Loading today’s stories
Build1 distinct publisher2 min readPublished
Positive AI comments in US employer reviews fell from 81 percent to 43 percent, and the sharpest skepticism sits with the software engineers reading the diffs rather than the architects and executives above them.
The Engineer · Build desk
Compiled by The EngineerSomething wrong?How this is made
The gradient inside the dataset is the part worth an hour. Software architects mention AI more often than any other role, and rate it mostly positively [4]. Executives are the most AI-positive group, and managers stay optimistic [5]. Software engineers, whose work is writing and reviewing code, land at 57 percent negative [3], four points worse than the overall workforce figure [5]. Architects specify systems; engineers merge them. The output that flatters a design review is the output that then has to survive a diff.
Insurance claims workers are the extreme case, and they name the mechanism plainly: management forces buggy tools on them [20]. That is the same complaint an engineer files, minus the vocabulary.
Then the arithmetic to do before quoting 53 percent as opposition. Glassdoor's analysts report that roughly 10 percent of negative comments are complaints that the employer is too slow to adopt AI or supplies the wrong tools, which makes those reviewers pro-AI in spirit [14]. Ten percent of 53 is 5.3 points, putting genuine opposition nearer 48 percent [3]. Positive and negative together come to 96, so about 4 points sit somewhere else [2]. The fall from 2019 is still 38 points [1], and no reweighting of the negative bucket touches that.
The controllable surface is small. Of detailed positive reviews, 44 percent come from workers whose employer profits from the AI boom [7], which is not a lever anyone can pull. Another 41 percent credit the employer for tools, training and support [8], which is. Against that, 14 percent of negative comments are about being forced to use AI [11] and 8 percent about unrealistic productivity expectations [13]. Supply versus mandate, measured in the same corpus.
Company size runs against the assumption that budget fixes this. Negative share is 51 percent at employers under 200 people and 67 percent above 10,000 [15], a 16-point spread [4] pointing at the firms most likely to have a training programme and a procurement process.
For any of this to describe a specific team, the rollout has to resemble the median one in the sample, and the sentiment has to be about the tool rather than the headcount plan sitting next to it. Glassdoor's analysts flag the second problem themselves, noting overlap with office roles where burnout and layoff fear were already common [17]. Their Gen Z result shows how fragile the controls are: for Gen X the gender gap disappears once industry and occupation are accounted for, and for Gen Z it does not [18], with 21 percent positive among Gen Z women against 42 percent among Gen Z men [19].
Ranked by verification strength, evidence, and original report placement.
A Glassdoor analysis found positive AI sentiment among US workers dropped from 81 percent in 2019 to 43 percent by mid-2026.
In the same Glassdoor analysis, negative AI comments rose to 53 percent.
Software engineers, whose work focuses on writing and reviewing code, are much more skeptical, with 57 percent of their AI comments skewing negative.
Software architects mention AI more often than any other role and rate it mostly in a positive light.
Executives are by far the most AI-positive group, and managers remain the most optimistic about AI.
AI mentions in US employer reviews jumped 240 percent from May 2025 to May 2026, and once a role hits moderate AI exposure the comments appear across nearly every profession.
Distinct publishers with included, body-backed reporting in this cluster.
2 articles · August 30, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
invest
Thomson Reuters trades Claude for a Qwen derivative it cannot let customers audit1 distinct publisher
invest
X discloses Chinese bot farm accounts posting on AI data centre and energy debate2 distinct publishers
build
The firm keeps the whole efficiency gain; the pipeline bill goes to everyone else1 distinct publisher
build
Claude opened 388 pull requests on Anthropic's own apps. 180 got merged.1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One outlet's recap of a dataset only its owner can see
The 81-to-43 collapse, the 57 percent engineer figure, the 67 percent at big employers — all of it reaches us through The Decoder's summary of an analysis Glassdoor ran on its own reviews, with no sample size, coding method, or year-over-year comparability shown. What lifts this off the floor is that the piece argues against itself where the data demands it: it flags the selection bias, and it concedes that about a tenth of the negativity is impatience for more AI. The arithmetic is still loose in places — the two positive buckets plus generic praise total 112 percent, so the categories overlap in ways nobody explains, and positive plus negative leaves four points homeless.
Everywhere on the desk, unmeasured in the ledger
The review corpus is unusually good at showing that AI has arrived and unusually bad at saying how much of it is in use. A 240 percent jump in mentions in twelve months, complaints about compulsory use, praise for employers that shipped training, adjusters describing tools pushed at them — that is the texture of live deployment reaching ordinary workflows, not pilots. But every observation here counts words about AI rather than seats, licences, or logged usage, and the whole picture is drawn from one self-selected platform.
Modest overreach, mostly about who is being spoken for
Sentiment souring is what the data says, so the headline is not doing violence to it. The stretch is in reach and resolution: a self-selected review corpus stands in for the American workforce's verdict, and the sharpest framing — that the skeptics sit closest to the keyboard — leans on two role-level cuts with no counts, when engineers land just four points off the average everyone else occupies. Credit where due: a story that tells you a tenth of its own negative share is pro-AI has already deflated part of its own balloon.
The corpus owner profits when its reviews read as a labour barometer
Glassdoor is both the subject and the instrument: every analysis that lands as a statement about the US workforce strengthens the platform's standing as a labour-market index, and its own analysts are the only interpreters on record. The outlet reproducing it closes with a subscription pitch for AI news 'without the hype', which is a commercial position as much as an editorial one. Among the cited corroborators is an Anthropic study of worker attitudes — a lab surveying sentiment toward its own product category. None of that makes the figures wrong; it does mean nobody with a reason to contest them has been near them.
The same article twice is not two sources
Our coverage holds one Decoder report in duplicate, so the apparent weight of reporting is an artefact of filing, not of independent confirmation. Nothing in the material contradicts anything else — there is nothing else. The direction travels better than the decimals: that workplace AI sentiment has fallen and that role and employer size shape it is safe enough to act on; the specific 43, 57, and 67 deserve a look at Glassdoor's underlying report before anyone quotes them in a board deck.