AI Doesn't Take Sick Days, But It Should

Anyone else notice AI has moods?

Some days it's sharp. On point. Exceeding expectations, no complaints.

Other days it's phoning it in like an employee who checked out three meetings ago. Same prompt. Same context. Same rules. Completely different output - and not in a good way.

It's like an employee that does just enough to not get fired, or an employee that doesn't even care if they get fired.

It's wild, but the more I use frontier models, the more I notice this. Great one day, lazy and sloppy the next. Same "employee", different attitude.

And here's the kicker - it's not the AI agents you built or the workflows you set up. Those are just following orders. It's the model underneath deciding, seemingly at random, how much effort it feels like giving today.

Yesterday I asked if you can trust your AI agents - should you trust your AI agents? Turns out that's only half the question. The bigger one is: can you trust the model they're built on to show up the same way twice?

Lately, my answer's been "not really". And it's part of why I've been leaning harder into open models and self-hosted setups. Say what you want about them - but at least when they're inconsistent, I know why. I'm not paying premium token prices for a model that decided to turn in an assignment that's only half complete.