3 sessions on qemu with a coding agent: the agent failed at least once in 3, and 0 finished without a human stepping in.
Takeaway: Next time I'd nail down the QEMU/OVMF/WSL boot pipeline first before layering framebuffer and GRUB experiments on top of it — half the pain was Windows/WSL path translation and stale OVMF vars, not the actual kernel or userland code.
Takeaway: Next time I'd start the prompt by stating upfront that I'm on WSL2/openSUSE driving QEMU from Windows, since half the wasted turns came from the agent trying WSL bash/PowerShell bridging that kept mistranslating paths.
Takeaway: Next time I'd specify upfront that MB1 must stay fully working as a real fallback, not just compile — that black-screen/no-console debugging loop cost the most time.