First week of beta: what broke, what we fixed

Our first tester never saw the second screen. An honest account of the first days: three real defects, what they taught us, and what remains open.

Seven days of beta, three real defects. The most serious blocked our first tester at the welcome screen: ‘no assistant detected’, on a machine where everything was installed. The fix went out that same day in an update published that evening. Here is what the first week broke, in order, without embellishment.

The bug nobody could see

On the tester's machine, the AI assistant worked in the terminal. Maestro, launched with a double-click, could not find it. The cause lies in a subtlety of macOS: an application opened from Finder does not see the same installation paths as the terminal on that machine. Tools installed by a version manager, a common practice among people who install many of them, became invisible.

This defect was invisible to us: our development machines launch everything from the terminal, where the problem does not exist. It took a real user, a real machine and an anonymous error report showing the exact category of blockage. Maestro now looks for paths where the terminal finds them, and the update shipped that same evening.

Half the feedback asked for things that already existed

The second lesson cost our egos more: in the feedback from the first full test, half the requests concerned existing capabilities. Resuming an existing project, switching assistants along the way, reverting to an earlier version: everything was there, and nothing was visible.

A capability the user cannot find does not exist. We reworked the welcome page and first screens so important actions present themselves at the right time instead of waiting to be discovered. It is the same principle behind our rejection of the terminal: the burden of clarity belongs to the product, not to you.

The counter that was talking nonsense

Third defect, found while checking everything else: the usage counter displayed incorrect figures. It ignored part of what assistants bill for, the working memory reused between exchanges, and therefore underestimated each project's consumption.

A tool promising to let you direct the work must first count accurately. The counter was rebuilt and checked again against the assistants' actual statements. If you retain one criterion for judging tools in this family, make it this one: ask where their figures come from.

What remains open

Not everything is resolved. One tester encountered macOS refusing to open their projects folder, and Maestro's error message sent them down the wrong path by referring to a missing tool. The real culprit was a system permission to grant in Settings. The honest message, naming the permission and the route to grant it, is written and awaiting the next update.

These accounts rely on anonymous error reports: categories of failures, never your project contents, as our page about your data describes. That let us see the first tester's blockage without ever seeing their screen. The beta continues by invitation, and each week will break something else: we will keep telling the story here.