"Stop thinking like a manager. Start thinking like a transformer"
Would be interesting to see how to synthesis these two viewpoints - consider complex task context, speed - parallelism (with different prompts/ expert agents) etc.
That article is talking about agents whereas my article is about basic LLM usage. Those are two very different things. Agents are too new, too raw, too untested, and have various failure cases (some of which are listed in the article you linked). As a result, I have not really written about agents in my newsletter—people on the cutting edge can explore agents, but I believe that most people can wait it out.
In my article, I'm talking about you as being a manager of a team of LLMs where each one is just the simple web-interface-based use of that LLM (i.e. via chatgpt.com or claude.ai), and not API or agent usage.
Other viewpoint from https://letters.lossfunk.com/p/dont-get-tempted-to-build-multi-agents
"Stop thinking like a manager. Start thinking like a transformer"
Would be interesting to see how to synthesis these two viewpoints - consider complex task context, speed - parallelism (with different prompts/ expert agents) etc.
That article is talking about agents whereas my article is about basic LLM usage. Those are two very different things. Agents are too new, too raw, too untested, and have various failure cases (some of which are listed in the article you linked). As a result, I have not really written about agents in my newsletter—people on the cutting edge can explore agents, but I believe that most people can wait it out.
In my article, I'm talking about you as being a manager of a team of LLMs where each one is just the simple web-interface-based use of that LLM (i.e. via chatgpt.com or claude.ai), and not API or agent usage.
Interesting take on using LLMs!