Around 2014 I was in school and working at a bar where schedules were done by hand each week. I figured I could easily build a little web app to handle this.

The rules were impossible to encode because they didn't really exist. Building the schedule by hand allowed the manager to draw on their knowledge of everything from employee work ethic and personality to expected rushes (e.g. hockey nights) and reservations.

Are you arguing that it’s not possible to articulate all the rules? It seems as though you’ve done a good job highlighting a good number of constraints and edge cases.

Further iteration and feedback when implicit/undocumented rules are violated will get you closer. It’s not like there’s an infinite source of rules, and many of these are just general knowledge that does exist online. You can eventually get there.

If we look at traditional project roles, it's the responsibility of the Programmer to articulate the Business Logic in a way that can be interpreted by a computer (i.e. write the rules in code, using their chosen programming language).

It's the responsibility of a Business Analyst to study a work process, and articulate the rules which govern the process.

With vibe-coding, the agent is usually playing the role of a Programmer. Are they also playing the role of a Business Analyst?

If we look at iterative development, that's usually the responsibility of a Product Owner.

With agentic development, we expect AI to perform the role of developer. In this example, is the bar manager able to perform the role of BA or PO? Is the agent able to accommodate and fill in the blanks that a BA or PO would normally fill?

There may also be human edge-cases, such as "Don't schedule Dave and Brian together, they don't get on".

Agentic AI doesn't have the nuance for all these factors. Many situations (such as a bar manager or restaurant manager with a small venue) may be better off with simply having a human plan the schedule on paper.