Control Algorithms: Your Workplace AI Might Be Working Against You

Every company deploying AI is making a choice they are not necessarily announcing. Most aren't even making the decisions consciously.

The most heated debate frames it as "jobs" – will AI replace them? The more urgent one is quieter and harder to perceive: Who is AI actually working for?

A digital assistant augments you. A control algorithm defines your work flow. The difference doesn't lie in the technology, but rather in who sets the parameters and whose interests the parameters serve.

On June 11, Alibaba fired Chen Hang, founder of DingTalk, China's dominant workplace app – a combination of Zoom, Microsoft Teams and Slack with over 700 million users.

The high-profile dismissal, at least outwardly, was triggered by a 75,000-word resignation letter from a product manager who accused Chen of creating a high-pressure work environment, top-down command and disregard for individual creativity as he tried to rebuild his team around AI agents. The resignation went viral across Chinese social media.

The obvious story is about management culture. Alibaba's public statement said Chen's "high-pressure approach" to his job is not in line with company culture. But the real story is about who AI is being built to serve, and it's a story playing out beyond China.

The manager who resigned wrote a subsequent online article after Chen was fired. In it, he explained that the reason for resignation was to focus on "the common and systematic problems many big organizations face when deploying AI.

"It's not just a problem for DingTalk, or Ali, or China," he wrote.

Indeed, what happened at DingTalk is symptomatic of decisions being made everywhere.

Alibaba's replacement for Chen Hang is Chen Yusen, a 34-year-old tech geek described as an "AI-native" thinker. He is now the youngest business unit chief in Alibaba's history.

The change in management is not just about Chen Hang being a control freak, but more the case of his direction for AI deployment. Chen was running a human control algorithm. Many a company wants a control algorithm, but it doesn't want a control freak designing it because that bakes excessive control into the product itself.

When your workplace platform decides which messages surface at the top of your feed, it is making a judgment about what demands on your attention are important. When it scores your response times, tracks your availability and sends that data to your manager, it is quantifying what it means to be a "good employee." When it routes tasks, flags anomalies and nudges behavior through interface design, it is encoding a boss' theory about "productive" work into the environment in which you operate.

None of this necessarily indicates malicious intent, but it does optimize a system to the boss' objectives – not yours. What many people want is an AI assistant that helps save them time and energy so they can focus their attention on more creative tasks. What many companies are deploying is an AI that is becoming the organization itself.

The question isn't whether your workplace AI is helpful. The question is who it helps, and to what ends as defined by whom.

Chen Hang's failure, under that understanding, wasn't that he was too controlling. It was that his control was the wrong kind – human, visible, resistible. You could push back against a demanding boss. You could write a resignation essay and go viral. What replaces him is something harder to pinpoint, harder to see, and considerably harder to fight.

DingTalk is a Chinese story. The control algorithm is not.

Think of workplace AI deployment as a layered stack, and similar structure appears everywhere.

At the top: governments and regulators, setting the outer boundaries of permissible AI deployment. China is openly discussing AI as an instrument of industrial policy and social governance instead of just a product feature. The US, not quite as explicitly, is simply less coherent while its regulatory frameworks lag years behind deployment reality.

In the middle of the stack is action central – corporations. They make the operative decisions about what gets optimized, what kind of behavior the system rewards or penalizes, and what gets measured and reported.

Microsoft has embedded Copilot across all its products. Salesforce AI agents live inside customer relationship management. German software giant SAP is rebuilding its personnel operations and finance software around AI workflows. Every one of these deployments encodes a theory of productive work. They are designed primarily to serve the firms, not the individual workers inside them.

And at the base of the stack are the AI systems trained on pre-set objectives that may not be delivered clearly through human managers. The question of assistant versus control algorithm isn't answered at any single layer, but emerges from how the layers interact and from whether workers, citizens and users have any meaningful input into that interaction. And so far, they don't.

China isn't running a different experiment in this realm. It's just running the same experiment faster. The speed of AI deployment in China compresses the timeline on decisions every major economy will face.

What takes a decade to surface in other markets plays out in two years here. DingTalk's 700 million users make it one of the largest real-world deployments of AI-controlled work on Earth. In the US, big tech tends to obscure architectural choices inside product roadmaps.

The DingTalk firing triggered a debate inside Alibaba about what DingTalk's AI should do and for whom. That has surfaced explicitly in internal memos public rebukes and personnel changes. The same gravitational pull exists everywhere. In artificial intelligence it is more measurable, optimizable and defensible to shareholders and boards. The assistant is harder to monetize than the control algorithm. The cage is easier to build than the tool that genuinely expands human capacity – simple corporate tendencies playing out first and most visibly in the world's fastest-moving AI deployment environment.

If AI is going to function as a genuine assistant rather than a sophisticated control algorithm, a few truths emerge but not from noble intentions or nicely worded press releases.

Consider these elements.

Transparency of objectives

Workers need to know what an AI system is being optimized for. "Productivity" may be the easiest answer but it doesn't really address core issues. Whose productivity? Measured how? Over what time horizon? At whose expense?

Meaningful options

An assistant whose help you cannot decline is not a true assistant. When AI workflow systems become the only viable path to doing your job and when opting out means falling behind on metrics you can't see being tracked, the relationship breaks down, regardless of how the product is marketed.

Accountability at the corporate level

The most consequential decisions about how AI shapes work are made by employers, not governments. Regulatory frameworks matter but consistently lag deployment reality. What matters more in the here and now is whether companies building these systems can answer honestly to workers, to users and to the public whom the system is designed to work for. At present, none of these conditions is currently standard. Most of the questions aren't even being asked.

AI native, which means artificial intelligence built into every layer of a platform, can be genuinely liberating. Agents absorb the friction. Humans operate with more autonomy, more creative bandwidth and less of their cognitive capacity consumed by administrative overhead.

The technology is genuinely capable of it. But capability is not destiny. The assistant and the control algorithm are built from the same materials. What differentiates them is not the sophistication of the model or the scale of the investment. It is about who sets the parameters, who can see them, and who has any power to change them.

So we have Chen Hang fired as a control freak in a very public way for his management style. I think he was essentially fired for not encoding that control deeply enough into the product. The humans who pushed back against him, who wrote the viral rebukes, won the visible battle.

The question is what gets built in his place by someone with a cleaner public image and a harder architectural vision. That question isn't being asked loudly enough. Not in Shanghai. Not in Seattle. Not anywhere the control algorithm is quietly going to work.