The Organizational Muscle Huawei and Midea Built Over Thirty Years—First to Depreciate When AI Arrives?

Deep thinking on AI and aspirations —— ByteDance Deep Thinking Circle

Midea air conditioners, Huawei phones, TCL televisions—during their fastest growth years, these companies shared one thing in common: sales compound growth rates above 30% for consecutive years, sometimes doubling in a single year. Management scholar Shi Wei tracked these companies long-term and noticed a counterintuitive phenomenon: after completing management standardization, their growth accelerated rather than slowed, even surpassing the wild early years of seizing opportunities. The market hadn’t improved—the organization had changed. Guerrillas had become a regular army.

Shi Wei coined a term for this organizational form: high-energy organizations, characterized by high performance, high energy, and high capability. In plain terms: earn more, fight harder, have stronger people. Huawei, Midea, Inovance Technology, and Kuka Home were all used as examples. This approach has a strong military flavor and deserves careful examination, because it represents the most thoroughly validated organizational formula for Chinese enterprises over the past thirty years. And now AI is pushing it into a corner.

What Problem Does an Army Actually Solve

The hierarchical structure of modern companies originates from Roman legions: 8-person squads, 80-person centuries, 480-person cohorts, 5,000-person legions, with commanders at each level responsible for aggregating information from below and transmitting decisions from above. This structure was designed around a human limitation: the number of people a leader can effectively manage, roughly 3 to 8. In the 1850s, railroad companies drew the world’s first organizational chart to manage systems spanning over 500 miles with thousands of workers; Taylor then broke work into specialized tasks, and the functional pyramid took shape. For the next two hundred years, management science basically tinkered within this framework.

Understanding this origin reveals what all the mechanisms of high-energy organizations are actually doing. Shi Wei’s playbook roughly consists of these moves: recruit top student soldiers from the best universities, using scholarships, internships, and pre-positioned courses to improve hit rates; offer compensation 30% to 50% above industry average, because excellent talent has a salary lever—10% above average pay can yield 20% more value creation; high targets plus asymmetric incentives, with ordinary supervisors earning 120,000 yuan annually, managers 500,000, directors 1 million, creating massive vertical gaps; multiple candidates for each position, with backup cadre programs making groups compete for succession, preventing incumbents from coasting; boldly use young people, keeping the average age of technical and management personnel 3 to 5 years younger than the industry, learning war through war.

These diverse mechanisms actually produce the same thing: certainty. In an era of limited skills and information, an organization’s combat effectiveness depends on how predictably it can compress the execution of ordinary people. Top recruitment buys malleability, high pay buys willingness to struggle, high pressure and backup programs buy stable supply of elimination and replacement. The essence of military organizations is a certainty production line, with discipline as its most important product.

Shi Wei has another judgment readers often overlook: the system matters more than people. When the soil is well-designed, towering trees will grow on their own; processes, mechanisms, and systems shape people in reverse. This judgment wasn’t remarkable at the time, but viewed today, it’s precisely the most durable part of the entire paradigm.

AI Begins Mass-Producing Execution, Discipline Depreciates

Now the execution side is being disrupted. YC partner Diana Hu observed in a batch of startups that teams cut engineering sprint times in half while delivering nearly ten times the volume; in more aggressive teams, code repositories can have no hand-written code—people only write specifications and tests, leaving implementation to agents to iterate. Starting January 2025, Jack Dorsey spent three hours every morning challenging whether tools could do things he thought impossible, feeling surprised every day for a year straight, after which Block laid off 40% of its people, going from idea to execution in under three weeks.

Putting these facts back into the military organization framework yields a rather harsh conclusion: when certainty execution can be purchased by the token, companies hand-producing certainty through organizational discipline are essentially manufacturing in-house what’s already mass-produced elsewhere. Disciplined execution is precisely what military organizations take most pride in producing, and it’s also the type of output where AI’s marginal cost is declining fastest.

My judgment is that scarcity is moving to the other end. Execution becomes cheap, while judgment, taste, and accountability for customer outcomes become expensive. This also explains why the most prominent AI companies this round are heading in almost the opposite organizational direction from high-energy organizations. Anthropic specifically includes culture interviews in hiring, asking 15 to 20 scenario questions in an hour, screening not for who’s smartest but who has small ego and willingly digs into details, even willing to pass on engineers with shinier backgrounds; Block replaces managers whose main job is information routing with player-coaches who still write code and still lead people, leaving coordination to systems.

High-energy organizations trade obedience for output; these companies trade voluntary commitment and identification for output. Two human-powered models are diverging, and the fork in the road is AI taking over execution.

Don’t Rush to Discard Everything, First Distinguish What’s Expired

Conversely, dismissing military organizations entirely isn’t honest either. Business nature determines organizational logic—Shi Wei himself explains this clearly: ByteDance makes short-cycle products, running dozens of versions per week, fixing mistakes quickly, so it dares to be bold and pioneering; Huawei makes base stations and chips where a single mistake’s recall cost could swallow an entire year’s profit, so discipline must trump everything. Applied to businesses with long chains and high error costs, militarized execution remains a moat. JD.Com manages 180,000 logistics workers relying on that precise reward-and-punishment system—it has no second choice. The first to fail are pure information businesses: companies whose output is documents, code, analysis, and content—the assets protected by this organizational discipline are themselves depreciating.

The judgment that system trumps people becomes even more valuable in the AI era, only the soil formula has changed. Previously it was processes, controls, backup programs; now it’s context, evaluation sets, queryable organizational memory. Diana Hu’s formulation is to make the entire company AI-readable, with every important action producing a learnable artifact. Those who design the soil still win—this hasn’t changed.

The salary lever still holds, but what it buys has changed: high pay used to buy execution intensity; going forward it buys judgment quality. Shi Wei’s observation from a few years ago looks even more accurate now: the worst ROI human capital model is hiring people with impressive credentials from big companies who mainly come to monetize their resumes. After execution can be outsourced to agents, the non-judgment parts of impressive credentials will only continue declining in value.

Boundaries need clarification. If your business relies on hundreds of thousands of hands and feet, military organization logic still holds—don’t be swayed by any new concept. If your output is information, then every mechanism in your organization used to produce obedience is yielding ground to competitors. Companies in the middle state are most numerous and most difficult—the correct approach is to judge by business line, not argue for a unified answer in conference rooms. This muscle built over thirty years—keep what should stay, replace what should change, the only thing not to do is copy or discard the entire set wholesale.