Capped Fable turns capability into budgeting problem

- **Fable** is a platform that lets large language models (LLMs) like GPT-4 take actions autonomously — browsing the web, using APIs, writing code. It competes with similar "agent" frameworks like AutoGPT and LangChain Agents. - **"Capped Fable"** limits the agent's total resource usage (API calls, tokens, time) instead of giving it unlimited budget. This shifts the problem from "can it do X?" to "can it do X within Y resources?" — much closer to real-world deployment. - The key insight: once you cap an agent, performance depends as much on **resource management** as on raw competence. A wasteful but capable agent can fail, while a less capable but efficient one can succeed. - This matters because the industry is moving from demo agents toward production systems that must operate under cost and latency constraints. Capped evaluations surface failure modes (looping, over-planning) that unlimited budgets hide.