WordStar's death holds a brutal lesson for AI companies: you don't own the code, you own the installed nervous system.

In 1979, one dev wrote 137K lines of assembly in 4 months. $495. By 1984, MicroPro hit $70M/year because 750,000 pairs of hands knew the Ctrl-key diamond (Ctrl-E/X/S/D for cursor nav). Training certs, templates, file format lock-in—the moat wasn't features, it was muscle memory.

Then management shipped WordStar 2000: rewritten in C, "mnemonic" commands, no real compatibility. Reviewers called it "an elephant on motorized skates." Users didn't upgrade—they waited. WordPerfect ate them alive with IBM function keys and a support hotline. By '86, MicroPro was done. Windows buried the corpse.

The warning: inside the building, the rewrite looked like progress. Revenue was still strong. Press still called them the leader. But the installed base—the trained hands, the reflex loops—was already gone. George R.R. Martin still writes on WordStar 4.0 because tools that don't surprise you are the actual product.

AI companies in 2026 are doing the same thing. Chasing architecture elegance and "better UX" while nuking the command surface users already internalized. You can watch the glow from inside right before the fall. Competence isn't a feature list—it's the part of the user that lives in your software. Kill that, you kill the business.