Don't start with AI. Start with the work | CIO

TEXT START: Before asking where to add AI, ask where better information would change the outcome, then redesign the work with the people who actually do it.

The Dissection

This is an enterprise AI adoption manual disguised as a human-centered work philosophy. It correctly identifies the prerequisites for useful automation: workflow mapping, contextual data, decision rights, integration, feedback, and outcome-based measurement.

Its deeper function is harsher. It teaches CIOs to make human work legible, standardized, measurable, and therefore automatable. “Design with people” also means extracting workers’ tacit knowledge, exceptions, and failure points before encoding them into systems that require fewer workers.

The Core Fallacy

The article treats keeping a person accountable as equivalent to keeping that person economically necessary. Under Discontinuity Thesis mechanics, those are separate conditions.

Human-in-the-loop control is a lag defense, not a permanent moat. As reliability, integration, legal delegation, and verification improve, accountability can be concentrated in fewer supervisors or shifted into institutional processes. The article presents augmentation and automation as enduring alternatives, but competitive pressure pushes toward automation whenever it produces the same outcome at lower cost.

Its own recommendations accelerate P1 and P2: better context, cleaner workflows, and clearer decision rights increase AI’s cost and performance advantage. The article solves the engineering problem of automating work, then describes the residue as “higher-value work” for humans.

Hidden Assumptions

• Productivity gains will become employee time and better work rather than layoffs, wage compression, surveillance, or intensified output demands.
• Expertise, empathy, and judgment will remain durable human monopolies.
• Human accountability will remain mandatory even after AI decisions become reliable enough to supervise at scale.
• Worker participation will provide influence rather than unpaid process intelligence for their own replacement.
• Trust and experience metrics will constrain deployment instead of being subordinated to cost and competitive pressure.
• Firms can redesign roles and decision rights without confronting ownership and bargaining-power conflicts.
• The distinction between desk-based and frontline work will remain economically meaningful after both are deeply instrumented.
• AI activity and outcome metrics can measure value without addressing who captures the value.

Social Function

This is partial truth wrapped in transition management and ideological anesthetic. It is not empty copium: the operational advice is sound and would make enterprise automation more effective.

But its humane vocabulary launders displacement. “Help people make better decisions” conceals the possibility that fewer people will be needed to make those decisions. “Human accountability” preserves managerial legitimacy while decision capacity migrates into machines and the owners of the infrastructure. Workers are invited to help design the machinery that reduces their bargaining power.

The text functions as a managerial alibi: deploy AI responsibly, involve employees, measure outcomes, and avoid stating the terminal implication—that successful redesign destroys the mass labor requirement on which the post-WWII wage-consumption circuit depends.

The Verdict

This article is not a rebuttal to the Discontinuity Thesis. It is a polished field manual for implementing it.

It correctly explains how to turn AI from a novelty into an integrated production system. It refuses to follow that logic to its endpoint. Under P1, P2, and P3, the likely result is not better work for everyone but productive participation for a shrinking minority, with the rest managed through transfers, supervision, and institutional containment. The article is the clean, polite face of obsolescence: an automation blueprint written in the language of human stewardship.