AI Business
Photo of author

AI Agents and Chat GPT 5

There’s been a recent OpenAI leak in the past few hours regarding ChatGPT 5’s capabilities and the release date.

The expected talking points are included in the article like a “really good, like materially better” model and training against increased datasets. But the portion that caught my eye even more was the aspect of the “AI Agent”. Summed up, it’s an agent that can perform tasks on your behalf.

I’ve seen parts of the marketing and programming community refer to this as an API but specifically for AI. Theoretically, you could ask an AI agent to perform a task like let’s say “Put together a vacation” and it could:

  1. Gather available hotels
  2. Time the flights to those available hotel times
  3. Gather (and reserve!) any dinner reservations or fun activities to do
Obviously we're a ways from an AI agent planning a whole birthday party for a child.

Something like this is some time away, but I really wanted to poke at the idea of what something like this would require.

For instance, in order to do #3 (reserve any dinner reservations), that would require a bit of a unified idea of a reservation system API. Or at bare minimum, a standardized Web API with a uniform concept of a reservation event, a cancellation event, etc.

Or even more simply (and maybe more likely?), we’d still use our existing API capable infrastructure and just slap an agent behind it.

Will this new API take over the Assistant API?

Seems like this idea of agents isn’t entirely new though. Take for example the “How Assistants Work” documentation from OpenAI. It’s a similar idea: Dispatch an event, let a specifically trained or nuanced AI agent look into it, and then come back with an answer. I think the real difference is now the agent will have access (or abilities) to actually use an external API of some kind.

The current flow of an ai agent doesn't really have any flow to external APIs

Right now though, it seems like it’s limited to just files on your filesystem and the data the LLM models have access to.

Wait and See

As usual a lot of AI is a ‘wait and see’ game. While experimenting with these features isn’t exactly free, it’s at least a metered usage model. So i’d say let’s play with it while we still can. Maybe in the future I can post little programs and models and see what we can get the new AI agent’s to do.