Jason, nice overview here of AI agents and this new class of "AI Scientists" that seem to be evolving very quickly in their sophistication. I was surprised looking at the AI2 Scientist and how lengthy and specific the planning prompt was. It makes total sense, and it encourages me to clean up my own agents to improve the formatting (and interpretation) of the code by the agentic tools.
I'm actually interested in hearing your take on an agent system I developed, so I'll email you, but this article was very helpful with the 2 very informative reference Scientists
The prompts in general are very detailed for all of the roles. They have very strong structured output indicators, long context histories stored in memory, detailed tool descriptions...etc. You should consider prompts as highly detailed SOPs
Jason, nice overview here of AI agents and this new class of "AI Scientists" that seem to be evolving very quickly in their sophistication. I was surprised looking at the AI2 Scientist and how lengthy and specific the planning prompt was. It makes total sense, and it encourages me to clean up my own agents to improve the formatting (and interpretation) of the code by the agentic tools.
I'm actually interested in hearing your take on an agent system I developed, so I'll email you, but this article was very helpful with the 2 very informative reference Scientists
The prompts in general are very detailed for all of the roles. They have very strong structured output indicators, long context histories stored in memory, detailed tool descriptions...etc. You should consider prompts as highly detailed SOPs