Why a good AI partner makes you think more - IBM

TEXT START: For twenty years, every designer, product leader and consultant chased the same goal: remove friction.

The Dissection

This is a transition-management memo disguised as a human-centered design argument. It admits that frictionless AI erodes expertise, judgment and the career ladder, then proposes engineered friction as the remedy.

The article identifies a real failure mode: automation bias, passive verification and the disappearance of apprenticeship work. But it relocates a system-level labor crisis into workflow design. IBM’s solution is to keep humans “in the loop” through checkpoints, explanations, training and recertification—creating demand for governance, consulting and organizational redesign while preserving the appearance of human control.

Its key sleight of hand is treating better human-AI cooperation as evidence that humans remain economically necessary. The cited revenue and margin gains show that some firms manage adoption better. They do not show that AI preserves aggregate employment or productive participation.

The Core Fallacy

The article confuses preserving human judgment with preserving human economic indispensability.

Under the Discontinuity Thesis, the question is not whether people can be made to think more while using AI. They can. The question is whether competitive organizations will continue paying for that thinking when AI can perform the same cognitive function faster, cheaper and at scale.

“Friction by design” is therefore a lag defense. It may slow skill atrophy and reduce error during the transition, but it also adds cost and delay. Once AI can handle verification, explanation, judgment and exception management well enough, competitive pressure removes the human checkpoints. A human who remains in the loop only because a policy requires it is an audit ritual, not a sovereign economic actor.

The slogan “AI won’t replace people, but people who use AI will replace people who don’t” is labor-market triage, not a refutation of displacement. It describes early relative advantage among workers before the system reduces the number of workers it needs.

Hidden Assumptions

• Human judgment will remain scarce, reliable and superior after AI systems improve at judging.
• Firms will voluntarily preserve expensive human involvement even when it reduces speed or margin.
• Explainability produces meaningful understanding rather than post hoc rationalization.
• A human reviewing an AI output remains productive rather than becoming a liability, rubber stamp or legal shield.
• Reskilling can keep pace with capability releases instead of chasing a moving target.
• Early-career workers can build expertise through redesigned workflows after the apprenticeship layer has been automated away.
• Better organizational adoption translates into broad labor demand rather than higher output with fewer employees.
• Institutional rules can stabilize human-only domains despite competitive pressure to automate them.
• “Human in the loop” means human power. Often it means human responsibility without human control.

These assumptions collide with P1, P2 and P3. If cognitive automation becomes durably superior, institutions cannot preserve large human-only economic domains, and the majority lose access to economically necessary labor. The article treats those conditions as design problems. They are structural constraints.

Social Function

Primary classification: transition management and elite self-exoneration.

Secondary functions: corporate prestige signaling, ideological anesthetic and product propaganda.

It gives executives a morally acceptable story: AI can be adopted aggressively while humans supposedly retain judgment. It gives workers a behavioral instruction: adapt, verify and reskill. It gives IBM a commercial territory—AI governance, change execution, explainability, operating-model redesign and continuous recertification.

The piece is not pure copium. Its account of skill erosion and automation bias is materially credible within the supplied text. Its failure is more useful than simple denial: it recognizes the wound, then sells a bandage as if it were a new skeleton.

The Verdict

This article does not solve the discontinuity. It manages the waiting room.

Designed friction can preserve competence, catch errors and delay dependency while AI remains imperfect. It cannot defeat the competitive logic that eventually automates the checkpoint itself. The human role moves from operator, to verifier, to accountable observer, and then—unless protected by law, institutional inertia or a temporary liability moat—out of the productive circuit.

The article’s “good AI partner” is therefore a hospice model for human economic participation: humane, useful and temporary. It may improve the transition. It does not reverse the death of the mass employment-to-consumption system.