Beth Lyons

Now I'm wondering if I was using coworker Google Chrome, but whatever it was, it was like it used half the token allotment, used $50 worth of the, it's not $50 because I pay the hundred and the hundred goes multiple days and multiple times in a day, but whatever it was, it used half my allotment.

to do something that i was like okay no that's not worth the time saved for me i mean you're taking 10 times as long as i would do it but if it's not me hey that's still worth it for me but not if you take half the tokens of the right fashion well that that's the uh uh that's sort of where we are in the crux of like all of those stories that are coming out of um

I agree.

And that is actually my process of going through things with Cloud Code and then sort of seeing, okay, so what part of what we just did should be spun off as a subagent, right?

Mm-hmm.

Because I swear to you, anything you do with the daily AI show content is giant because we're on for an hour.

One of the first things was downgrade that MP3 so that it is mono and the minimum that Scribe can work with.

Because we don't need high fidelity bitrate for transcribing word for word the way that Scribe does our show to get that transcript.

And the next piece is like, okay, maybe the transcript that then comes out of that, a diarized transcript...

Maybe that's assessed by a sub-agent as well, buddy, because I need you to have full context of what we're doing.

Those kinds of things.

And then the next piece is, oh, that's an automation, right?

Like that, the file drops, there's an automation that switches the bitrate, makes it a smaller file, sends it.

There's an automation that recognizes that that's complete, checks it periodically, and then launches the other pieces.

And that, like, I was talking to people yesterday.

I feel like my work, the more comfortable work that I do is a little more observation of a process than think the process out before we get into it.

And that's, for me, the shift from needing to be really good at prompt engineering and process, like, mentally working through it

And and being willing to fail at processes like it's it's fine.

The worst that happens is that they have no idea what I'm talking about.

And that is good information for me to figure out both ways versus trying to figure out why the custom GPT didn't take the next three actions that it was supposed to take.