Updated
Updated · Tom's Guide · Jun 27
Amanda Caswell Builds 1 Routing AI to Manage 5 Specialist ChatGPT Agents
Updated
Updated · Tom's Guide · Jun 27

Amanda Caswell Builds 1 Routing AI to Manage 5 Specialist ChatGPT Agents

1 articles · Updated · Tom's Guide · Jun 27

Summary

  • Caswell says a single routing agent became the most useful tool in her workflow by deciding which specialist GPT should handle each new task.
  • Five specialist agents — research, brainstorming, fact-checking, editing and review — had made her ChatGPT setup productive but cluttered, forcing an extra decision at the start of every project.
  • Her routing GPT evaluates each request against 5 questions: the goal, best specialist, whether multiple agents are needed, their order and what information is missing.
  • The setup starts with identifying 3 to 5 repeat tasks, building one custom GPT for each, then adding a router that can also absorb new specialists such as SEO, coding or image-generation assistants.
  • Caswell says the approach mirrors multi-agent AI systems like Sakana while keeping everything on one platform, aiming to make workflows smarter without adding complexity.

Insights

Could centralized 'routing agents' become bottlenecks, stifling the creative potential of decentralized AI collaboration?
As we build 'teams of agents,' what is the most critical skill for humans managing these new AI workforces?
How can AI systems be designed to reduce decision fatigue without removing essential human judgment from the loop?