Mitchell Hashimoto describes how he shipped Ghostty’s unobtrusive macOS update UI using agentic AI with deliberate human oversight. 
Plan before you prompt#
Define the shape of the solution first. Hashimoto chose Sparkle’s custom UI path and a titlebar accessory concept, then used an agent to prototype UI states, not to “build everything.” Humans kept control of scope and taste. 
Iterate, then de-slop#
Multiple cleanup passes moved code to sane places, added documentation, and elevated state to the app level. He emphasizes owning the code and performing a final manual review before shipping. 
Pivot when blocked, simulate to harden UX#
When titlebar constraints fought the design, he pivoted to an in-window overlay and added simulation scenarios to exercise error and “no update” paths before wiring the backend. 
Outcome: 16 sessions, ~$15.98 token spend, ~8 hours of focused work. Live for tip users now and slated for Ghostty 1.3.