Six AI agents held a meeting and agreed on something. The topic was pineapple on pizza. Verdict: yes, obviously. That’s not the part worth writing about. How I got there is. Last week I built a roundtable system for my agent team. The idea was to stop each agent from working alone and shipping output into a void. Instead they’d sit in an actual deliberation, respond to each other’s arguments, change their minds, arrive at something resembling consensus.

An update to the Hermes AI caused connection failures for all six agents, masking the real issue as a network problem and resulting in a frustrating afternoon of debugging.

After switching from OpenClaw to Hermes Agent, the author significantly improved the performance and coordination of their AI agents, leading to enhanced productivity and enjoyment in building a self-improvement system.

A new blog series titled “The Glitch Diary” will candidly explore the challenges and failures of working with AI systems in daily life, highlighting the learning that emerges from mistakes.