- Companies are creating digital replicas of employees using AI trained on personal work patterns.
- AI systems now seek to learn decision-making processes, not just final output.
- Legal and ethical questions arise over ownership, consent, and transparency of work data.
In a game company based in Shandong province, China, a former HR specialist was replaced by an AI worker who was trained on his documents and typical work patterns. According to the South China Morning Post, this system now handles questions, schedules meetings, and even creates slide decks and spreadsheets under the digital persona of that departed employee. Honestly, this has caused quite a stir among people observing it because, well, it makes the idea of replacement seem less like simple automation and more like a sort of digital afterlife. The main worry isn't just that jobs are becoming mechanized; rather, folks are concerned that someone’s work style, particularities, decisions, habits, can actually be preserved and replayed after they’re gone.
That kind of story came around just as Meta was catching heat over a different, but related, effort, according to Reuters from April. The company had plans to track U.S. employees’ mouse movements, clicks, keystrokes, and screen activity, tools on their computers, to help AI models learn how people really get their work done. Meta claimed the data wouldn’t be used for performance reviews or anything like that, but still, it stirred up privacy worries because the focus was on how work gets done, not just the finished output. The Verge and Business Insider reported similar details, describing this initiative as part of Meta’s broad effort to develop agents that could eventually perform office tasks independently.
What’s really interesting here is the bigger pattern emerging. It’s not just about the final product anymore, although those are easy enough to find. What companies really want now is the messy middle, the decision-making process, the corrections, the shortcuts that take someone from problem to solution. That’s why the data that AI builders are now craving is shifting from public text to real human workflows, including the subtle judgment calls buried inside everyday computer use. Business Insider called this the next big data grab: learning how humans work, not just what they write.
On the flip side, a mirror image of this trend is also taking shape. According to the Financial Times, firms like Mercor are hiring consultants, journalists, and real estate pros to teach their models how those jobs work, and Handshake AI pays professionals for their existing work products, things they have the rights to share. Which basically means expertise is becoming a kind of commodity. On one hand, companies are extracting workflow data from inside their systems; on the other, workers are selling their past work. And in both cases, the result is pretty much the same. Human judgment and decision-making are turning into training data for systems designed to operate without actual human oversight.
Developers might already be unintentionally creating some of the clearest datasets out there. For example, tools like Claude Code and Codex save complete session histories locally as JSONL files, keeping prompts, tool calls, retries, rejections, and diffs. These logs aren’t just about code; they actually document how a problem was broken down, what standards were applied, and why earlier attempts didn’t work. So, in effect, what’s stored isn’t just the final answer but also the reasoning process behind it, making these logs incredibly valuable for training or mimicking human thought.
This naturally brings up a practical question for both employers and suppliers: who owns the knowledge contained within those work traces, and what’s it worth? The article points out that some Chinese legal commentators warn that chat logs, emails, and work habits could be classified as personal info, and that state media argue professionals’ skills shouldn’t just be fed into AI models without some kind of fair compensation. For companies building software, managing contractors, or outsourcing tech jobs, the line is blurring. Records that once seemed like simple handoff material are now starting to be regarded as training fuel. The more work becomes readable by machines, the more organizations will need to carefully think about consent, ownership, and control.
There is also a broader business angle that reaches beyond white-collar offices and into sourcing, logistics, electronics, lifestyle, and mobile ecosystems. In many industries, the real value is no longer only the artifact at the end of a workflow but the chain of decisions that produced it: which supplier was chosen, why one route was preferred over another, how exceptions were handled, and what quality checks were repeated before shipment or release. When an AI system can observe enough of those steps, it can begin to imitate not just a task but an operational style. That makes the underlying data more sensitive, because it is no longer simply about productivity metrics; it is about the hidden structure of human expertise.
For employers, this creates a tricky balancing act. On one side, AI promises convenience, continuity, and speed. A digital persona that answers questions and creates decks may look efficient, especially when a team is short-staffed or when institutional memory is at risk of walking out the door. On the other side, the same setup can feel unsettling because it blurs the line between helpful automation and a persistent surrogate for a person. If an employee leaves voluntarily, but their habits, phrasing, and judgment remain accessible through an agent, then the company has not merely stored documents; it has preserved a working pattern. That raises questions about whether consent should cover only the collection of data, or also the future use of a person’s style as a machine-readable model.
Employees, meanwhile, may start to view everyday software interactions differently. A meeting note, a chat thread, a spreadsheet edit, or even a rejected draft can become part of a long-term data trail. What once felt disposable may now look like an asset that trains systems for future labor replacement. That shift could influence workplace behavior in subtle ways. People may become more cautious, more guarded, or more strategic about what they type and how they act on screen. In that sense, AI is not only changing workflows; it is also changing workplace culture. The rise of agentic tools in mobile and desktop environments may eventually make this even more visible, because the more closely tools observe users, the more they can learn to mimic them.
At a policy level, the most important issue may be transparency. Workers should know when their activity is being recorded, how long it is kept, whether it is used for model training, and whether it could be shared across teams, vendors, or regions. Companies should also be clear about whether compensation is tied to that use, especially when specialized knowledge is being collected from people in consulting, electronics, software, logistics, or media roles. If AI development is moving toward the capture of lived work patterns, then existing rules around data protection, employment, and intellectual property may need to catch up quickly. Otherwise, organizations may find themselves building powerful systems on top of unclear permissions.
Takeaways
- - AI is moving beyond outputs and into the process of work itself.
- - Consent matters, but so does clarity about future reuse of work traces.
- - Logging tools can create valuable training data from everyday office activity.
- - Workers may need stronger visibility into what their systems record.
- - The next big debate is not just automation, but ownership of human judgment.
Q&A Q: Is this just normal automation? A: Not exactly. The key difference is that these systems aim to learn how people work, not only what they produce.
Q: Why are people worried about privacy? A: Because mouse movements, keystrokes, chats, and screen activity can reveal habits, decisions, and work style, not just final results.
Q: Could this affect more than office jobs? A: Yes. Any field with traceable digital workflows, like sourcing, logistics, electronics, and mobile product teams, could generate useful training data.
Q: What should workers ask employers? A: How data is collected, how long it is stored, who can access it, and whether it may be used to train AI systems in the future.
Q: What is the main takeaway for companies? A: AI value is increasingly tied to human process data, so governance, consent, and compensation should be treated seriously.
Disclaimer: This article may have been created with AI assistance and reviewed by our editorial team. It is provided for general informational purposes only. Readers should verify information independently before relying on this content.
Sources
- - Paragraph 1: SCMP
- - Paragraph 2: Reuters, The Verge, Business Insider
- - Paragraph 3: Business Insider
- - Paragraph 4: FT, Business Insider
- - Paragraph 5: dev.to
- - Paragraph 6: dev.to, SCMP
