Stop trying to get AI to agree with you. Make it disagree with itself...productively!
Here's how steel-manning + contrarian agents produce 10x better output than consensus.
Excited to try this - have been using Claude subagents to “red team” ideas, but not steel manning.
I’ve gotten value from “user” agents providing feedback on a code prototype wither other agents building solutions. Some of the ideas from that process have been specifically called out (positively) from carbon-based users.
Hey Robb. Let me know what you find. I've been experimenting with how much precision can I get upstream in the process and is it possible to just one shot the coding off the deliverable at the end? So far what I described here has yielded the best results. Can't imagine where we will be in a year.
Excited to try it out. One question I had from your piece is what types of problems you were applying this method to - a refactor, a new user-facing feature, a new data pipeline, etc ? I’d assume different problems may benefit from some difference in approach, but a “golden requirements” approach up front is compelling. And yes, it’s exciting to think where we can be a year from now!
Excited to try this - have been using Claude subagents to “red team” ideas, but not steel manning.
I’ve gotten value from “user” agents providing feedback on a code prototype wither other agents building solutions. Some of the ideas from that process have been specifically called out (positively) from carbon-based users.
Hey Robb. Let me know what you find. I've been experimenting with how much precision can I get upstream in the process and is it possible to just one shot the coding off the deliverable at the end? So far what I described here has yielded the best results. Can't imagine where we will be in a year.
Excited to try it out. One question I had from your piece is what types of problems you were applying this method to - a refactor, a new user-facing feature, a new data pipeline, etc ? I’d assume different problems may benefit from some difference in approach, but a “golden requirements” approach up front is compelling. And yes, it’s exciting to think where we can be a year from now!