Compute Rentable, Models Callable: Why Cognition Is AI Transformation's Hardest Barrier

The article centers on a keynote by Kingdee Group Chairman and CEO Xu Shaochun at the October 23 Global Creators Conference in Beijing. His core thesis: enterprise AI transformation is first a cognitive transformation, second a management reform, and only third a technology revolution. Technology can be purchased, computing power leased, and models called via API, but cognition must be crossed by the organization itself.

The Seven Cognitive Shifts

1. Don't Use AI to Optimize Old Processes — Eliminate Them

Traditional enterprises cling to processes refined over decades. The instinct is to apply AI for auto-filling forms, auto-pushing notifications, and auto-approving requests — essentially running the old process faster. Xu argues the real question is whether the process itself still needs to exist. AI should be an opportunity to redefine the business, not a life-support system for legacy inefficiency.

2. Product Innovation Is Not "Plan Then Act" But "Act Then Plan"

Conventional R&D follows a waterfall: extensive research, repeated reviews, finalized spec, then build. AI drives trial-and-error cost to near zero. The winning approach is to move first, validate quickly, iterate continuously, and converge direction through action. The article asks whether your R&D system still waits for a perfect plan or has built an AI-driven rapid validation loop.

3. Business Model Is Not Usage-Driven But Value-Creation-Driven

Past digital and SaaS models measured success by login duration, document count, active frequency — charging by usage volume. In the AI era, customers pay for tangible outcomes: cost reduction, risk control, profit growth. The challenge is to articulate the incremental value created, not just count API calls.

4. Not Building Walls for Moats — Open Symbiosis for Vitality

Enterprises historically hoard data, close capabilities, and build isolated systems. In the agent era, no single company can close the loop on all capabilities. True defensibility comes from continuous connection, capability exchange, and co-evolution. The strategic choice: reinforce data silos or design composable, connectable business capabilities.

5. Organize Around Capabilities, Not Power

Pyramid organizations arrange roles by hierarchy, authority, and approval chains; many positions exist solely to move approvals. When AI handles execution, verification, and statistics, the organizing principle must shift to who judges, who creates, who collaborates — not who holds higher rank. The question: does your structure serve approval authority or efficient capability flow?

6. Scarce Talent Is Not "AI Users" But "Cognitive Flywheel Drivers"

Many firms run company-wide prompt-engineering training. Calling AI is table stakes, not core competency. The rare talent builds a personal cognitive flywheel: hypothesize, delegate verification to AI, review results, iterate mental models, and maintain independent judgment amid information overload. Are you training prompt workers or developing self-renewing decision-makers?

7. The CEO Must Be "Chief Practice Officer," Not Just Supporter

In past transformations, the boss endorsed, funded, and kicked off, then handed off to CIO/IT. AI transformation penetrates business, KPIs, accountability, and interests — IT alone cannot drive it. The CEO must personally experiment, rewrite rules, and break ingrained habits. The litmus test: are you applauding from the sidelines or the first to practice and break the old ways?

Sequence Matters: Cognition → Management → Technology

The article stresses that the ordering cognitive transformation → management reform → technology revolution is itself the most critical insight. Reversing it — buying tech first, fixing mindset later — leads to flashy launches and quiet failures because AI transformation reshapes organizational form, value judgment, and human roles, not just tools.

Implementation Roadmap

First, the CEO completes the identity shift from supporter to Chief Practice Officer — this is the keystone.

Re-examine core processes with the courage to "eliminate processes," not merely optimize. Without this resolve, don't start AI.

Update talent evaluation: measure not AI tool proficiency but ability to continuously iterate cognition.

Only then select models, deploy compute, and build the technical foundation.

"Technology can be purchased, computing power can be leased, models can be called, but cognition must be crossed by oneself."

The author, drawing on years of ERP, MOM, and supply-chain implementations across factories and groups, observes that digital projects used to stall on mental inertia. Traditional digitalization could lock processes via systems and enforce compliance through policy. AI differs: agents execute, but humans still choose, judge, and direct. An organization's cognitive ceiling becomes its AI transformation ceiling. Many firms aren't short of technology; they're trapped by yesterday's successful patterns. The article concludes that AI is not a trend-chasing procurement but a quiet, difficult, self-driven cognitive revolution that determines the next decade.