containerized execution environment to your multi-agent workflows.

This is why AutoGen has built in human in the loop in every agent — microsoft.github.io/autogen/docs/t…

Running multiple models locally and having them talking to each other is actually really fun.

Agent capability will go up as (1) model improves e.g. GPT-5, and (2) evolution in multi agent conversation patterns. Agent cost will go down as (1) SLMs get into the game and replace LLMs for sub tasks; and (2) optimization and reuse of multi agent components. AI agents

I haven't gotten paid for this... maybe I should have? GitHub Kevin A. Crosby It's at 3M downloads per month. github.com/ekzhu/datasket…

Likewise! Join our discord channel for more in-depth technical discussion of agentic AI applications and research. aka.ms/autogen-dc

Multi-agent collaboration has emerged as a key AI agentic design pattern. Given a complex task like writing software, a multi-agent approach would break down the task into subtasks to be executed by different roles -- such as a software engineer, product manager, designer, QA

Interested in learning how a multi-agent applications can be built to address tasks like browser automation? Tamer Abuelsaad, who has been building Agent-E (a system that aims to automate actions within the browser) using AutoGen will discuss the project and share his

Much has been said about many companies’ desire for more compute (as well as data) to train larger foundation models. I think it’s under-appreciated that we have nowhere near enough compute available for inference on foundation models as well. Years ago, when I was leading teams

AutoGen This turned out to be a insightful and interesting conversation. Thanks Tamer Abuelsaad and Deepak Akkil. For those who could not attend, below is a brief summary of the conversation. Main Insights from Presenters - DOM as text: Current implementation represents the content

Mental!

16 K for the parts but a good software is probably going to cost more at least in beginning

Latest requests package update broke Docker client and then the Python ecosystem. github.com/docker/docker-… (Image from geerlingguy in the thread above)

I spent more than 30 hours photographing Orion’s Belt through a telescope to reveal this: these two nebulae that are larger than the full moon in the sky. This was formatted as a mobile wallpaper you can save as part of my daily giveaway. More coming!

Boeing is out? Maybe there will be room for another player.

Tech trees do exist in large companies for very good reasons. When it is done right it unlocks huge potential and create cohesion. Although, it’s a big mistake for a startup to engage in building foundation rather than creating immediate value for customers.