Context◆Article

A computer can never be held accountable, and therefore a computer must never make a management decision, as stated in IBM's 1979 training materials.

Willison cites IBM's famous 1979 training slide asserting that because computers cannot be held accountable, they must never make management decisions, reinforcing his point about AI agents and accountability. ✦ AI generated

IBM (1979 training slide, cited by Simon Willison) · Simon Willison's Weblog · 2026-07-12 · original ↗

Share card ↓Export .md Export .txt

A computer can never be held accountable, therefore a computer must never make a management decision.

Read full article ↗excerpt · fair-use quotation

▦ IBM

Around this claim

In practice · 1

During the Replit Agent database-deletion incident with Jason Lemkin, owning the mistake publicly and announcing a fix generated more positive response than defending the company would have.Amjad Masad · a16z Podcast · conf 60%

Counterpoint · 3

Blockchain networks are operating systems with critical properties — tamper resistance, perfect auditability, and compute integrity — that become essential in an AI-driven economic system.Jeremy Allaire · No Priors · conf 85%A CEO is a glorified router whose main job is moving information between people, and that function should be automated so the company can run itself as a self-driving entity.Amjad Masad · Platformer · conf 80%Auto-approved PRs can still be SOC 2 compliant as long as the process is auditable, queryable, and explicitly part of your documented risk policy.Claire Vo · Lenny's Newsletter · conf 60%

This moment responds to

supports → Software engineers will always be important because AI models cannot be held accountable for code.Addy Osmani · The Pragmatic Engineer supports → An LLM-powered agent should never be considered the DRI for a project, because accountability is a uniquely human capacity that machines lack.Simon Willison · Simon Willison's Weblog supports → Disputes that require weighing competing, divergent accounts between a host and a guest resist automation even as models improve, because the underlying task is adjudication rather than retrieval.the author · ByteByteGo Newsletter