China’s new AI employment guidelines raise a question for every economy adopting the technology: can workers shape an investment before its consequences are fixed?
On 23 October, China announced guidelines intended to make AI development more employment-friendly: https://english.www.gov.cn/policies/latestreleases/202610/23/content_WS6adb4e53c6d00ca5f9a0deb7.html
Their significance lies partly in timing. The full text calls for employment-risk assessments during review of potentially disruptive AI projects in market-access areas: https://www.mee.gov.cn/zcwj/gwywj/202610/t20261024_1184583.shtml
It also encourages advance notice and transition support for affected workers. These are policy directions, not evidence of jobs saved or a guarantee that workers can veto a deployment. But they put a consequential question on the table: who gets to influence the decision before the software contract is signed?
An employment assessment is useful only if it can change the options being considered. Otherwise, it risks becoming a well-written explanation of a decision already taken.
Consider a hypothetical Indian business-services firm buying an AI system to handle routine customer requests. One plan uses it to reduce staffing. Another keeps more staff and pursues customers previously too expensive to serve. A third reduces queues and gives employees time to resolve difficult cases. All three use the same technology. They make different bets about demand, service and whose time matters.
None is automatically viable. More customers may fail to appear; the software may disappoint; keeping every existing task may make the firm uncompetitive. Yet comparing these alternatives before committing money is more informative than announcing a training programme after selecting the staffing cuts.
Workers have knowledge that a purchasing team may lack: which apparently repetitive calls conceal distress, which exceptions prevent costly mistakes, and where customers need a person who speaks their language. Asking for that knowledge can improve the investment itself. Paying people for the time needed to test alternatives makes participation practical, especially for staff who cannot afford unpaid meetings.
There is some evidence for taking participation seriously. An OECD study published in 2023, covering manufacturing and finance in seven countries, found that training and worker consultation were associated with better worker outcomes: https://www.oecd.org/en/publications/the-impact-of-ai-on-the-workplace-main-findings-from-the-oecd-ai-surveys-of-employers-and-workers_ea0a0fe1-en.html
It does not establish that consultation caused those outcomes, and it cannot stand in for evidence from an Indian workplace. It does challenge the assumption that employees’ involvement is merely an obstacle to efficient adoption.
The economic issue is broader than compensation. If managers measure an investment mainly by payroll savings, they may overlook ways of using AI to increase quality or reach new customers. If they count promised new business without testing demand, they can make the opposite mistake. A useful assessment puts both claims under scrutiny, including licensing, integration, supervision and training costs.
India has particular reason to care. Its Economic Survey 2025–26 identifies the tension between higher productivity and absorbing labour, while warning that delaying adoption can also trap firms in low productivity: https://www.indiabudget.gov.in/economicsurvey/doc/eschapter/echap14.pdf
That is an argument for examining feasible choices in specific industries. It offers no easy national formula.
The assessment’s boundary matters, too. Imagine the firm retains its permanent employees but sharply reduces work sent to an outside agency. A reassuring internal headcount could conceal losses elsewhere. Contractors should therefore be heard alongside employees, and participation should be possible through local-language discussions, not just an online questionnaire.
The strongest objection is that this could become another expensive approval ritual. Large incumbents can employ compliance teams; smaller rivals cannot. A poorly designed requirement might protect established businesses while slowing useful tools that make work safer or less exhausting.
That objection should shape the design. Reviews should be proportionate, time-limited and focused on consequential changes. Small firms could share independent technical advice through industry or public institutions. Employers should document alternatives and explain their choices; they should not have to pretend that every task can survive.
The ILO’s 2025 assessment stresses that exposure to generative AI is not a count of actual job losses and that many jobs are more likely to change than disappear: https://www.ilo.org/publications/generative-ai-and-jobs-2025-update
There is room for deliberate design, though no assurance of a painless transition.
A better future would use paid pilots and worker knowledge to improve both services and jobs. An uneven one would give permanent staff a voice while contractors absorb the adjustment. A harmful one would make employment assessments ceremonial—or use them to shield incumbents from competition. Which emerges depends partly on whether anyone can still alter the plan when the consultation begins.
Questions for founders and investors
Before approving an AI investment, ask which alternatives were tested, whose earnings and hours were counted, and what changed because workers spoke. Fund the pilot and publish the reasons for the decision. The most revealing evidence of participation is a decision that was still open to revision.
Sources
China announcement: https://english.www.gov.cn/policies/latestreleases/202610/23/content_WS6adb4e53c6d00ca5f9a0deb7.html
Full policy text: https://www.mee.gov.cn/zcwj/gwywj/202610/t20261024_1184583.shtml
OECD workplace surveys: https://www.oecd.org/en/publications/the-impact-of-ai-on-the-workplace-main-findings-from-the-oecd-ai-surveys-of-employers-and-workers_ea0a0fe1-en.html
India Economic Survey, chapter 14: https://www.indiabudget.gov.in/economicsurvey/doc/eschapter/echap14.pdf
ILO research brief: https://www.ilo.org/publications/generative-ai-and-jobs-2025-update
On 23 October, China announced guidelines intended to make AI development more employment-friendly: https://english.www.gov.cn/policies/latestreleases/202610/23/content_WS6adb4e53c6d00ca5f9a0deb7.html
Their significance lies partly in timing. The full text calls for employment-risk assessments during review of potentially disruptive AI projects in market-access areas: https://www.mee.gov.cn/zcwj/gwywj/202610/t20261024_1184583.shtml
It also encourages advance notice and transition support for affected workers. These are policy directions, not evidence of jobs saved or a guarantee that workers can veto a deployment. But they put a consequential question on the table: who gets to influence the decision before the software contract is signed?
An employment assessment is useful only if it can change the options being considered. Otherwise, it risks becoming a well-written explanation of a decision already taken.
Consider a hypothetical Indian business-services firm buying an AI system to handle routine customer requests. One plan uses it to reduce staffing. Another keeps more staff and pursues customers previously too expensive to serve. A third reduces queues and gives employees time to resolve difficult cases. All three use the same technology. They make different bets about demand, service and whose time matters.
None is automatically viable. More customers may fail to appear; the software may disappoint; keeping every existing task may make the firm uncompetitive. Yet comparing these alternatives before committing money is more informative than announcing a training programme after selecting the staffing cuts.
Workers have knowledge that a purchasing team may lack: which apparently repetitive calls conceal distress, which exceptions prevent costly mistakes, and where customers need a person who speaks their language. Asking for that knowledge can improve the investment itself. Paying people for the time needed to test alternatives makes participation practical, especially for staff who cannot afford unpaid meetings.
There is some evidence for taking participation seriously. An OECD study published in 2023, covering manufacturing and finance in seven countries, found that training and worker consultation were associated with better worker outcomes: https://www.oecd.org/en/publications/the-impact-of-ai-on-the-workplace-main-findings-from-the-oecd-ai-surveys-of-employers-and-workers_ea0a0fe1-en.html
It does not establish that consultation caused those outcomes, and it cannot stand in for evidence from an Indian workplace. It does challenge the assumption that employees’ involvement is merely an obstacle to efficient adoption.
The economic issue is broader than compensation. If managers measure an investment mainly by payroll savings, they may overlook ways of using AI to increase quality or reach new customers. If they count promised new business without testing demand, they can make the opposite mistake. A useful assessment puts both claims under scrutiny, including licensing, integration, supervision and training costs.
India has particular reason to care. Its Economic Survey 2025–26 identifies the tension between higher productivity and absorbing labour, while warning that delaying adoption can also trap firms in low productivity: https://www.indiabudget.gov.in/economicsurvey/doc/eschapter/echap14.pdf
That is an argument for examining feasible choices in specific industries. It offers no easy national formula.
The assessment’s boundary matters, too. Imagine the firm retains its permanent employees but sharply reduces work sent to an outside agency. A reassuring internal headcount could conceal losses elsewhere. Contractors should therefore be heard alongside employees, and participation should be possible through local-language discussions, not just an online questionnaire.
The strongest objection is that this could become another expensive approval ritual. Large incumbents can employ compliance teams; smaller rivals cannot. A poorly designed requirement might protect established businesses while slowing useful tools that make work safer or less exhausting.
That objection should shape the design. Reviews should be proportionate, time-limited and focused on consequential changes. Small firms could share independent technical advice through industry or public institutions. Employers should document alternatives and explain their choices; they should not have to pretend that every task can survive.
The ILO’s 2025 assessment stresses that exposure to generative AI is not a count of actual job losses and that many jobs are more likely to change than disappear: https://www.ilo.org/publications/generative-ai-and-jobs-2025-update
There is room for deliberate design, though no assurance of a painless transition.
A better future would use paid pilots and worker knowledge to improve both services and jobs. An uneven one would give permanent staff a voice while contractors absorb the adjustment. A harmful one would make employment assessments ceremonial—or use them to shield incumbents from competition. Which emerges depends partly on whether anyone can still alter the plan when the consultation begins.
Questions for founders and investors
Before approving an AI investment, ask which alternatives were tested, whose earnings and hours were counted, and what changed because workers spoke. Fund the pilot and publish the reasons for the decision. The most revealing evidence of participation is a decision that was still open to revision.
Sources
China announcement: https://english.www.gov.cn/policies/latestreleases/202610/23/content_WS6adb4e53c6d00ca5f9a0deb7.html
Full policy text: https://www.mee.gov.cn/zcwj/gwywj/202610/t20261024_1184583.shtml
OECD workplace surveys: https://www.oecd.org/en/publications/the-impact-of-ai-on-the-workplace-main-findings-from-the-oecd-ai-surveys-of-employers-and-workers_ea0a0fe1-en.html
India Economic Survey, chapter 14: https://www.indiabudget.gov.in/economicsurvey/doc/eschapter/echap14.pdf
ILO research brief: https://www.ilo.org/publications/generative-ai-and-jobs-2025-update