A Consensus Algorithm

𝐈𝐦𝐚𝐠𝐢𝐧𝐞 𝐲𝐨𝐮 𝐡𝐚𝐯𝐞 𝟐-𝟓 𝐚𝐠𝐞𝐧𝐭𝐬 𝐫𝐮𝐧𝐧𝐢𝐧𝐠 𝐨𝐧 𝐭𝐡𝐞 𝐬𝐚𝐦𝐞** 𝐜𝐨𝐝𝐞𝐩𝐚𝐭𝐡.

How do you consolidate?

1. I find it is easiest to pick one of the agents. And copy one of the methods let's say they worked on.

2. Take that method, and copy it over to your main worktree.
3. In that main worktree, you now have something to work off of, and can then look at 'diffs' between agents [1,3-5].

This is actually in my opinion the way that a human+ai pairing could work.

This same process can actually be used by an orchestrator agent...

Orchestrator Agent
1. Prompt: `Spin up 5 agents to do Task task...`
2. Take a random agents output on the same method, and paste it into main worktree.
3. Now take diffs between Agents 5 - Agent Chosen
4. Continously improve main worktree by running diffs against the set, remove each from Set.

Now this process can be done over and over in a tree like manner, with Orchestrators managing Orchestrators and so on.

This algorithm could in effect work... With the first layer being Orchestrators for finding method names etc. Orchestrators for Simplifying the code, and so on and so forth.

Quote of the Week.

Again, what this shows is that it is difficult to Simultaneously ‘collapse’ five trains of thought. We still aren’t living in a Quantum Compute world. We need to pick one, and diff, diff, diff, diff, until all 5 ‘options’ are looked at. Its a simple algorithm but ultimately is the way that we think.