In 1964, while most computers were room-sized beasts fed by punch cards, engineers Tom Ellis, Malcolm Davis, and Keith Uncapher at the RAND Corporation in Santa Monica unveiled something extraordinary: a flat, 10.24-inch-square writing surface you could draw on with a stylus, with the computer recognizing your handwriting in real time. They called it the RAND Tablet, sometimes marketed as the Grafacon.

The hardware was an engineering marvel for its era. Beneath a Mylar surface lay a printed circuit grid of 1,024 × 1,024 capacitively-coupled conductors — a true 1024×1024 digitizer at a time when most CRT displays struggled to show 512 lines. A stylus tipped with a small capacitor picked up Gray-coded pulses from the grid, achieving roughly 0.01-inch resolution. Sample rate: about 200 points per second. Connected to an SDS 940 timesharing system, it ran GRAIL (Graphical Input Language), developed by Gabriel Groner in 1966 — software that recognized hand-drawn flowchart symbols, printed letters, and gestures like scratch-out-to-delete. Watch the original 1968 demo film: it looks uncannily like an iPad.

Why did it die? Cost and context. Each tablet cost around $18,000 in 1964 dollars (roughly $180,000 today), and required a timesharing mainframe to do anything useful. The CRT display sat separately from the writing surface — the eye-hand disconnect that plagued graphics tablets until the iPad finally fused screen and digitizer in 2010. ARPA funded a handful for research labs; commercial buyers balked. By the 1970s the mouse, vastly cheaper to manufacture, won the input wars. Alan Kay saw a RAND Tablet at Utah in 1968 and it became one of the seeds of his Dynabook concept — but Xerox PARC ultimately chose the mouse for the Alto.

The deeper tragedy: GRAIL's gesture recognition was better than what shipped on the Apple Newton in 1993. Ellis's team had solved scratch-to-delete, character segmentation, and stroke-direction parsing by 1968. That research literally sat in DTIC reports for thirty years.

Why revisit it now? Three things have changed:

• Capacitive grid manufacturing is essentially free. Every smartphone touchscreen is a descendant of the RAND grid. The 2026 BOM cost for a 1024×1024 digitizer is under $5.
• On-device neural handwriting models (Apple's Scribble, Goodnotes, the open-source mathpix-ocr family) now exceed GRAIL's recognition by orders of magnitude — and run locally on an M-series chip.
• The "command surface" idea was the real invention. RAND wasn't building a drawing pad; they were building a programming environment where you sketched flowcharts and the computer compiled them. That's exactly what tools like tldraw computer, Excalidraw + LLM pipelines, and visual programming environments are rediscovering in 2025–2026 — except RAND had the integrated stylus/recognition stack 60 years earlier.

A modern reconstruction would cost about as much as a mid-range Wacom and could run GRAIL's entire 1968 feature set on a Raspberry Pi 5. Several research groups (notably Bret Victor's Dynamicland and the Ink & Switch lab) are explicitly mining this vein. The RAND Tablet wasn't a failed product — it was a correct product released into a world that couldn't afford it.