token impact

[bashfulrobot]

Hi there, I've been testing OpenSpec with success on Codex using the 2x entitlements I currently have until April. I'm just curious what others have seen in terms of token consumption using OpenSpec over time. And given how OpenSpec is architected, do archive items ever come into play for token consumption? Or any of the OpenSpec files for that fact. My assumption is that the OpenSpec files don't come into any token consumption unless you call the prompt. And even then, I may assume that anything archived is no longer in scope.

Is this a fair understanding?

Cheers!

[bashfulrobot]

I think I may have answered my own question here. I just tried it out with Claude and blew through my daily allotment in a relatively short time. Granted, I'm only on the pro plan at the moment.

Definitely drives better outputs, but just goes to town on the token and limits.

[bashfulrobot]

Maybe the thing to do here is transition the conversation to which models to use for what, to optimize token usage.

[TabishB]

Yes, getting the agent to be more accurate tends to use more tokens, but I feel like sometimes these kinds of questions tend to miss a few other things:

• How many tokens would you have spent otherwise, course-correcting and getting the agents to fix their own mistakes?
• How much extra time would you be spending as a developer otherwise? (How much is your time worth?)

My personal view is that not all token usage is the same. Burning tokens to ensure correctness alignment, etc, is well worth it compared to burning tokens to get inconsistent outputs.

[bashfulrobot]

You are 100% right. I think what I was looking for was simply to gauge what an increase from nothing to something might look like. Then I can strategically choose when to use it and when not to. The reality is, I probably just need to use it a couple of times and just see how it goes. But I was just hoping to capture some insight from someone who had a period.

Appreciate the feedback.