Agentic infra can beat LLMs alone on compositional generalization
prasanna_says · x · 2026-07-21
The post argues that the output of agentic infrastructure is greater than the sum of an LLM plus its harness.
It connects this idea to compositional generalization: if a model can solve one task, it does not necessarily generalize to a structurally similar unseen task. A harness can help by making two problems look the same through context offloading and programmatic sub-calling, which may let models trained on easy tasks generalize better to harder ones.
The author suggests this pattern matters not only for closed-loop domains like coding and math, but also for other decomposable tasks where humans may currently provide the missing loop via taste and selection of outputs.
More from coding & agent
- DeepMind paper: LLMs can't sample randomly on their own, only convert given seeds — rohanpaul_ai · 2026-10-27
- DeepMind paper: LLMs can't sample randomly on their own — pass random numbers from code instead — rohanpaul_ai · 2026-10-27
- BELL: Open-Source Stack Combines Browserbase, Eve, iMessage and Stripe Link for Personal Agents — jeff_weinstein · 2026-10-27
- Legora runs 2,000 parallel agent evals on Modal Sandboxes for legal AI — blelbach · 2026-10-27
- Two engineers rebuilt a FedEx supplier's billing core in 3.5 months vs an estimated 7-8 — alex_verem · 2026-10-27
- GPT-6.1 Terra scores 48 on AA Coding Agent Index at one-third the cost of GPT-5.6 Terra — ArtificialAnlys · 2026-10-27