FIRST COMING · 1962-1973
Vetoed by Politics
OGAS (USSR) / Cybersyn (Chile). Technically entirely feasible, it died from power structures' instinctive resistance to transparency.
In 1962, Ukrainian scientist Viktor Glushkov submitted a groundbreaking proposal to Soviet leadership: to build OGAS (National Automated Economic Management System)—a computer network connecting all factories, stores, and banks across the Soviet Union, collecting production and consumption data in real time, with central algorithms dynamically adjusting resource allocation.
OGAS promised to replace bureaucratic intuition with real-time data and mathematical optimization. Glushkov estimated it would take 20 years and approximately 20 billion rubles—far less than the 50 billion rubles per year in military spending at the time. This was an entirely feasible project.
But the proposal was rejected in 1965. Rejected again in 1970. Shelved entirely in the 1980s. OGAS was never built.
The reason for rejection was not technology—the Soviet Union was the world's second-largest computer producer in the 1960s. Nor was it economics—the cost was far lower than military spending.
The real reason was politics.
Once OGAS was operational, it would give the center unprecedented precise information, meaning the "information intermediary" role of ministry bureaucrats would be dissolved. The gray power that local party committees gained through falsifying data would be completely lost. Middle managers at every level would lose their irreplaceability.
A system that could make everyone transparent
threatened those who derived their power from opacity.
Cybernetics was established by Norbert Wiener in his 1948 book of the same name. The core insight was:
Organisms, machines, and social organizations can all be described using the same mathematical language—the core being "feedback loops" and "information flows."
The cybernetics tradition was profoundly influential in the 1950s–60s:
Cybernetics gradually faded from mainstream discourse in the 1980s. But today's AI agent systems provide a perfect application scenario: autonomous decision-making, real-time feedback, multi-objective optimization, cross-domain collaboration—these are precisely the problems the cybernetics tradition is best at describing.
When a machine begins to pursue a defined objective,
it will precisely optimize that objective,
not the one we truly want. — NORBERT WIENER · 1960 · "MORAL & TECHNICAL CONSEQUENCES OF AUTOMATION"
OGAS (USSR) / Cybersyn (Chile). Technically entirely feasible, it died from power structures' instinctive resistance to transparency.
Amazon supply chains / Walmart inventory / Palantir data integration—doing almost exactly the same thing as OGAS, only serving capital rather than a planned economy.
Autonomous decision-making + real-time feedback + multi-objective optimization + cross-domain collaboration. Translating human fuzzy preferences, ethical judgments, and risk tolerance into machine-executable constraints—this is exactly the challenge Glushkov faced back then.
"Goal alignment" = "reward hacking," "feedback loop design" = RLHF training, "multi-agent coordination" = Beer's Viable System Model, "observability" = Ashby's requisite variety—every agent engineering problem has a cybernetic counterpart.
The biggest resistance to deploying agents within enterprises is not technical but organizational—middle managers realize that agents make the "information intermediary" role redundant. This is the same resistance Glushkov encountered in Soviet ministry offices in 1965.
The truly important factors remain unchanged: goal alignment, feedback design, accountability, political acceptability—these have been studied by the cybernetics tradition for 60 years. What has changed is:
The cost of ignoring these truths has shifted from a "chronic disease" to an "acute explosion."
In traditional software, poor feedback loop design causes systems to slowly decay. In AI agent systems, poor constraint and feedback design causes agents to go out of control within minutes—the real bottleneck is externalizing human high-level judgment into calibratable rules and feedback signals.
This is precisely why Glushkov failed: not because his technology was inadequate, but because he could not externalize the implicit power dynamics of the Soviet bureaucratic system into computable constraint conditions. Today's agent engineers face the exact same challenge.
Study cybernetics seriously. Not as historical knowledge, but as first principles for agent system design. Wiener's Cybernetics and Ashby's An Introduction to Cybernetics are more valuable than most agent framework documentation.
OGAS's failure was a political failure, not a technical one. Who has the right to define an agent's objective function? Who bears the consequences of an agent's decisions? Technical design cannot avoid these questions—avoiding them only causes political conflicts to erupt at a higher-cost position.
Cybertonia is a mirror. A system that is theoretically entirely feasible may never be realized due to resistance from power structures. The real bottleneck for AI agents may not be model capability, but organizational tolerance for transparency.