I actually think its the inversion of what Andrej Karpathy said at AI Startup School. Applications can use multiple llms models and how they go about making those tradeoffs in speed (latency), accuracy (intelligence and tool use) gives them their edge or taste as we like to call it. It's your positioning of your product which attracts a certain type of user.

This is vastly different his characterization on cursor using a single model which is never the case. The future is multi-modal (providers) and multi modality IO.

Prepare for “teamwork” agents/apps powered by multiple communicating MCPs, at DuckAi.ai and OpenPond.ai we’re tackling this with the upcoming new releases 🦆

software artist | building /outpaint | ex: @coinbase ex: bisontrails | seeker of the flow state | artivilla.com