Real Automation Eventually Reaches a Screen
APIs cover a lot of infrastructure. Installers, recovery environments, graphical tools, and legacy software still put a screen in the way.
Infrastructure APIs only take you so far. Operating-system installers, BIOS and UEFI screens, recovery environments, login prompts, graphical configuration tools, and older applications still require someone to interact with what is on the screen. If infrastructure AI is going to operate real systems, eventually it has to deal with those interfaces too.
During one of our development cycles, STRATUM AI had to recognize that a command-line window was running inside a graphical desktop, click inside the terminal to establish keyboard focus, capture a fresh screenshot, type the required command without immediately submitting it, verify that the exact command was visible on screen, and only then press Enter and continue the Mission.
That sequence is small, but it captures the architecture we want. Screen observation and computer use are additional tools inside the governed Mission. Installed STRATUM Guest Tools remain the preferred interface when structured guest operations are available, while command execution, Serial, and graphical console control give the AI additional paths when the software or installer only exists on a screen.
This lets the AI operate the same awkward systems humans operate today while keeping every step inside STRATUM's scope, authorization, credential, and verification boundaries.