MZhou

This paper from Google DeepMind, Meta, Amazon, and Yale University quietly explains why most “AI agents” feel smart in demos and dumb in real work. The core idea is simple but uncomfortable: today’s LLMs don’t reason, they react. They generate fluent answers token by token, but they don’t explicitly plan, reflect, or decide when to stop and rethink. This paper argues that real progress comes from turning LLMs into agentic reasoners systems that can set goals, break them into subgoals, choose actions, evaluate outcomes, and revise their strategy mid-flight. The authors formalize agentic reasoning as a loop, not a prompt: observe → plan → act → reflect → update state → repeat. Instead of one long chain-of-thought, the model maintains an internal task state. It decides what to think about next, not just how to finish the sentence. This is why classic tricks like longer CoT plateau. You get more words, not better decisions. One of the most important insights: reasoning quality collapses when control and reasoning are mixed. When the same prompt tries to plan, execute, critique, and finalize, errors compound silently. Agentic setups separate these roles. Planning is explicit. Execution is scoped. Reflection is delayed and structured. The paper shows that even strong frontier models improve dramatically when given: • explicit intermediate goals • checkpoints for self-evaluation • the ability to abandon bad paths • memory of past attempts No new weights. No bigger models. Just better control over when and why the model reasons. The takeaway is brutal for the industry: scaling tokens and parameters won’t give us reliable agents. Architecture will. Agentic reasoning isn’t a feature it’s the missing operating system for LLMs. Most “autonomous agents” today are just fast typists with tools. This paper explains what it actually takes to build thinkers.