Brute force engineering

In computer science, a brute force solution is an algorithm that enumerate all possible candidates or combinations to find the right answer to a problem. So instead of using an algorithm like quick sort to sort a list, you would check every single possible list ordering until one happens to be sorted.

With agents, because the cost of implementing one off prototypes is so low now, if I come across some problem and the best solution is not clear, I implement all of them and compare them. The point isn't to be mindless, this is an opportunity to learn the details you were missing that didn't allow you to make the decision from first principles.

The difference between this and normal iterative development is that you intentionally explore ideas you wouldn't have otherwise and by using agents you can create agents that don't have your own bias.

This works for other things too:

Have your agent implement a design 20 times with different variations of different components and compare them. Then filter the things you like and don't like. Often you end up combining things from different designs.

Are you stuck between multiple ideas? Launch dedicated agents to give their best arguments for and against the different ideas and evaluate them.

When doing this, it's best to have one orchestrator agent and keeping the agents completely separate so they don't influence each other and you get truly independent implementations. The orchestrator can also ensure the full space is being explored.

Like the list sorting case, this can only ever work for small N. You're not going to be able to do N choose M or N! implementations or designs for any N > 4 so you have to use your existing knowledge and skills to whittle down the solution space before you try this approach.