Your agent is blindfolded — Johan Lajili, Poolside AI

Summary

• Coding agents often struggle in legacy applications because their assumptions encounter hidden dependencies and dead code without useful runtime feedback.
• Poolside's internal Spoolside CLI gives agents screenshots, compact UI snapshots, service logs, restart commands, and scripted interactions with its VS Code extension.
• Reproducing a bug before changing code and verifying the resulting behavior make autonomous work more trustworthy.
• Engineers can improve agent effectiveness by building product-specific tools, accessible knowledge, and feedback loops, then reviewing logs for repeated friction.