AI
“For years, many bosses have been adamant that employees use AI at all costs, despite research suggesting there's little to gain from it in terms of productivity, evidence that flies in the face of the tech industry's repeated promises of a workplace revolution brought on by the tech.”
Plain restatementOver a multi-year period, a substantial number of employers have pushed or required staff to use AI tools, and the research literature indicates that AI produces little productivity benefit.
This post correctly reports that many employers pushed staff to use AI, and the study it cites is real, peer reviewed, and open access, with the 68 percent versus 24 percent trust figures matching the university's own announcement. The problem is the framing. The post says research shows little productivity gain from AI, but the research is split rather than settled: controlled experiments on specific tasks generally find gains, including a 15 percent improvement among more than five thousand customer support agents and a moderate positive average across coding studies, while studies of company profits and national earnings data find close to zero. Both findings are real and they measure different things. The post also leaves out the most important result in the study it relies on: the loss of manager trust largely disappeared when the company's policy explicitly encouraged AI use, and was worst where AI was banned or where no policy existed. That finding cuts against the headline, since it suggests clear guidance helps rather than that telling staff to use AI was the error. One further stretch: the study measured whether supervisors said an employee seemed promotable in an experiment, not whether anyone actually lost a promotion. What remains genuinely unresolved is how much AI helps overall, because that answer changes depending on whether you measure individual tasks, company profits, or wages.
[drifted from the evidence:] For years, many bosses have [drifted from the evidence:] been adamant that employees use AI [drifted from the evidence:] at all costs, despite research [drifted from the evidence:] suggesting there's little [drifted from the evidence:] to gain from it in terms of productivity, [drifted from the evidence:] evidence that flies in the face of the tech industry's repeated promises of a workplace revolution brought on by the tech.
[added by the neutral restatement:] Over a multi-year period, a substantial number of employers have [added by the neutral restatement:] pushed or required staff to use AI [added by the neutral restatement:] tools, and the research [added by the neutral restatement:] literature indicates that AI produces little productivity [added by the neutral restatement:] benefit.
Red-tinted words in the claim drifted from the evidence. Green-tinted words are what a neutral restatement needs.
The trace / claim to source
- Employer pressure to adopt AI is real and documented at named companies. Shopify's April 2025 memo made AI use a baseline expectation tied to performance reviews and headcount requests.
- Formal AI policies spread quickly among employers, rising from 38% to 68% in one employer survey between 2025 and 2026.
- Part of the research literature does support the "little to gain" reading, specifically at the firm and labor-market level. The Danish administrative-data study rules out earnings and hours effects larger than 1% while measuring average time savings of 2.8%, and METR's RCT measured a 19% slowdown for experienced open-source developers on their own repositories.
- The post's cited study exists, is peer reviewed, is open access, and is a registered report. The 68% versus 24% trust figures match the Hebrew University release, and the image caption's "almost triple" is arithmetically right at a ratio of about 2.8.
- The Business Insider anecdote is real. The CEO described is Michelle Volberg of tech recruiting firm Twill.
- The Inc. anecdote is real. An anonymous CEO threatened to fire the next person sending unedited ChatGPT output, reported by Joe Procopio.
- The caption presents "research" as pointing one way. The literature is split along a measurement level. Controlled experiments report gains of 14 to 55 percent with a sample-size-weighted mean near 21 percent, a pre-registered meta-analysis of coding assistants finds a moderate positive effect, and a study of over five thousand support agents finds a 15 percent gain, while administrative and firm-level data find near-zero. Selecting only the null-result half and calling it "research" manufactures a settled consensus that does not exist.
- The study the post uses as its "case in point" found that the trust penalty was not significant when the organization's policy encouraged AI use, and was present when use was forbidden or when no policy existed. That qualifier runs against the post's thesis. On the study's own finding, bosses who told staff to use AI are not the ones producing the trust penalty; the penalty is largest where there is a ban or silence. The post omits this entirely.
- The same study found AI-assisted work was trusted more than plagiarized work, a comparison that bounds the size of the stigma and is left out.
- The caption says AI use is "actively hurting their chances of promotion." The study measured supervisors' perceived promotability inside experiments, not actual promotion outcomes in real workplaces. Perceived promotability in an incentivized game is not a promotion decision.
- A trust study is used as evidence for a productivity claim. The paper measures whether supervisors rely on or verify an employee's report. It does not measure output, speed, or quality, and so it cannot support "little to gain in terms of productivity." The authors' own framing treats performance benefits from LLMs as given and asks why trust falls anyway. Bigman's quoted line makes this explicit in the post's own text: an employee may use AI to produce better work more efficiently yet still lose their manager's trust.
- "Many bosses have been adamant that employees use AI at all costs" compresses a spectrum into a mandate. Roughly 6 percent of workers in two separate 2026 surveys say AI use is actually required, while a much larger share report policies, encouragement, or performance-review weighting. Encouragement, policy, and requirement are different things, and the strongest word is applied to all of them.
- The headline frames executives restricting AI in internal messages as bosses realizing a mandate was a mistake. The Business Insider case describes a company that had deployed an enterprise Claude subscription and then restricted AI for internal communications specifically while still permitting it for brainstorming. That is a scoping of use, not a reversal of adoption.
- Whether the research literature as a whole supports "little to gain" cannot be resolved as stated, because the claim never specifies the measure. At the task level the weight of evidence is positive; at the firm and earnings level it is near zero. Both readings are defensible for their own measure, and neither is defensible as a blanket statement.
- How widespread genuine AI mandates are remains unsettled. Survey estimates of required use, policy existence, and performance-review weighting range from about 6 percent to 68 percent depending on what question was asked and of whom, and the exact question wording behind most of these figures was not available.
- Which of the six studies in the paper produced the 68 versus 24 split, and its sample size, is not confirmed from the paper itself.
- Whether the measured trust penalty persists outside incentivized experiments and into real promotion and performance decisions over time is not established by this study design.
On the employer-pressure half of the claim, the push is real and documented. Shopify CEO Tobi Lütke published an April 2025 memo making reflexive AI use a baseline expectation, counted in performance reviews, and requiring teams to show why AI could not do the work before requesting headcount. Survey aggregation puts formal policy adoption high and hard mandates low: Littler's 2026 Annual Employer Survey of 303 legal and HR professionals found 68% of employers report a formal AI use policy, up from 38% in 2025, while both the CNBC/SurveyMonkey Q3 2026 survey and the Founder Reports survey land on 6% of workers saying AI use is required by their company. A separate survey found 47 percent of respondents said they are already factoring AI usage into performance evaluations. On the productivity half, the research is split rather than settled. Pointing one way: METR's randomized controlled trial found that when developers were allowed to use AI tools they took 19% longer to complete issues, against an expectation of a 24% speedup and a post-hoc belief that they had been sped up by 20%, and Danish administrative data using difference-in-differences and employer policies as quasi-experimental variation estimated precise zeros, with AI chatbots having no significant impact on earnings or recorded hours in any occupation, confidence intervals ruling out effects larger than 1%, alongside average time savings of 2.8%. Pointing the other way: a study of 5,172 customer-support agents found access to AI assistance increased productivity, measured by issues resolved per hour, by 15% on average, with substantial heterogeneity across workers. A pre-registered synthesis found a moderate positive effect of GenAI coding assistants on productivity (Hedges' g=0.33, 95% CI 0.09 to 0.58, p=0.008) alongside substantial heterogeneity, and a cumulative meta-analysis showing a consistent, modest productivity gain that remains stable as studies are added. A systematic review frames the tension directly: the only study using nationally representative administrative records rules out earnings effects larger than two percent, yet controlled experiments over the same period report productivity improvements of 14 to 55 percent, with a sample-size-weighted mean improvement of 20.9 percent. On the study the post builds its case around, the retrieved abstract states: existing leadership theory cannot meaningfully explain how subordinates' LLM use influences supervisors' trust in them, and in six studies spanning five incentivized economic games and one field setting the authors investigated how subordinate LLM use affects supervisory trust. Use of LLM-generated output led to less trust than self-generated output but more trust than plagiarized output, and the authors explored organizational LLM policy, task criticality, and supervisors' prior LLM experience as boundary conditions. Critically: where policy encouraged use rather than forbidding it or stating nothing, LLM use did not significantly reduce trust, and there were significant downstream consequences of LLM use via supervisory trust on perceived subordinate performance, organizational citizenship behavior, and promotability. The press release confirms the headline numbers: employees who completed the work themselves had a 68% chance of earning their supervisors' trust, compared with just 24% when they used AI.
Complete reasoning
The reply is formatted for pasting into the thread where the claim is circulating.
Ask this case
Answers come only from the case file above; nothing is added.
Is it true that bosses have been pushing employees to use AI?
Yes. This part is well documented, including Shopify's 2025 memo making AI use a baseline expectation tied to performance reviews, and survey data showing formal AI policies at employers jumped from 38 percent to 68 percent in one year.
Does research actually show AI gives little productivity benefit?
The research is split, not settled. Studies of company profits and national earnings data find close to zero effect, but controlled experiments on specific tasks, like a study of over 5,000 customer support agents, generally find real gains, including a 15 percent improvement in that case.
What does the study behind this claim actually say about trust and AI use?
It found that using AI led to less trust than doing work yourself, but the trust penalty was not significant when a company's policy encouraged AI use. The penalty was worst where AI was banned or where there was no clear policy, which runs against the post's framing.
Does the study prove AI use hurts people's actual chances of promotion?
No. The study measured whether supervisors said an employee seemed promotable within an experiment, not whether anyone actually lost a real promotion.
So is the claim accurate overall?
It is rated partially accurate but misleading. The employer-pressure part is true, but the claim treats a mixed research record as if it were one-sided, and it leaves out the study's key finding that clear policies encouraging AI use largely prevented the trust penalty.