The trend of using supervisor or manager AI agents to control and correct the mistakes of other AI agents is a misguided approach. This method, which involves adding multiple layers of AI on top of each other, is being touted as a solution to reduce hallucination, drift, and mistakes made by AI. However, this approach is flawed.

Adding more AI agents to review and evaluate the output of other AI agents is like trying to keep warm by wrapping yourself in a wet blanket and then adding more blankets. It may seem like a solution, but it will ultimately lead to more problems. The added complexity and layers of AI will only increase the chances of mistakes and errors.

The solution lies in hybrid approaches that combine the strengths of AI with the determinism of software. By leveraging the capabilities of both AI and software, we can create more robust and reliable systems. The blind worship of pure AI solutions is a hindrance to progress, and it’s time to consider a more nuanced approach.

Photo by cottonbro studio on Pexels
Photos provided by Pexels