🔍 Read the full analysis: The Effort Behind Trusting AI: What Workplace Studies Show on ThorstenMeyerAI.com
Get the little things that make your day delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
A November 16 essay by Georgia Tech cognitive engineer Karen Feigh argues that checking AI output can absorb time and mental effort, a burden workplace productivity measures may miss. Studies she discusses report slower completion in one trial of experienced developers, faster work on certain writing and consulting tasks, and lower accuracy when AI is used beyond its capabilities.
Workplace studies cited by Georgia Tech cognitive engineer Karen Feigh show that AI can speed up some professional tasks while adding review work or lowering accuracy in others, complicating claims that faster first drafts mean higher productivity. In an essay published November 16 in The Conversation, Feigh calls the effort spent checking whether AI output is accurate and appropriate the “verification burden.”
The clearest example of a time cost comes from a randomized trial by METR, a research group that studied 16 experienced open-source developers completing 246 real tasks in large software projects. Tasks were randomly assigned to allow or disallow AI tools. In the study, developers took 19% longer on average when AI was allowed. The researchers measured task completion time, including work after code review, rather than only the time needed to produce an initial draft.
The result differed sharply from participants’ expectations. Before the trial, developers predicted AI would reduce their completion time by 24%; after it, they still estimated that it had saved them 20%. The study reports that developers accepted fewer than 44% of AI-generated code suggestions. In a survey, 75% said they read every line of AI-generated code, while 56% said they often had to make major changes to clean it up. Screen recordings for a subset of tasks indicated that about 9% of working time went to reviewing and cleaning AI output.
Other studies cited by Feigh show that these results are not a general measure of AI’s effect on all work. A study of mid-level professional writing tasks found that ChatGPT reduced completion time by 40% and increased quality by 18%. In a Boston Consulting Group experiment, participants using AI completed tasks judged to be within its capabilities 25.1% faster on average. On a separate task outside those capabilities, however, AI users were 19% less likely to produce a correct answer than participants without AI.
AI and work · explainer
The verification loop: where AI’s time savings can go
The draft is fast. Knowing whether it is right is the work that follows.
Time for one task (schematic)
Draft: secondsgood and bad parts in the same confident voice About 9% of the timewent to reviewing and cleaning AI output in METR’s trial with AI allowed Speed is not productivitycount the whole loop, errors included
How the hidden cost builds up
- DraftThe AI answers in seconds, fluent and confident“few clues about where scrutiny is needed”
- CheckDecide what to check and how, without knowing if or where an error iscognitive load
- FixFinding an error does not end the work, it starts a correction cycletime
- Re-checkEvery fix needs checking again; when nothing is wrong, the effort feels wastedmotivation
- Count it allProductivity is the whole human-AI workflow, not the first draftnet, not gross
Source: Karen Feigh, Georgia Tech, The Conversation, Nov 16, 2026 (the three burdens, the quotes); METR, arXiv 2507.09089, p. 20 (time spent reviewing and cleaning AI output). The bars are a schematic, not data.
Graphic: Powered by Thorsten Meyer AI
Why Review Time Changes Productivity
The findings matter because workplace claims about AI productivity can focus on how quickly a tool generates an answer, rather than the time needed to check, correct, or discard it. If employees remain responsible for the result, that follow-up work is part of the task. A polished response may save time when it is reliable and suited to the work; when it is wrong or incomplete, the apparent saving can shrink or disappear.
The studies also point to a risk that is different from simply making more mistakes. Errors that require understanding a task’s rules may be harder to catch than visible typos. In an experiment involving 2,784 people checking AI-style suggestions, participants caught typos most of the time but were correct only 31% of the time on errors requiring rule-based understanding. That finding suggests review quality cannot be assumed from the fact that a person looked at the output.
For employers, the evidence supports measuring the full workflow, including review and correction, and distinguishing tasks where AI helps from those where it may mislead. It does not establish that AI universally reduces productivity, or that every use creates a net cost. The results instead show why performance depends on the task, the tools, and the person’s ability to verify the answer.
Products Worth Considering

Shark Rocket Pet Plus Corded Stick Vacuum, Bordeaux/Silver, HV322

Dyson V15 Detect Cordless Vacuum

Levoit Core 300S-P Smart WiFi Air Purifier for Home and Pets
As an affiliate, we earn on qualifying purchases.
Different Tasks, Different Results
Feigh’s piece is an expert essay, not a newly reported experiment. It draws on earlier research, including the METR developer trial and the Boston Consulting Group study of knowledge workers. The examples cover different occupations and measures, so their results should not be treated as directly comparable estimates of one workplace-wide effect.
The METR trial involved experienced developers working in mature repositories and used Cursor Pro with Anthropic’s Claude 3.5 or 3.7 Sonnet during the study period from February to June 2025. Its authors cautioned that the result does not show that current AI systems are unhelpful in every economically relevant setting, nor does it rule out future tools improving performance in that particular setting.
Feigh connects the review problem to a broader concern about automation: a system may exercise practical influence over work while a human remains accountable for the outcome. She describes AI as resembling a fast, polished intern whose confident delivery does not reliably signal where scrutiny is needed. That is her analogy and interpretation, rather than a measured result from the cited trials.
“AI’s productivity gains “are often countered by time and mental energy spent determining whether the output is accurate and appropriate.””
— Karen Feigh, cognitive engineer at Georgia Tech, writing in The Conversation
Where the Evidence Does Not Reach
The studies do not establish one overall productivity effect for workplace AI. They examine different tasks, participants, tools, and definitions of speed and quality. The METR result applies to the developers and projects in that trial; it cannot by itself predict outcomes for other software teams or occupations.
It is also unclear how much review time workers spend across a typical day, or how that time changes as tools and workplace practices evolve. The 9% review estimate came from a subset of recorded tasks, while several other figures came from participant surveys. Feigh’s essay argues that verification deserves attention, but it does not supply a new, broad measurement of that burden across industries.
The 2,784-person error-checking experiment also does not mean that workers will miss 69% of all AI errors: its 31% accuracy figure refers specifically to errors that required understanding task rules. The available material does not provide a general error-detection rate for everyday workplace AI use.
Measure the Whole Work Process
The practical implication in Feigh’s account is to evaluate AI across the complete workflow, rather than counting only how quickly it creates a draft or suggestion. That means tracking the time spent reviewing and correcting output, as well as whether the finished work is accurate and appropriate for the task.
Employers also need to account for who remains responsible for the final result and whether those workers have enough time and training to check AI assistance. The cited material does not announce a new policy or a follow-up study. Further evidence across different jobs and tools will be needed to show how often verification offsets time savings and which review practices catch consequential errors.
Key Questions
Did the developer study find that AI always slows workers down?
No. The 19% longer completion time was the average result for 16 experienced open-source developers working on 246 tasks in one METR trial. The researchers said it does not show that AI is unhelpful in every setting.
What does “verification burden” mean?
It is Karen Feigh’s term for the time and mental effort workers spend deciding whether AI output is accurate and suitable, and checking or correcting it before relying on it.
Can AI make work faster?
Yes, in some measured settings. A professional-writing study reported tasks completed 40% faster, and a consulting experiment found a 25.1% average speed advantage on tasks within the AI’s capabilities. Those results do not apply to every type of work.
What happened when the consulting task was beyond AI’s capabilities?
Participants using AI were 19% less likely to produce a correct solution on one task outside the system’s capabilities than participants who did not use AI.
What should employers measure?
Feigh’s argument is to measure the full workflow: drafting, review, corrections, and the accuracy of the final work, rather than treating the speed of an AI-generated first answer as the productivity result.
Source: ThorstenMeyerAI.com
