You have to add a parameter rootDeviceHints to agent-config.yaml to make the installer know the disk to install to. If you have an enterprise grade infrastructure you will find multiple disks of different flavor and usage on a host. Figuring out the right device means booting a RHEL9 or CoreOS and looking through /dev/disk/by-path or issue commands like
udevadm info --name abcd|grep -i "serial\|wwn"
to get the right device hints. If you fail, and in more complex environments, you might need some tries, you have to change the yaml, create the image again and reboot your hardware, which needs at least 15 minutes with modern hardware. Put your shoes into a VSphere admin and you know, that this will never work!
Instead of telling during the installation, "dev tralalal is not available - installation failed", we have now (since 25 years) the possiblity to open up a text ui to select the disk. We already do this, when the network configuration is wrong. We should be able to do the same, when the disk is not found. In fact we should, like with the network, ask all the time for sanity reasons, whether the device is the correct one, with a timeout of 15 secs.

    1. Create an agent-config.yaml.
    2. Add a wrong rootDeviceHints.
    3. Create and boot the image.
    

The installer fails without any hints, besides that it can't find the defined disk.

    The installer should just ask for the right disk to push the image to. Like with the network it should ask this all the time for sanity check.

    Could be an RFE, but for VSphere admins it is seen as a severe bug!