AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: How To Weigh AI’s Workplace Benefits Against The Cost Of Reviewing It on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

An expert commentary by Georgia Tech cognitive engineer Karen Feigh argues that AI’s productivity gains can be offset by the time and effort workers spend checking its output. Research cited in the commentary shows mixed results: a trial of experienced developers found they took longer with AI, while other studies found faster work on tasks suited to the tools.

Georgia Tech cognitive engineer Karen Feigh argued in a November 16 commentary that workers’ time spent checking AI output can offset its productivity gains, a concern supported by studies that found both faster work on some tasks and slower or less accurate results on others. The findings suggest employers should measure the full workflow, including review and correction, rather than treating a quick AI-generated draft as proof that a task took less time.

Feigh, writing in The Conversation, calls the effort to determine whether AI output is accurate and appropriate the “verification burden.” Her commentary is expert analysis, not a new experiment. It draws on research from software development, consulting and professional writing, where results differ according to the task and the AI tool’s capabilities.

The clearest example of a potential time cost comes from a randomized trial by METR, a research group. Sixteen experienced open-source developers completed 246 tasks in established software projects. In the trial, tasks that allowed AI tools took developers 19% longer on average than tasks without them. The measured time included work after the AI generated code, such as reviewing and changing it.

That result came with a gap between expectation and measurement: participants had expected AI to reduce completion time by 24%, and after the trial estimated that it had saved them 20%. METR reported that developers accepted fewer than 44% of AI code generations; 75% said they read every line of generated code. In a subset of tasks with usable screen recordings, about 9% of AI-allowed working time went to reviewing and cleaning output. Those figures describe this study and its participants, not every workplace or AI system.

At a glance
analysisWhen: Commentary published November 16, 2026;…
The developmentA November 16 commentary by Karen Feigh brings together research showing that AI’s workplace productivity benefits vary by task and can be reduced by the cost of reviewing its output.

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)

AI draft Checking Fixing Re-checking What a speed metric sees What the work costs

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

  1. DraftThe AI answers in seconds, fluent and confident“few clues about where scrutiny is needed”
  2. CheckDecide what to check and how, without knowing if or where an error iscognitive load
  3. FixFinding an error does not end the work, it starts a correction cycletime
  4. Re-checkEvery fix needs checking again; when nothing is wrong, the effort feels wastedmotivation
  5. Count it allProductivity is the whole human-AI workflow, not the first draftnet, not gross
The real measure: “whether the entire human-AI workflow produces trustworthy work at a lower total cost”

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 the Productivity Case

For employers, the practical question is not just whether AI produces a draft faster, but whether it reduces the time needed to deliver accurate, usable work. If staff have to inspect, test, correct or replace generated material, those tasks belong in any assessment of productivity. Ignoring them can make a tool appear more beneficial than it is across the complete process.

Accuracy also matters because review does not automatically catch mistakes. Feigh cites an experiment involving 2,784 people checking AI-style suggestions: participants usually caught typos, but were correct only 31% of the time on errors requiring knowledge of task rules. That result suggests polished output may still require expertise to assess, particularly when errors are subtle or depend on specialist context.

Feigh describes workers as supervising a fast, polished intern that can produce good and bad work in a similar confident voice but does not take responsibility for consequences. The analogy points to an organizational issue: people remain accountable for output even when the system contributes to it. Employers may need to account for that responsibility through appropriate review time, training and workflow design.

Products Worth Considering

Sony WH-1000XM5 Wireless Noise-Canceling Over-Ear Headphones, Black, International Version

Sony WH-1000XM5 Wireless Noise-Canceling Over-Ear Headphones, Black, International Version

As an affiliate, we earn on qualifying purchases.

Results Vary Across Workplace Tasks

The METR trial focused on experienced developers working in large, mature open-source projects. Participants mainly used Cursor Pro and Anthropic’s Claude 3.5 or 3.7 Sonnet, tools available during the study period from February to June 2025. METR cautioned that its findings do not show that AI is unhelpful in all economically relevant settings, or that future systems could not improve performance in the same coding tasks.

Other cited research found gains in different settings. A study of mid-level professional writing tasks reported that ChatGPT reduced completion time by 40% and increased quality by 18%. In a preregistered Boston Consulting Group experiment with 758 knowledge workers, participants using AI completed 12.2% more tasks and finished 25.1% faster on tasks within the tools’ capabilities. On a separate task designed to fall outside those capabilities, AI users were 19% less likely to produce a correct solution.

These results are not interchangeable: they involve different people, tasks, tools and measures. Taken together, they support a limited conclusion that AI can help with some work while adding risk or overhead elsewhere. Feigh also points to a longstanding concern in automation research: systems may exercise practical influence over work while human workers retain responsibility for the result.

“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, in The Conversation

Where the Evidence Still Falls Short

The cited studies do not establish a single average productivity effect for AI across workplaces. Their results depend on task type, worker experience, available tools and how time or quality is measured. The METR trial, in particular, was small and concerned experienced developers working on specific open-source projects; its 19% slowdown should not be treated as a prediction for all software teams.

It is also unclear how the balance changes as AI tools, user experience and workplace processes evolve. The studies cited here do not provide a universal estimate for the time needed to review generated work, or a common method for comparing review costs across industries. The commentary’s evidence supports the need to measure those costs, but does not settle what the net benefit will be for a particular organization.

Measure the Complete Work Cycle

Feigh’s practical recommendation is to track the entire process: time spent prompting, reviewing, checking, correcting and completing a task, alongside the quality of the final work. Employers can compare AI-supported workflows with existing ones on specific task types instead of assuming that results from one setting will transfer to another.

Organizations also need to decide who is responsible for checking AI-assisted work and give those workers enough time and training to do it. Further evidence will be needed to show how review costs change with newer tools and with repeated use. Until then, the studies point to a cautious measure of productivity: count usable, accurate work delivered, not just the speed of the first draft.

Key Questions

What does “verification burden” mean?

It is Karen Feigh’s term for the time and mental effort workers spend checking whether AI output is accurate and appropriate, and correcting it when needed.

Did AI make developers slower in the METR trial?

Yes. In a randomized trial involving 16 experienced developers and 246 tasks, tasks that allowed AI took 19% longer on average. The finding applies to that trial’s participants, tools and projects, not necessarily to all coding work.

Does research show AI always reduces workplace productivity?

No. Other cited experiments found faster work on professional writing and consulting tasks suited to AI. Results varied by task, and the consulting experiment also found lower accuracy on a task outside the tools’ capabilities.

What should employers measure when evaluating workplace AI?

They should measure the full workflow, including review, testing, correction and final work quality, rather than timing only how quickly AI produces an initial response.

Source: ThorstenMeyerAI.com

You May Also Like

How to Choose Wireless Noise-Canceling Headphones

Learn how to set up, connect, and optimize wireless noise-canceling headphones for the best sound experience with this step-by-step guide.

OpenAI’s ChatGPT Voice Plans: What Ends On December 11

OpenAI will remove Advanced and Standard voice from ChatGPT on Dec. 11, 2026, leaving GPT-Live as the default. Some features remain on Advanced for now.

Source Control At Anthropic: What’s Changing?

Anthropic’s headline points to a source-control redesign, but available information does not explain its scope, motivation, rollout or results.

Laptop Graphics Explained: 10 Best External GPUs To Explore In 2027

A 2027 eGPU roundup compares 10 docks and enclosures, focusing on GPU inclusion, connection types and host compatibility.