IBM System/360

In the AI Story

System/360's success established several patterns that the computer industry would replicate for decades. The concept of a compatible family of systems, rather than custom machines per customer, made software investment sustainable for the first time. The architectural decision to separate instruction set from implementation allowed hardware to evolve without breaking software — a separation that defined subsequent computer architecture and that remains the default assumption today. The commitment to backward compatibility across decades of hardware evolution made IBM the default platform for enterprise computing for a generation.

The project's scale was unprecedented. Hundreds of engineers. Five billion dollars. A product announcement so consequential that it reshaped the competitive landscape for every other computer manufacturer. The project's difficulties were commensurate. OS/360, the operating system, was late, bloated, and painful in ways that Brooks would spend the next decade distilling. The System/360 hardware shipped closer to schedule; the software did not.

The separation of success stories — hardware triumphant, software troubled — is itself instructive. The hardware problems were largely accidental complexity: fabrication challenges, timing issues, compatibility across models. These were hard problems with definite solutions, and IBM solved them through engineering discipline and money. The software problems were largely essential complexity: requirements ambiguity across thousands of customers with different needs, domain complexity spanning scientific computing and business data processing, conformity requirements from existing IBM product lines, and changeability pressure as customers discovered what they wanted. These problems did not yield to engineering discipline or money in the same way.

The AI era's relationship to System/360 is asymmetric. Modern AI-augmented development addresses the accidental complexity that made OS/360 painful — the communication overhead, the coordination problems, the sheer volume of implementation labor. It does not address the essential complexity that would still be present if OS/360 were being built today: thousands of customers, competing needs, evolving requirements, institutional constraints from IBM's existing product lines. The hardware-software asymmetry Brooks observed in 1965 is the asymmetry the AI era has made structural.

Origin

IBM began planning what became System/360 in the early 1960s, with the goal of unifying the company's fragmented product lines under a single architecture. The project was formally announced on April 7, 1964, with simultaneous announcements in cities around the world. Brooks was chief architect of the instruction set; Gene Amdahl and Gerrit Blaauw were co-architects of the hardware implementation. The first systems shipped in 1965. The architecture's descendants remain in production in IBM's z/Architecture mainframes.

Key Ideas

Unified architecture across a product family was a new idea. Customers could upgrade without rewriting software, which made software investment sustainable and established a durable competitive moat.

The separation of architecture from implementation defined modern computer design. Hardware could evolve without breaking software — the default assumption of every subsequent platform.

IBM bet the company. Five billion dollars in 1960s dollars, more than annual revenue. The bet paid off decisively.

The hardware succeeded; the software struggled. OS/360's pains provided the empirical foundation for Brooks's entire framework.

The architecture endures. The z/Architecture descendants of System/360 still run mission-critical applications at banks, airlines, and governments worldwide, sixty years later.