TLDR: While many AI agents, like Manus, are built around large language models (LLMs), a new player, Pokee AI, is taking a different approach. Founded by former Meta lead Zheqing (Bill) Zhu, Pokee AI leverages reinforcement learning (RL) as its core decision-making engine, aiming for smarter planning and the ability to chain more tools for workflow automation.
In a significant shift within the burgeoning AI agent sector, Pokee AI, a new startup founded in October 2024 by Stanford Ph.D. and former Meta reinforcement learning lead Zheqing (Bill) Zhu, is challenging the prevailing reliance on large language models (LLMs). Unlike prominent AI agents such as Manus, which are deeply integrated with LLMs, Pokee AI is pioneering an approach centered on reinforcement learning (RL) for its core intelligence.
Zhu, who previously led applied reinforcement learning at Meta, where his team’s work on optimizing recommender systems and ad delivery reportedly ‘generated hundreds of millions in revenue,’ believes RL offers a superior path to smarter planning and decision-making in AI agents. He left Meta to establish Pokee AI, positioning RL as the primary decision-maker for the agent, rather than a mere fine-tuning layer for LLMs.
Pokee AI aims to automate complex workflows, performing tasks ranging from scheduling meetings to posting social media content across a diverse array of web tools. A key differentiator highlighted by Zhu is the agent’s enhanced capability to ‘chain more tools together in one go.’ In an interview with Tech in Asia, Zhu stated, ‘RL can be a very valuable model by itself when you train it properly. You’re no longer just generating tokens. You have to plan at a very abstract level.’ This means Pokee’s model can break down ambitious goals, such as executing a marketing campaign, into granular steps and then autonomously select and utilize the appropriate tools for each stage.
Zhu’s decision to build Pokee outside of Meta was driven by the need for broader access to external APIs. He questioned, ‘If Meta were to ask for YouTube’s API, would they ever get it? If I were to do it at Meta, [Pokee] would never be internet scale. It would be [within] Meta’s ecosystem.’ This strategic independence allows Pokee AI to integrate with a wider ecosystem of web services, enhancing its versatility and reach.
Also Read:
- Pokee AI Secures $12 Million Seed Funding to Revolutionize Online Workflow Automation with Advanced AI Agents
- Meta’s Advanced AI Agents Achieve Significant Gains in Kaggle Competitions Through Enhanced Search Strategies
The company’s innovative ‘RL brains’ approach, which swaps what Zhu refers to as ‘LLM bloat’ for a more focused RL-driven intelligence, is already attracting investor attention, signaling a potential new direction for AI agent development.
