Ask HN: Do subagents stop making sense at some point?

1 pointnemath3 comments

If we have really powerful models (AGI) - would it still make sense to have subagents that have access to specific tools and has specific context?

Wouldn't the overlord model just do everything?

Do you have GI? Does it makes sense for you to use a subagent?

Anyhow, a solar-powered calculator (cost: approximately $0) gives a more reliable and more efficient answer than $100 billion and 100GW of AI.

The overlord model might be expensive to run, take a long time, etc.

So even if things become cheaper to run and faster to run there will always be a use case for a submodel?