Scott Watermasysk - September 25, 2026

One thing I’ve really leaned into over the last couple of weeks is having the agent (local/cloud/etc.) prove that its changes meet the desired goal.Codex excels here with computer use, and Claude has also done a great job using Chrome.But this isn’t the only way. Encourage your tool of choice to build something that helps it confirm, for sure, that it was successful.Anytime you find yourself doing something twice, it’s probably an opportunity to automate it—even if it’s as simple as refreshing the browser.