Unitree Disputes the 100 Yuan Approval Claim, Exposing a Bigger Management Problem

Unitree Robotics disputed claims that CEO Wang Xingxing must approve expenses above 100 yuan and that its employee system emphasizes punishment over rewards. The company told Chinese media that many details were untrue and should not be taken seriously.

That denial matters, but it does not settle the controversy. Unitree did not publicly identify which allegations were inaccurate or provide an alternative description of its approval and reward systems.

The argument is therefore larger than one disputed expense threshold. It tests whether a fast-growing technology company can preserve founder-level discipline without turning every decision into a referendum on the founder.

What Unitree Actually Said About the 100 Yuan Claim

The verified story is narrow: allegations circulated on September 1, 2026, and Unitree issued a broad denial without a detailed rebuttal.

The controversy began with a Chinese media account of Unitree’s internal management. Employees quoted in that reporting described Wang as a highly involved founder who watched costs and product details closely.

Among the most widely repeated claims was that an expense above 100 yuan required Wang’s approval. Another alleged that Unitree’s reward and punishment system offered penalties with almost no corresponding rewards.

The reports also described Wang as direct and demanding. He reportedly discussed small engineering choices with frontline employees, including the length of individual screws used in products.

These details spread quickly across Chinese social media. Hashtags about the 100 yuan threshold and the alleged punishment-heavy culture reached trending lists on September 1.

Unitree then responded through comments reported by Chinese outlets. The company said, “Many of the claims are untrue. Do not take them seriously,” according to the company response published that evening.

That statement establishes two facts. Unitree rejected at least part of the circulating narrative, and it advised readers against accepting the reports at face value.

It does not establish that every allegation was false. The statement did not specify whether Unitree disputed the 100 yuan figure, the approval process, the reward system, or all three.

Public reporting has also not produced an internal expense policy, a written performance framework, or a named executive explanation. The employees behind the original descriptions were not publicly identified.

That leaves a significant verification gap. Readers should treat the claims as allegations, not as confirmed descriptions of Unitree’s current policies.

Some later online posts asserted that Unitree uses a tiered authorization system. Those assertions should receive the same scrutiny because no accessible primary document accompanied them.

The timing is clear even if the underlying policy is not. The first major reports appeared on September 1, and Unitree’s response followed that day.

A separate management account published on September 1 attributed the allegations to an earlier report by Caijing magazine. It repeated the expense claim alongside descriptions of Wang’s product involvement.

The controversy reached English-language readers shortly afterward. However, most English coverage ultimately traces back to the same small group of Chinese reports and the same brief denial.

That source concentration matters. Ten articles repeating one allegation do not create ten independent confirmations.

The strongest responsible conclusion is limited. Unitree faced allegations about centralized approvals and punishment-heavy management, then rejected many details without offering a point-by-point account.

That unresolved space is the real starting point. The important question is not whether one viral sentence sounds believable. It is what effective controls should look like when a founder-led hardware company becomes much larger.

Why Founder Control Stops Scaling

A founder can inspect critical decisions, but a growing company cannot route routine judgment through one person without creating delays and hidden workarounds.

Founder attention often helps an early hardware business. Product failures can originate in small component choices, supplier substitutions, or testing shortcuts.

A chief executive with deep engineering knowledge can spot problems that a general manager might miss. That ability can become part of the company’s advantage during its earliest stages.

Expense discipline also matters in robotics. Hardware businesses manage components, prototypes, contract manufacturing, inventory, logistics, repairs, and warranty exposure.

Small costs can multiply across production. A founder who teaches teams to question waste is not automatically micromanaging them.

The problem begins when personal attention becomes an operating system. An organization needs to distinguish founder expertise from founder permission.

If every ordinary reimbursement truly required CEO approval, the direct cost would extend beyond the executive’s time. Employees would also spend time preparing, checking, escalating, and waiting.

Managers would learn that they do not possess real authority. Employees would learn that a tiny administrative mistake can attract attention from the highest level.

Those lessons shape behavior. People begin optimizing for approval instead of outcomes.

Some employees would delay legitimate purchases. Others might combine expenses, divide them, or use informal arrangements to avoid the official route.

Engineers could become reluctant to run small experiments. Team leaders could stop making timely decisions because they expect the founder to reverse them.

This is why approval thresholds should reflect risk, not symbolism. A transaction’s amount is only one part of its risk.

A modest payment to an unfamiliar vendor can create fraud or security concerns. A much larger payment under an established contract can be routine.

Good controls therefore examine category, vendor, budget ownership, documentation, and exceptions. They do not treat every transaction as equally important.

The right design is delegated authority with traceability. Team leaders receive defined limits, finance teams monitor patterns, and unusual cases move upward.

The CEO should review aggregate spending, high-risk exceptions, and material commitments. That creates oversight without requiring personal approval for ordinary work.

Technology makes this separation easier. Expense systems can flag duplicate claims, unusual vendors, missing receipts, policy exceptions, and repeated threshold avoidance.

Dashboards can show where spending rises unexpectedly. Random audits can test whether managers are following policy.

These mechanisms provide more information than a long approval queue. They also preserve a record that another executive or auditor can examine.

A founder can still intervene when data reveals a problem. The difference is that intervention becomes selective and evidence-based.

This approach also reduces key-person risk. A company should continue operating when its chief executive travels, becomes ill, or concentrates on a product crisis.

If routine transactions stop when one person becomes unavailable, the approval design is fragile. That remains true even when the person rarely causes a delay.

Centralized approval can appear efficient because it creates one clear decision maker. Yet it often moves complexity away from the policy and into employees’ waiting time.

That waiting time rarely appears in an accounting report. It appears as slower purchasing, postponed testing, duplicated communication, and frustrated managers.

For Unitree, the allegation creates reputational pressure even without confirmation. The company now has an opportunity to describe which decisions belong to Wang and which belong to operating teams.

A transparent answer would not require disclosing sensitive financial controls. It could simply explain that routine expenses follow documented authority levels and independent finance review.

Without that explanation, the public debate will remain attached to the disputed 100 yuan number. The number is memorable because it symbolizes a broader fear about founder dependency.

Punishment and Reward Shape Different Behaviors

Punishment can stop a defined violation, but it is a weak primary tool for producing experimentation, collaboration, and difficult engineering work.

Companies need consequences. Fraud, harassment, data misuse, safety violations, and deliberate quality shortcuts cannot be managed with praise alone.

A fair system should state prohibited conduct, investigate consistently, and apply proportionate consequences. Employees also need a genuine appeal route.

However, penalties work best when the required behavior is clear. They are less useful when employees must explore uncertain technical paths.

Robotics development contains repeated failure. Engineers test mechanisms, motion systems, control software, sensors, batteries, materials, and manufacturing processes.

Many experiments will not work. If an organization routinely treats unsuccessful attempts as individual failures, employees learn to conceal uncertainty.

They choose safer projects. They report problems late. They avoid decisions that could produce visible mistakes.

A punishment-heavy environment can therefore improve superficial compliance while weakening technical learning. The company sees fewer reported errors but may not experience fewer errors.

Reward systems influence a different question. They tell employees which valuable behaviors deserve repetition.

Recognition might reinforce a well-documented experiment, a supplier improvement, a safety intervention, or assistance that helped another team complete a difficult release.

The reward does not need to be theatrical. It needs to be specific, credible, and connected to work that the company genuinely values.

An evidence review from the Chartered Institute of Personnel and Development found that incentives and recognition should reflect job realities. The research review also emphasizes consistent links to performance and perceived fairness.

That fairness has three practical dimensions.

First, distributive fairness asks whether outcomes are allocated reasonably. Employees compare their contribution and recognition with those of colleagues.

Second, procedural fairness asks whether the decision process is consistent and open to relevant evidence. Employees need to know how judgments are made.

Third, interpersonal fairness concerns how managers treat people while applying the process. A technically correct decision can still damage trust when delivered disrespectfully.

These dimensions explain why simply adding bonuses would not resolve allegations of punishment-heavy management. A large reward can still feel arbitrary.

Likewise, a penalty can be accepted when the rule was clear, the investigation was credible, and comparable cases received comparable treatment.

Companies should separate misconduct from performance. Misconduct concerns behavior that violates a known rule or obligation.

Performance concerns whether an employee produced expected results under actual working conditions. It often requires coaching, resources, role clarification, or revised goals.

Combining both categories into a single punishment ladder creates confusion. Employees cannot tell whether they broke a rule or missed a target.

Managers then gain excessive discretion. One employee receives coaching while another receives a penalty for a similar outcome.

A better framework begins with explicit categories.

Safety, ethics, security, and financial integrity require nonnegotiable controls. Violations need documented investigations and proportionate consequences.

Ordinary performance should use clear expectations, regular feedback, and improvement periods. Managers should record the support provided before escalating consequences.

Innovation work needs a separate treatment. The company should distinguish a responsible failed experiment from negligence or repeated disregard for evidence.

A responsible failure follows an approved hypothesis, budget, safety boundary, and documentation standard. It produces knowledge that the next team can use.

Negligence ignores established controls or conceals relevant information. Treating these outcomes identically discourages experimentation without improving accountability.

Recognition should mirror the same distinctions. Companies can reward results, but they should also recognize behaviors that make future results more likely.

Those behaviors include sharing negative findings, improving test coverage, documenting supplier defects, mentoring colleagues, and raising safety concerns early.

Gallup’s workplace framework includes clear expectations and recent recognition among its core engagement conditions. Its workplace analysis connects those conditions with measurable organizational outcomes across business units.

That does not mean every employee needs constant praise. Generic recognition can feel manipulative and lose its meaning.

Recognition works when it identifies the action, explains its value, and comes from a credible observer. It should not become an attendance prize.

The design target is balance, not equal numbers of rewards and penalties. One serious violation might justify a major consequence, while many ordinary contributions receive routine recognition.

The relevant test is whether employees can predict how the system will respond. Predictability creates accountability without requiring fear.

For Unitree, the company’s broad denial leaves this central issue unanswered. It has not publicly described whether employees receive formal recognition, bonuses, promotions, equity, or team rewards.

Absence of public information is not evidence of absence. Yet the controversy shows how quickly an information vacuum can define an employer’s reputation.

Fair Rules Matter More Than Generous Rewards

Employees judge a management system by its consistency, evidence, and appeal process, not only by the size of its rewards.

A company can spend heavily on bonuses and still create distrust. That happens when employees do not understand why one project counts and another does not.

It also happens when managers change targets after work has begun. Employees experience the reward as discretionary, even if executives consider it generous.

Vague criteria increase this risk. Terms such as “ownership,” “impact,” and “attitude” can describe useful qualities, but they require observable examples.

Otherwise, managers can use them to rationalize almost any decision. Personal familiarity and communication style begin influencing judgments that should concern performance.

Research on fair performance reviews warns that open-ended evaluation criteria allow bias to enter the process. Structured evidence and consistent standards reduce that exposure.

A workable system should tell employees what outcomes matter before the review period. It should also explain which constraints managers will consider.

Hardware teams cannot control every supplier delay, certification decision, or component shortage. Sales teams cannot control every customer budget change.

Goals should therefore combine individual outcomes, team outcomes, and controllable behaviors. No single metric should determine the entire evaluation.

Individual measures preserve accountability. Team measures reflect the collaborative nature of complex products.

Behavioral measures protect work that produces long-term value but does not appear immediately in revenue or delivery statistics.

That mix also reduces gaming. People have less reason to hoard information when collaboration affects the result.

Managers should document evidence continuously instead of reconstructing a year from memory. Short review cycles make corrections possible before a final rating.

Employees should see the evidence and respond. A disagreement should move through a defined review route that does not depend solely on the original manager.

Calibration meetings can compare standards across teams. However, they should not become secret negotiations that change ratings without explanation.

The final decision should include a short written rationale. It should connect observable work with previously stated criteria.

The same requirement should apply to penalties. A consequence should identify the rule, available evidence, decision authority, and appeal process.

Companies should also publish a decision-rights map. This document states who can approve spending, hiring, vendor changes, product exceptions, and performance actions.

The map should define escalation triggers. It should also name a backup authority when the primary approver is unavailable.

For routine expenses, employees need a short policy that answers common questions. It should cover eligible categories, required records, approval limits, and processing times.

Any automated flags should be explainable. Employees should know why a claim was held and what information resolves the issue.

Finance teams can then audit patterns instead of manually treating every request as suspicious. Managers remain accountable for their budgets.

The organization should monitor whether the policy creates unnecessary friction. Useful measures include processing time, rejection rate, exception frequency, and repeated resubmission.

Employee surveys can test perceived fairness. Exit interviews and internal transfers can reveal whether a specific manager applies rules differently.

Recognition requires similar monitoring. Leaders should examine whether awards cluster around visible teams, headquarters staff, or employees with frequent executive access.

The goal is not mathematical uniformity. Different roles produce different kinds of contributions.

The goal is to identify unexplained patterns before they become cultural assumptions. Transparent criteria make those patterns easier to challenge.

Companies can support transparency with a shared policy repository. A searchable knowledge base helps teams find current rules, decision records, and technical documentation without relying on hallway explanations.

Documentation does not replace judgment. It gives employees a common reference when judgment differs.

This matters especially in founder-led companies. Employees often interpret a founder’s preferences as rules, even when the founder intended only a suggestion.

Written decision rights prevent that ambiguity from spreading. They clarify where founder review is mandatory and where managers should act independently.

The best reward and punishment system is therefore not the most elaborate one. It is the one employees can understand before a conflict occurs.

Unitree’s Denial Leaves a Governance Question

Unitree does not need to prove every online story false, but a broad denial cannot resolve a controversy about opaque internal rules.

The company faces a difficult communications choice. Publishing detailed internal policies could expose information that belongs inside the business.

Responding line by line could also extend the life of weak allegations. Companies should not let every trending post dictate their disclosure schedule.

However, silence has costs when a claim becomes a symbol of organizational culture. Prospective employees, suppliers, investors, and customers may use that symbol when evaluating the company.

A concise factual explanation would be more useful than a general dismissal. Unitree could state whether routine expenses use tiered approval and whether Wang reviews only exceptions or aggregate reports.

It could also explain whether employee performance decisions include positive recognition, documented criteria, and an appeal path.

Such a statement would not confirm confidential compensation terms. It would address the system rather than individual cases.

The company should avoid declaring the matter resolved merely because some details are inaccurate. A mixed report can contain errors while still identifying a real management concern.

Critics face the opposite responsibility. They should not convert anonymous descriptions into established policy without documents or independent confirmation.

The specific 100 yuan claim remains unverified. So does the stronger assertion that Unitree’s system contains almost no rewards.

There are plausible reasons to doubt a literal interpretation. A large operating company would generate many routine transactions, making universal CEO review inefficient.

Yet plausibility is not evidence. Some organizations maintain surprisingly low approval thresholds or use systems that route approvals quickly.

The word “approval” can also describe different actions. Wang might approve a budget category, an exception, a batch, or every individual claim.

Those possibilities have very different management implications. Public reporting has not distinguished among them.

Similarly, “almost no rewards” might refer to formal cash awards, everyday recognition, promotions, or one employee’s experience.

The phrase is too broad for confident judgment. It still deserves attention because it describes how at least some reported sources perceived the system.

Perception affects behavior even when leaders believe the policy is balanced. If employees cannot identify rewards, the system may not communicate them effectively.

That is why corporate culture cannot be evaluated solely through policy documents. Leaders must examine how rules operate across teams.

A company might possess a written recognition program while managers rarely use it. Another might lack a formal program but provide frequent advancement and meaningful acknowledgment.

The correct audit compares stated policy, actual decisions, and employee experience. Gaps among those three layers create reputational and operational risk.

Unitree’s technical reputation raises the stakes. The company operates in a field where safety, manufacturing quality, cost, and rapid iteration all matter.

Strict controls can protect those priorities. Excessive centralization can undermine them by slowing decisions and discouraging early problem reporting.

The core tension is therefore discipline versus scalable trust. Unitree needs both.

Discipline without delegated trust makes the founder a bottleneck. Trust without controls can produce inconsistent spending, weak quality, and avoidable risk.

The controversy offers no basis for declaring which extreme describes Unitree. It does show that outsiders lack enough verified information to evaluate the balance.

That uncertainty should shape coverage. The story is not proof of a dysfunctional company, and the denial is not proof of a healthy management system.

Both interpretations go beyond the available evidence.

Three Signals Will Show What Happens Next

The next meaningful evidence will come from policy clarification, employee patterns, and Unitree’s ability to operate without visible founder bottlenecks.

The first signal is a more specific company statement. Unitree can strengthen its denial by identifying the central factual errors.

A useful clarification would explain the approval structure without revealing individual expenses. It would distinguish routine decisions from exceptions requiring executive review.

The same statement could summarize how performance, recognition, and discipline are governed. Clear process information would weaken the claim that arbitrary founder control defines daily work.

A continued broad denial would leave the central uncertainty intact. It would not validate the allegations, but it would allow the 100 yuan story to remain the dominant public shorthand.

The second signal is consistent employee evidence over time. One anonymous account can identify a lead, but repeated independent accounts can reveal a pattern.

Future reporting should seek employees from different functions and periods. Engineering, manufacturing, finance, sales, and human resources experience management systems differently.

Specific examples matter more than general labels. Reporters should ask who approved a decision, how long it took, which written rule applied, and whether an appeal existed.

They should also seek evidence of rewards, promotions, recognition, and employee ownership. A punishment narrative cannot be assessed without examining its other side.

Employee turnover can offer context, but raw departures are difficult to interpret. Fast-growing technology companies often experience restructuring and competitive hiring.

More useful signals include repeated descriptions of the same process, documented policy changes, or unusual concentration of decisions at the executive level.

The third signal is operating scalability. Unitree’s management model should support faster production, reliable quality, and timely product development as the company grows.

Founder involvement is not automatically a weakness. Many technically accomplished founders remain close to product decisions.

The test is whether that involvement targets high-value questions. A founder should spend attention where specialized judgment changes the outcome.

Routine matters should move through accountable managers and systems. If ordinary work repeatedly waits for executive attention, the company has not converted founder knowledge into organizational capability.

Customers and partners can watch practical indicators. Product delays, inconsistent support, repeated quality issues, or frequent reversals can reveal coordination problems.

Stable execution would support Unitree’s position that viral descriptions misrepresent its internal reality. It would not prove that every employee experience is positive.

Companies outside robotics should watch the controversy for a different reason. The same design problem appears in startups, family businesses, research laboratories, and rapidly expanding software teams.

Early habits often survive long after their original purpose disappears. The founder keeps approving small costs because that process worked when the company had ten people.

Managers continue requesting approval because taking independent responsibility feels riskier. Employees then treat escalation as the safe default.

Breaking that cycle requires more than raising an expense threshold. Leaders must define authority, measure decisions, and protect managers who exercise sound judgment.

Rewards and consequences should reinforce that model. Managers should receive recognition for responsible delegation, not only for avoiding visible mistakes.

Employees should face consequences for concealment or negligence, not for every well-designed experiment that fails.

The disputed Unitree story is valuable precisely because the underlying facts remain incomplete. It forces a distinction between a viral claim and the management principle hidden inside it.

No available evidence confirms that Wang personally approves every expense above 100 yuan. No detailed public evidence disproves the claim either.

Unitree’s brief response establishes its disagreement, not the exact operating truth. Readers should resist filling that gap with certainty.

The better question is measurable: can Unitree show that cost control, employee accountability, and technical rigor operate through fair systems rather than personal intervention?

Over the next several months, specific clarification, independent employee evidence, and scalable execution will provide the answer. Until then, the 100 yuan claim should remain labeled as disputed, while the governance issue stays open.