Symbolics

Contribution to AI

Lisp ran badly on hardware designed for Fortran, and the response was to stop compromising: build the machine around the language instead. Tagged memory words let the processor check types and bounds in hardware rather than in software, and the 3600 shipped an incremental collector with hardware assistance for ephemeral objects — David Moon's design, the first commercial use of generational scavenging, and now the default assumption in every managed runtime from the JVM to JavaScript engines. Genera made a different argument: one language all the way down, a single address space, sources shipped to the customer, and a running system that could be inspected, edited and recompiled without stopping. Much of what is now taken for granted in an integrated development environment was demonstrated there first and then lost for a decade. The Flavors object system, extended in-house as New Flavors, supplied the generic-function design that the Common Lisp committee absorbed into CLOS. Commercially the machines proved that AI research was a market someone could sell into, at roughly a hundred thousand dollars a seat — and then proved the reverse when stock workstations from Sun and the RISC vendors caught up on price and performance, taking the specialised-hardware case with them. The company's decision to keep its operating system changes from MIT's other licensee is also the direct provocation for Stallman leaving to start GNU.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).