Thank you for this review, I always like seeing different perspectives. That being said, I don't agree with your final assessment of Agent Zero. For example benchmarks step 1 through 6 could use a little more analysis. I'm thinking that cost and execution time depends on desired goal as interpreted by the framework. So if normalization of data before comparison wasn't taken into account, I'd probably reconsider what the data is actually telling us if we don't adjust or adapt for discrepancies between the different frameworks. Finally, step 7 and 8 contradict each other and so do steps 9 and 10. Step 11 which is the summary, declares the framework not to be "agentic". I think this surprised me the most and made me question what "agentic" really means when referring autonomous AI agent frameworks. This is all IMHO of course, so any feedback or insight would be appreciated!
Thanks for watching :) Concerning your question, check some videos of Andrew Ng, he explains the word "agentic" very well. It can be missleading in many cases as its still a very new area, I am pretty sure that it will change and evolve a lot in the near feature.
You are comparing apples and oranges here. Agent zero is a single autonomous agent, not a multi agent agentic framework. There are a bunch of others now popping up in that space. Prompts and which llm you use, determine everything. Agent zero prompts can be easily improved, very visible in the code, not hidden deep diwn in pip package ...
Yes, I realized that too, you are right, concerning agent zero the comparasion does not fit that well, anyway I wanted to make it part of the serie as its pretty popular right now. You know any other frameworks similar to agent zero, I tried autogpt, but was not that happy with it.
Thank you for this review, I always like seeing different perspectives. That being said, I don't agree with your final assessment of Agent Zero. For example benchmarks step 1 through 6 could use a little more analysis. I'm thinking that cost and execution time depends on desired goal as interpreted by the framework. So if normalization of data before comparison wasn't taken into account, I'd probably reconsider what the data is actually telling us if we don't adjust or adapt for discrepancies between the different frameworks. Finally, step 7 and 8 contradict each other and so do steps 9 and 10. Step 11 which is the summary, declares the framework not to be "agentic". I think this surprised me the most and made me question what "agentic" really means when referring autonomous AI agent frameworks. This is all IMHO of course, so any feedback or insight would be appreciated!
Thanks for watching :) Concerning your question, check some videos of Andrew Ng, he explains the word "agentic" very well. It can be missleading in many cases as its still a very new area, I am pretty sure that it will change and evolve a lot in the near feature.
You are comparing apples and oranges here. Agent zero is a single autonomous agent, not a multi agent agentic framework. There are a bunch of others now popping up in that space.
Prompts and which llm you use, determine everything. Agent zero prompts can be easily improved, very visible in the code, not hidden deep diwn in pip package ...
Yes, I realized that too, you are right, concerning agent zero the comparasion does not fit that well, anyway I wanted to make it part of the serie as its pretty popular right now. You know any other frameworks similar to agent zero, I tried autogpt, but was not that happy with it.
@@FlorenzErstling @aicodeking has several recent reviews