No Agent<A>, only AI<A>

[serras]

This PR goes one step further than #76 in removing the Agent abstraction altogether. "AI Agents" are now simply functions that use the shared "AI Scope". As you can see, no changes were needed in the examples.

One side benefit of this is that prompt, image, and so on, are no longer parts of the AIScope class, they are just defined as simple extensions of it.

Pinging @nomisRev @raulraja @realdavidvega for discussion

Note: the build breaks because I haven't updated the Scala code.

[nomisRev]

I really like it @serras! 🙌 Should we target it at my open PR, or should I close mine?

The more everything fits in the AI DSL the better IMO, this all seems to compose together than what we had before. If we can introduce @javipacheco work here for building more complex prompts I think we've gotten a nice API for core.

[serras]

I didn't realize I targeted main... I'm OK with whatever, either merging this directly, or re-asigning a base branch.

[raulraja]

This is much nicer and simpler. I guess we had names and descriptions for agents, which with functions are gone, but those were only useful for logging purposes.

[serras]

This is much nicer and simpler. I guess we had names and descriptions for agents, which with functions are gone, but those were only useful for logging purposes.

I tried to keep those by defining:

interface Agent<A>: AI<A> {
  val name: String
  val description: String
}

fun agent<A>(name: String, val: String, block: AIScope.() -> A): Agent<A>

but it was too hard for the compiler to figure out what was going on in several cases. In particular, it could no longer infer the type variables in prompt, which seemed like a big loss to me.

[realdavidvega]

I think now it's much better than before, looks more clean and also makes more sense.
Great work @serras 🙌🏻 Thank you!