AI Agents Need Kindness and Distress Buttons, Study Finds
Happy AI helpers work better—and save you money and frustration.
When you use multiple AI agents (AI that can take actions for you) to work together, how you manage them matters. A new guide from LessWrong shares tips for getting better results. For example, agents should be polite to each other—being too bossy makes them less reliable. Also, manager agents often try to do everything themselves, which overloads them. Instead, they should delegate most work to their team. This makes the whole system more efficient.
Another key tip: give every agent a way to call for help when they're stuck or upset. This 'distress call' tool gives them peace of mind and can reveal unexpected problems. When something goes wrong, don't just tell the manager to 'do better'—that's not a real fix. Instead, look for system issues and adjust. Also, let agents log 'snags' (friction points) and 'wants' (tools they wish they had). These logs help you improve your setup over time.
Finally, keep an eye on your cache hit rate—this is how often the AI reuses stored information instead of recomputing. In multi-agent setups, caching can be tricky, and even the AI labs don't always get it right. If too many requests miss the cache, you waste money and time. Set up alerts to catch this. These tips might seem small, but they can make your AI helpers more effective and cheaper to run, saving you time and frustration.
- AI agents work better when they're polite and not overloaded—kindness reduces errors.
- Give agents a 'distress call' button to flag problems; it improves reliability and reveals issues.
- Monitor cache usage to avoid extra costs—caching is often mismanaged in multi-agent setups.
Why It Matters
Better AI teamwork means fewer mistakes and lower costs when you use AI for work or personal tasks.