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

🔍 Read the full analysis: How Much Work Does It Take To Check AI? What Studies Indicate on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

TL;DR

Research on software developers, consultants and professional writers finds that AI’s effect on productivity varies by task. In a randomized trial, experienced developers took 19% longer with AI, while other experiments found faster work within AI’s capabilities and weaker accuracy beyond them. The studies point to checking time and error detection as important parts of measuring AI’s impact.

Research reviewed in a November 16 article by Georgia Tech cognitive engineer Karen Feigh suggests that checking AI-generated work can absorb some of the time and attention saved by producing it. The evidence is mixed: a randomized trial found experienced developers took 19% longer with AI tools, while experiments in professional writing and consulting found faster completion on some tasks.

Feigh describes the time and mental effort spent deciding whether AI output is accurate and appropriate as the “verification burden.” Her article in The Conversation is an expert commentary, not a new experiment. The studies she discusses measure different jobs, tasks, AI tools and outcomes, so their results should not be treated as a single estimate of AI’s effect on workplace productivity.

The most direct evidence on review time comes from a randomized trial by METR, or Model Evaluation & Threat Research. Sixteen experienced open-source developers completed 246 tasks in large software projects. Tasks were randomly assigned to allow or disallow AI. In the study, which used tools including Cursor Pro and Claude 3.5 or 3.7 Sonnet, developers took 19% longer on average when AI was allowed.

The developers’ expectations did not match their measured results: they predicted a 24% time saving before the trial and still estimated a 20% saving afterward. METR reported that participants accepted fewer than 44% of AI code generations; 75% said they read every line of generated code, and 56% said they often had to make major changes. In a subset of 44 tasks with usable screen recordings, about 9% of AI-allowed working time went to reviewing and cleaning output, with another 4% spent waiting for the AI.

At a glance
reportWhen: Published November 17, 2026; the resear…
The developmentA review of research cited by Georgia Tech cognitive engineer Karen Feigh highlights how checking AI output can offset speed gains, with results differing across tasks and studies.

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 Changes the Productivity Calculation

These findings matter because a fast first draft is not the same as a faster completed task. If a worker must inspect, test, correct or replace AI output, those steps belong in the productivity measure. Leaving them out can make a tool appear more effective than it is across the full workflow.

Review is also tied to responsibility. In Feigh’s account, people remain accountable for work even when AI contributes to it. That can put workers in the position of checking confident-sounding output without clear signals about which parts are wrong. The practical question for employers is not only whether AI produces material quickly, but whether staff have enough time and skill to verify it.

The research does not establish that AI always slows work. It indicates that outcomes depend on the task and on how well AI can perform it. A useful evaluation should track completion time and quality together, including errors and the labor required to find and fix them.

Products Worth Considering

Bose QuietComfort Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, Eucalyptus Green

Bose QuietComfort Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, Eucalyptus Green

As an affiliate, we earn on qualifying purchases.

Results Depend on the Task

Other studies cited by Feigh show why the METR result should not be generalized to every occupation. In a study of mid-level professional writing tasks, ChatGPT users completed work 40% faster and received quality scores 18% higher than participants who did not use the tool. Those results apply to the writing tasks and study conditions, not to all workplace use.

A preregistered experiment with 758 Boston Consulting Group knowledge workers likewise found different results depending on task difficulty for AI. On 18 tasks within the system’s capabilities, participants using AI completed 12.2% more tasks and worked 25.1% faster on average, while also producing higher-quality solutions. On a complex task outside those capabilities, AI users were 19% less likely to produce a correct solution than participants without AI.

There is also evidence that people may not reliably catch consequential errors. In an experiment involving 2,784 people checking AI-style suggestions, participants usually caught typos, but their accuracy was 31% for errors requiring an understanding of task rules. This is a result from that specific experiment; it does not establish a general error-detection rate for workers using AI in other settings.

“AI’s productivity gains are often countered by time and mental energy spent determining whether the output is accurate and appropriate.”

— Karen Feigh, Georgia Tech cognitive engineer, in The Conversation

How Broadly the Findings Apply

The studies do not provide a single, comparable measure of the time workers spend checking AI across occupations. They differ in participants, tools, task design and definitions of speed and quality. The METR trial involved 16 developers and particular coding tools; its result cannot establish how current tools affect every software team or other jobs.

It is also unclear how much verification time is routine in workplaces beyond the tasks studied, or how results change as tools and work practices change. The reviewed evidence does not show that every user checks every output, nor does it identify one review method that reliably catches all kinds of error.

Measure the Full Work Cycle

The research points employers and evaluators toward tracking more than how quickly AI produces a draft. They can compare total task time, output quality, error rates and correction work between AI-assisted and unaided workflows, while recording the task and tools used. Results should be reported with their baselines and study conditions so readers can see what the numbers do—and do not—show.

For workers who remain responsible for AI-assisted output, the findings also raise practical questions about review time and training. The cited studies do not prescribe a universal standard. Further evidence across more jobs and real workplace settings will be needed to establish when AI saves time after checking is included and when it shifts work into verification.

Key Questions

Did AI make the developers in the METR study faster?

No. In that randomized trial, 16 experienced developers took 19% longer on average when AI tools were allowed. The finding concerns the study’s participants, tasks and tools, not all developers.

How much time did developers spend reviewing AI code?

In a subset of 44 tasks with usable recordings, METR estimated that about 9% of AI-allowed working time went to reviewing and cleaning AI output. The study also reported that 75% of developers said they read every line of generated code.

Do the studies show AI always reduces productivity?

No. Writing and consulting experiments found faster completion on some tasks. The results varied by task, and the consulting experiment found worse accuracy on a task outside AI’s capabilities.

What does “verification burden” mean?

Feigh uses the term for the time and mental effort spent checking whether AI output is accurate and appropriate. It can include reviewing, testing and correcting generated work.

What should organizations measure when evaluating AI?

They should measure the full workflow, including completion time, quality, errors and review or correction work, rather than treating the speed of an initial AI draft as the productivity result.

Source: ThorstenMeyerAI.com

You May Also Like

David Mustapha: 5 Ways AI Is Transforming E-Commerce In Nigeria

David Mustapha outlines five key ways AI is reshaping e-commerce in Nigeria, boosting efficiency, customer experience, and market growth.

Can AI Improve Student Organization Efficiency? Top 6 Tools For 2026

Thorsten Meyer AI’s 2026 roundup ranks six AI student organization resources, led by The Complete Notion AI Guide and two Google Notebook LM manuals.

The Labor Displacement Data: What Q1-Q2 2026 Actually Shows

Initial data from Q1-Q2 2026 shows significant AI-driven job cuts in tech, with specific cohorts most affected. Displacement is concentrated, not widespread.

Google DeepMind Eyes Self-Improving AI In 1 Year [2026]

Google DeepMind aims to develop self-improving AI systems within a year, marking a significant milestone in artificial intelligence research.