Your Agent Doesn't Need an LLM for Every Decision - Raymond Weitekamp, OpenProse

Categories: AI, Tools

Summary

Decision models (system one AI) that make fast, convergent judgments are the counterbalance to generative LLMs' infinite divergence—enabling AI agents to reliably pick winners rather than endlessly generate options. Raymond Weitekamp argues reasoning tasks can be solved through many tiny fast decisions instead of one expensive LLM call per problem.

Key Takeaways

  1. Decision models rank and pick among options rather than generate open-ended text, creating a yin-yang dynamic with LLMs. This convergence property solves the 'slop problem' where generative outputs still need human or agent filtering.
  2. Recursive decision models combine recursive language models (MIT research, 1+ year old) with decision model architecture. This enables hierarchical reasoning where sub-models call other sub-models to solve complex tasks through nested small decisions.
  3. Reframe expensive reasoning tasks: instead of querying an LLM for one big deliberate decision, decompose into many fast system-one judgments. This mirrors how human brains operate (Kahneman's dual-process theory) and reduces token costs.
  4. Generative AI has infinite divergence (always more tokens, more variation); decision models have forced convergence. This matters for production systems where you need deterministic choices, not endless options requiring downstream filtering.
  5. Typesafe's Jamba sparked engineer excitement not for generation but for convergence—positioning decision models as the missing counterweight to generative dominance. Most community calls them 'decision models' rather than Typesafe's 'system one' branding.

Related topics

Transcript Excerpt

Hey, my name is Raymond Whitampamp and my talk today is originally titled recursive decision models. And what I think I've realized since submitting that title is what it's really about, what we're really going to be talking about today is operating system one. Okay. So maybe some quick definitions or what what does that mean? Uh so system one models the the um naming of system one models from type safe when they released Jev um not even two months ago now I think um came from uh this concept that Danny Conaman popularized uh in thinking fast and slow. this idea of like the system one part of our uh brains is making very fast and intuitive like judgment calls, gut reactions, reflexes, maybe maybe reflex is more uh purely neurological but um that idea of like a lot of the decisions that we …

More from ai.engineer