Question Is people's reluctance to use AI at work better explained by worries about their jobs, or by doubts about the quality of what it produces?
Across the 18 interviews (6 creative professionals, 6 scientists, 6 general workforce), find each reason an interviewee gives for avoiding, limiting or not relying on AI, and each mention of AI taking work, and say which of the two explanations accounts for more interviewees' own reluctance.
Doubts about the quality of what AI produces explain far more of these interviewees' own reluctance than worries about their jobs. The evidence for this judgement is strong. In 14 () of the interviews, the person limits, checks or holds back from AI because its output may be wrong, miss context or lack nuance. This covers all of the general workforce, 6 (), and 4 () scientists and 4 () creative professionals. Not one interviewee gives worry about their own job as a reason for avoiding, limiting or not relying on AI. AI taking work does come up, in 9 () interviews, but as a concern about their field or about other people, never as the reason they hold back.
18 interviews in 3 groups · 37 passages coded · 14 October 2026 · Run 2026-10-14-zq5p
Verification in this report:
9 numbers, each written by the AI and shown beside the count code made from the passages it cited, which it must equal. Click any number to see who it counts.
37 quotations, every one found word for word in its interview. Click an interview's ID beside a quotation to read it in place.
These definitions were my own and set after reading, because the evaluator was not available. A reason for reluctance is a reason the person gives for their own avoiding, limiting or double-checking of AI. A mention of AI taking work is any reference to AI replacing people or work, whether it happened or might happen.
What the quality doubts look like
The quality reasons are direct and tied to specific tasks. A healthcare administrator keeps billing entries away from AI because "currently it is too vital to entrust to AI as there is too much nuance to the entries" []. A physicist will not let AI summarise papers: "the trust in an AI model to do this correctly, with absolutely no mistakes just isn't there for me yet" []. A microbiologist's experiences with tacit lab cues "have made me hesitant to use AI tools for this type of task" []. A teacher writes their own worksheets because "AI never quite hits the mark" [], and an investor-relations writer says "AI still isn't good enough to capture nuanced language" []. Some of these doubts are mild, such as having to "check its math" []. One has gone away with use: a seamstress found that AI "is always accurate so this is no longer a concern for me" [].
How job loss appears instead
Where AI taking work comes up, it does not turn into reluctance. The two people who actually lost work to AI both use it heavily. A copywriter had "one client replace me as her blog writer in favor of ChatGPT" [], and a teacher's legal freelance writing "dried up the second AI could produce an article that's 75% as good as mine" []. The teacher's response was to plan for working with AI. Four () interviewees reject the worry outright, from "This isn't something I worry about" [] to "the risk is overblown" []. Others speak about their field: fewer opportunities for voice actors [], streams lost by real musicians [], and software being written by AI [].
The voice actor comes closest to linking the two. They fear "lessening opportunities for many voice actors" and keep their performances free of AI. But the reason they give for that boundary is authenticity, "to maintain authenticity" [], not protecting their job.
The other reasons
The main alternative to quality is not job fear but the value of human work. Three () creative professionals keep AI out of their core craft because of the human element or because they enjoy doing it [] [] []. A few reasons appear only once or twice:
wanting to understand one's own code []
confidentiality or data privacy, in 2 () interviews
wanting control over business data []
no need, because a task already uses established methods []
Limitations
Prompted answers. The interviewers asked directly about concerns and about "maintaining your skills in areas where AI could potentially take over". Some mentions of job loss were prompted. Some quality reasons came in answer to a question about tasks people prefer to do themselves.
Who was interviewed. Every interviewee uses AI. These are people who limit AI, not people who refuse it, and job fear may weigh more among non-users.
Single coder. One reader coded the interviews, under definitions made after reading. Borderline cases, such as the voice actor, are judgement calls.
Annexes
How this report was written
What was read. The AI that wrote the answer was given all 18 documents; what it read of them is not recorded.
The workflow x-ray beside Step by step, below, draws the same procedure, a box for each stage and what it made.
Step by step
The answer was written first, and the passages each of its numbers rests on were cited afterwards. Each count was then made directly from the passages cited rather than by the AI, so anyone can check a number against the passages behind it. Whether the passages cited are all that the documents hold on each point is not checked.
What was read
All 18 documents, given to the AI that wrote the answer: 6 creative, 6 scientist, 6 general workforce.
Coding: C
one row for each passage that fits; 37 rows.
The instruction a second coder would follow
Rows cited by an answer written by the calling model: each is a passage that a stated count rests on.
Claim
the label of the stated count this passage supports
The coded passages 37 rows
Writing the rhythms and melodies by hand is the crucial human element. I'm much more interested in it helping with technical tasks.
taking away streams from real music that someone slaved over
I already had one client replace me as her blog writer in favor of ChatGPT
I already had one client replace me as her blog writer in favor of ChatGPT
I occasionally have to check its math, but I'm still saving quite a few hours by using AI.
Sometimes, it will tell me I've made mistakes which aren't really mistakes.
I'm not concerned that it will take my place as a creative writer.
It may mean I have fewer jobs as an editor, but I also don't think that's a bad thing.
My boundaries won't shift because I love writing so much. It would be no fun if a computer got to do it all.
I would prefer to do everything myself to maintain authenticity
I will often have to double check their work to make sure they didn't do anything unnatural to the audio
I fear that this will lead to a lot of content that will sound very similar, while also lessening opportunities for many voice actors.
I was apprehensive about its abilities and needed to double check its maths for peace of mind, but I have found that it is always accurate so this is no longer a concern for me.
Getting an AI to write this code would not be useful for me, as I would not understand it as well as if I had written it myself.
it would be hard to trust an LLMs response since it wouldn't necessarily have a source to reference
maybe all of this software will be produced by AI and human developers will have minimal involvement
an AI tool would not understand the difference to look for. Multiple experiences like this have made me hesitant to use AI tools for this type of task.
Those are both historically studied and known methods.
I think the trust in an AI model to do this correctly, with absolutely no mistakes just isn't there for me yet.
The AI doesn't have this information, so it might remove those points.
currently it is too vital to entrust to AI as there is too much nuance to the entries
I feel I have better control over the data.
AI may not fit in well and can be less efficient and make thing worse
it is important to develop skills in relating technology to business, which is less likely to be taken over by AI
it can't add the touches that are going to make it sound like it's perfectly in my voice
I spent a couple of years doing legal freelance writing, and that work dried up
that work dried up the second AI could produce an article that's 75% as good as mine
I'm nervous about the safety of giving too many companies data that makes me easier to market to
AI never quite hits the mark
This isn't something I worry about.
AI still isn't good enough to capture nuanced language
there is some confidentiality issues in place
AI is meant to make you more productive rather than put you out of a job altogether.
AI still can't replace customer relationships and the human element of the job.
AI is fallible, and I would not want it automatically acting on my behalf.
I feel like there is so much to gain in the realm of using AI to be a more effective employee that the risk is overblown.
that's not to take away from the volume of workers who have lost their roles in customer service and the like
Counting: T
Documents by claim
The table 9 cells
Claim
Documents
of 18
of 18
of 18
of 18
of 18
of 18
of 18
of 18
of 18
Counting: Kind
Documents by claim and kind
The table 27 cells
Claim
Kind
Documents
General workforce
of 6
Creative
of 6
General workforce
of 6
Creative
of 6
Scientist
of 6
Creative
of 6
General workforce
of 6
General workforce
of 6
Scientist
of 6
General workforce
of 6
Creative
of 6
General workforce
of 6
Scientist
of 6
Creative
of 6
Scientist
of 6
Creative
of 6
Scientist
of 6
Creative
of 6
Scientist
of 6
General workforce
of 6
Scientist
of 6
Scientist
of 6
Creative
of 6
General workforce
of 6
Scientist
of 6
Creative
of 6
General workforce
of 6
Writing: Answer
The instruction the answer was written to
What the evaluator was asked to agree
Nothing was put to the evaluator to agree before the answer was written.
The checks that were run
37 of 37 coded passages found word for word in their document.
14 of 14 quotations in the answer found word for word in the documents (a quotation of under twelve characters, such as one word, is not checked).
9 numbers in the answer, each written by the AI beside the count code made from the passages it cited, and equal to it; 0 written by the AI without a count behind them.
0 citations to passages that do not exist; 0 quotations not in the passage they cite.
The record
This is run 2026-10-14-zq5p. Its ID stays the same through every revision, and the zip's fingerprint, below, names this version of it. This report was drawn directly from the run's zip, is-peoples-reluctance-to-use-ai-at-work-better-2026-10-14-065cca5b.zip. The zip's fingerprint is 065cca5bb322b81a55768dba9873b1ce636905137024da8559fce22cd6e86d4a. It is the SHA-256 of the zip's SHA256SUMS file, which lists the SHA-256 of every other file in the zip, so changing any file breaks the match. To check that the zip is the one this report was drawn from, unzip it, run sha256sum -c SHA256SUMS, then sha256sum SHA256SUMS, and compare the result with the fingerprint above. Rubicon's own code, which is public, checks the rest: python verify.py <report> <zip> recounts every number from the coded passages in the zip, draws the report again from the zip and says whether it matches this one exactly.
Every interview at a glance
Each row is one interview, each column a reason as the coding defines it. Click a mark to read the passages.
🔥 Carbon cost of this report: Between 0.00088 and 0.14 km driven in an average UK family car
Making this report used between 1.2 and 190 watt-hours of electricity, which is the carbon of driving an average UK family car for that distance. A ballpark only: Anthropic publishes no energy figures, and the run, made in the evaluator's own Claude, records no token counts, so the estimate is made from the run's size. The 18 documents were read in full once by each of the 1 coding steps, and the writer read the tables; everything the run wrote counts as output. That comes to about 47,350 tokens read and 2,655 written, at 4 characters a token.
High: 0.39 Wh per thousand tokens read and 1.95 per thousand written, with what was read counted 10 times over for the agent's own overhead. Simon P. Couch, 'Electricity use of AI coding agents', 20 January 2026, https://simonpcouch.com/blog/2026-01-20-cc-impact/: about 390 Wh per million input tokens and 1,950 per million output tokens for Claude Opus, back-solved from Epoch AI's estimate for a 100,000-token prompt with output at five times the input rate, as Anthropic's prices are. The author calls them napkin maths. Our assumption, not a measurement: an agent re-sends its instructions, earlier turns and tool output with every request, so it reads far more than the documents themselves. The low case counts only what was read and written; the high case counts what was read ten times over.
Low: the same rates divided by 20, what was read counted once. Oviedo et al., 'Energy use of AI inference, efficiency pathways, and test-time scaling', arXiv:2509.20241: a median 0.31 Wh per query for frontier models, and widely cited estimates overstated by 4 to 20 times. The low case takes the high case's rates divided by the top of that range.
0.124 kg CO2e per kWh. DESNZ greenhouse gas reporting conversion factors, UK grid electricity, 2024.
0.16601 kg CO2e per km. DESNZ greenhouse gas reporting conversion factors, average car of unknown fuel, 2024.