About System One in Production AI
The industry is moving toward dual-model pipelines, highlighted by TypeSafe's launch of Jev: fast System 1 outputs from a fixed set, backed by a slower System 2 model for detailed reasoning 🤯
Plot twist: We’ve already built Sally around this exact two-pass pipeline since last year 👀
Running fast classification ahead of natural language generation solves two core issues:
- Latency
- Non-determinism
Watch how it works in practice.
Experience it here: https://t.co/GK54fMUJ1T 🚀
Our team already knows Claude Opus’s strengths and weaknesses. Switching now means paying that learning cost again.
We're small, so we optimize for jobs done, not cost. If we grow, we'll look at other open models.
We'd reconsider if you: drop the computer history, ship an alternative to Claude Design, and improve the model (we don't trust evals, we judge on results)