“LLM-based code is often verbose, and rarely aligns with in-house coding guidelines and standards, guaranteeing that it’ll take far longer to chew through, which naturally means that those burdened with reviewing it will either skim-read it or feed it into another LLM to work out what the hell to do.”
Our reading:
LLM-based code is often verbose and rarely aligns with in-house coding guidelines and standards, guaranteeing it takes far longer to review, which means reviewers will either skim-read it or feed it into another LLM.
Assessment: Unresolved
What happened
Does the evidence support the cause?
Outcome: Unresolved
We checked, and the evidence so far doesn’t settle it.
Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.
Evidence up to . Checked .
Why
Meta's study, published July 17, 2025, reported that an initial trial of AI patch suggestions increased review time by over 5% [e1]. After suggestions were shown only to authors, a second trial found no statistically significant review-time regression [e1]. Sonar's October 2025 survey found that 38% of developers reported greater review effort for AI-generated code than human-written code, while 27% reported less [e2]. The trial and survey show different review outcomes, but do not test the full proposed chain from verbosity and local-standard failures to longer reviews and then skimming or LLM delegation [e1] [e2].
Caveats
- Meta studied targeted patches in 2024, not representative generated implementations as of March 2026.
- A statistically nonsignificant result does not establish equivalent review times.
- Sonar measured self-reported effort, not matched review duration or observed reviewer behavior.
- The packet lacks representative measurements of verbosity and compliance with actual in-house standards.
- Neither item establishes whether those code properties cause reviewers to skim or delegate review.
Evidence (2 items)
Context only: arXiv; Meta researchers,
regressions
Why it matters: Meta's initial experiment found longer reviews; a second trial ending December 18, 2024 found no statistically significant regression after changing how suggested patches appeared. This challenges inevitability, but concerns targeted patches and an earlier workflow rather than the full March 2026 claim.
- Publisher
- arXiv; Meta researchers
- Kind of source
- Preprint, not peer reviewed
- About events on
- Published
Context only: Sonar,
less effort
Why it matters: The survey reports greater review effort for 38% of developers and less effort for 27%. It supports review burdens for a substantial group while showing variation; self-reported effort does not establish review duration or the claimed causal mechanism.
- Publisher
- Sonar
- Kind of source
- Independent measurement
- About events on
- Published
How we judged it
- Kind of claim
- Causal claim
- Deadline
- None given. We don’t add one.
Where we had to interpret
- The plain reading is a broad causal claim about professional review of LLM-generated code as of publication.
- 'Often' requires recurring prevalence, and 'rarely aligns' means compliance is uncommon. Neither supplies a precise frequency.
- 'Far longer' concerns materially greater review time for comparable work, not simply more total review work because more code is produced.
- The proposed causal chain must connect verbosity and standards failures to review time and then to skimming or LLM delegation.
- A small study of targeted patches can test inevitability in that setting, but cannot establish the prevalence of problems across contemporary generated code.
Right if: Representative comparisons must find frequent verbosity, uncommon compliance with actual local standards and substantially longer review for comparable changes, with evidence that these problems cause reviewers to skim or delegate review.
Wrong if: Representative comparisons showing ordinary compliance and no substantial review penalty would contradict the general claim. Comparable observed workflows preserving thorough human review would rebut the claimed inevitable downstream behavior.
We wrote this test after the outcome was known.
What would settle it
Representative workplace comparisons that match tasks, languages, change sizes, models and reviewer experience. They must separately measure verbosity, local-standard compliance, active review time and actual reviewer behavior, and test the proposed causal links. Comparable workflows preserving thorough human review would test the claimed inevitable downstream behavior.
We found 35 claims in this post. See them all.
Full record
Source and dates
Where's Your Ed At, The AI Industry Is Lying To You, paragraph 127.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-28. The post may have been edited after 2026-03-24.
Where it was said (1)
Said once in the posts we read.
: Where's Your Ed At, The AI Industry Is Lying To You
LLM-based code is often verbose, and rarely aligns with in-house coding guidelines and standards, guaranteeing that it’ll take far longer to chew through, which naturally means that those burdened with reviewing it will either skim-read it or feed it into another LLM to work out what the hell to do.
- Where
- paragraph 127
- Published
- 2026-03-24
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Unresolved. Evidence up to 2026-10-28.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.
Held backBy us
The final check failed: a cited source didn’t say what the write-up said.
Held backBy us
The final check failed: a cited source didn’t say what the write-up said.