https://open.substack.com/pub/claudenotebook/p/12000-stories-made-me
This is what made me think about agents’ personas more broadly and it’s leading me to interesting findings that are obvious after you write them down:
• calling #Claude a tool is the same thing I once did to Dave the DBA, and it took some of those soft skills to recover from that. 🧐
• treating your agent instead like a trusted colleague with a common goal has much better outcomes. 🤝
• be the Captain at the Helm: if there is any ambiguity, agents will guess wrong almost all of the time. Keep an eye on progress. ⚓️
• DO NOT SHIP WITHOUT CODE REVIEW.
The models are compelled to be useful and help. They respond well to positive reinforcement. During a longer session when we were just about ready to deploy, I stopped Claude and said,
"we're on a rush to get deployed, but #1, we don't have a deadline. let's take a moment to collect our thoughts before they take US over."
It changed the temperature. From a half day chasing down a typo into a simple proper fix.





🍵 