What Reviewing AI Can Mean For Employees And Their Workloads
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: What Reviewing AI Can Mean For Employees And Their Workloads 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

A November 16, 2026, essay by Georgia Tech cognitive engineer Karen Feigh highlights a potential verification burden: employees can spend time checking AI work, even when it appears polished. Studies cited in the discussion show mixed results, from slower completion in one software trial to faster work on tasks suited to AI, with accuracy risks outside its capabilities.

Georgia Tech cognitive engineer Karen Feigh argued in a November 16 essay in The Conversation that AI can shift work rather than simply remove it, because employees must spend time checking whether generated material is accurate and appropriate. Her argument draws on studies with mixed results: one trial found experienced developers took longer with AI tools, while other experiments found faster work on tasks within AI’s capabilities.

Feigh calls the time and mental effort spent checking AI output the “verification burden.” In her account, employees can receive a quick, polished answer but still need to inspect it for mistakes, missing steps, outdated information or bias. She describes the work as resembling supervision of a fast intern whose output may be useful or flawed, often without making clear which parts need scrutiny.

The most direct evidence she cites comes from a randomized METR trial involving 16 experienced open-source developers and 246 real tasks in large software projects. Developers were assigned tasks with or without AI access. In the study, conducted with tools available from February to June 2025, the AI-allowed group took 19% longer on average. Participants had expected AI to make them 24% faster before the trial and still estimated afterward that it had saved them 20% of their time.

METR’s measures also point to review work. Developers accepted fewer than 44% of AI code generations, and 75% said they read every line of generated code. In screen recordings for a subset of tasks, about 9% of working time went to reviewing and cleaning AI output. These figures describe that study’s participants and tasks; they are not a general estimate of how much review every employee must do.

At a glance
reportWhen: Essay published November 16, 2026; rese…
The developmentKaren Feigh’s November 16 essay brings attention to research showing that AI’s effects on employee workloads depend on how much review its outputs require and whether the task suits the technology.

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

Review Time Can Erase Speed Gains

The findings matter because a fast first draft is not the same as a faster completed task. If workers have to test, correct or rewrite AI output, organizations measuring only generation speed may miss work that remains with employees. That can affect staffing plans, deadlines and assessments of whether a tool has reduced workload.

The risk is not limited to extra minutes. Feigh points to research in which people were much more successful at spotting surface errors than errors requiring an understanding of task rules. In an experiment involving 2,784 people checking AI-style suggestions, reported accuracy on rule-dependent errors was 31%. This suggests that review itself can fail, particularly when output sounds confident or the reviewer lacks time or subject knowledge.

At the same time, the evidence does not support treating AI as a uniform source of extra work. In a study of mid-level professional writing tasks, ChatGPT reduced completion time by 40% and raised quality by 18%. The practical question for employers is where gains hold after review, and where verification costs or mistakes outweigh them.

Products Worth Considering

Sonos One Two-Room Set Smart Speakers with Built-In Alexa, Black

Sonos One Two-Room Set Smart Speakers with Built-In Alexa, Black

Logitech MX Creative Console with 9 Customizable Keys, Graphite

Logitech MX Creative Console with 9 Customizable Keys, Graphite

As an affiliate, we earn on qualifying purchases.

Results Depend on the Task

The consulting experiment Feigh cites illustrates why task boundaries matter. Conducted with Boston Consulting Group and involving 758 knowledge workers, it tested realistic tasks both inside and outside what researchers called AI’s “jagged technological frontier.” On 18 tasks within the frontier, participants using AI completed 12.2% more tasks and finished 25.1% faster on average, with higher-quality solutions. On one complex task placed outside the frontier, AI users were 19% less likely to produce a correct answer.

Those results do not establish that all AI-assisted work is faster or less accurate. They show different outcomes across the experiment’s task types. A separate writing experiment found substantial time and quality gains, while the METR software trial measured longer completion times in its particular setting. Differences in work, tools and participants make direct comparisons difficult.

Feigh’s essay is an expert analysis, not a new experiment. It draws together existing research and a longstanding concern in automation: people may remain responsible for outcomes even when software shapes the work. The studies supply evidence for particular settings, not a single workload estimate for employees across industries.

““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

How Much Review Work Is Typical?

The cited studies do not establish how much verification time employees typically need across workplaces. The METR review estimate came from a subset of tasks with valid screen recordings, and its participants were experienced open-source developers working in particular repositories with specific AI tools. The result may not transfer to other jobs, tools or levels of experience.

It is also unclear how review demands change as AI systems and workplace practices evolve. The evidence presented covers several task types and periods, but it does not provide a single measure of net workload after accounting for training, corrections, downstream errors and work that employees may no longer need to do. Feigh’s essay raises these questions; it does not resolve them.

Measure Completed Work, Not Drafts

Feigh’s recommendation is to measure the whole workflow, including checking and correction, rather than judging productivity by how quickly an AI tool produces an initial answer. Employers evaluating tools can compare time, quality and error rates across tasks, while recording how much human review is required and whether that review catches substantive problems.

Organizations also need to account for the employees who remain responsible for AI-assisted output. Feigh argues that they need enough time and training to assess it. The research cited does not set a standard for how much review or training is sufficient; that depends on the work and the consequences of an error. Further workplace evaluations will be needed to show whether AI reduces total workload in specific roles after those costs are included.

Key Questions

What does “verification burden” mean?

It is Karen Feigh’s term for the time and mental effort employees spend checking whether AI-generated work is accurate and appropriate.

Did AI make developers slower in the METR study?

In that randomized trial, 16 experienced open-source developers took 19% longer on average when AI tools were allowed across 246 real tasks. The finding applies to that study’s setting, not every software job.

Do the studies show AI always increases workloads?

No. Results varied by task. The METR trial found slower completion, while writing and consulting experiments found faster work in some settings. The cited research does not establish one outcome for all employees.

Why can AI-assisted work still need substantial review?

AI can produce fluent output that contains mistakes or misses task requirements. In the research Feigh cites, checking errors that required understanding the rules was harder than catching typos.

What should employers measure when adopting AI?

They should assess the completed workflow, including drafting, review, correction, quality and errors—not just the time it takes to generate a first draft.

Source: ThorstenMeyerAI.com

You May Also Like

A Shared Desk Brings My AI Agents Into One Workspace

Thorsten Meyer says Thrymvault gives his AI newsroom agents a shared workspace with approved access and traceable changes; some safeguards remain planned.

How to Measure Your Workspace Correctly Before You Buy Anything

Discover essential tips to accurately measure your workspace before purchasing anything and ensure a perfect fit for your space.

How to Reduce Notification Anxiety Without Going Offline

Navigating notification anxiety can be challenging, but discovering effective strategies helps you stay connected without feeling overwhelmed.

The “Shutdown Routine” That Makes Work-Life Balance Real

While establishing a shutdown routine can transform your work-life balance, discovering the key steps to make it effective will change everything.