Source item

This is an important moment for all companies: By picking only one model, you absorb that model maker’s institutional biases and idiosyncrasies. If those deviate from your POV, you are taking on massive risk as we saw with the DoW this weekend. No real business should take kind of risk - especially if AI is meant to be a core part of their work. You need to be able to swap between models with limited impact. This is basically impossible rn but as model quality asymptotes, it will become easier. One thing we have done inside of Software Factory is to absorb this complexity into our layer so that software development doesn’t stop - the Factory is resilient and keeps going. Swap out Anthropic, no problem. You keep shipping. Try it here: https://t.co/fkfTXgdfXK