It waits on hold for you and calls when a human answers

Build it waits on hold for you and calls when a human answers.

Context: You need to reach your bank, an airline, an insurer. You know it's forty
minutes of hold music. So you put it off for three weeks and the problem gets
worse.

This calls for you. It sits through the menu, presses the right options, waits
on hold, and rings your phone the moment an actual person picks up. You answer
already connected to a human.

Everyone has this problem and nobody has solved it for normal people. The
existing versions are enterprise call-centre software sold to the companies
doing the holding, which is the wrong end entirely.

The hard part is the phone menus. Every company has a different tree, they
change without notice, and some deliberately make it hard to reach a person.
That's not a machine-learning problem so much as a maintained map — the top
few hundred companies cover most calls, and users correcting a wrong path is
the cheapest way to keep it current.

Second hard part: dropping you in at the right moment. Too early and you're
still listening to hold music. Too late and the agent has already hung up on
silence. It probably needs to say something like "connecting the customer now"
so the agent waits three seconds.

And it must never speak on your behalf. The moment it starts negotiating with
your bank, the trust is gone and the liability is real. It waits. That's all.

Requirements:
- Ship the smallest version that delivers the core value above, nothing more.
- Pick a stack you can deploy in a day, and state the choice before writing code.
- Handle the empty state, the error state and the offline case explicitly.
- Pricing to build toward: $3 per successful connection, or $9/mo for unlimited. Pay-per-use fits better —
this is something people need four times a year and would resent subscribing
to.

Start by listing the screens, then build them one at a time.