Pet recovery specialist assists in locating dogs missing after a car accidentIsraeli drone strike near Syria-Lebanon border injures sixSecret Service Director Describes Current Threat Environment as UnprecedentedIntercommunal clashes in South Sudan result in dozens of deaths as security forces retake RumbekUnclear Details Surround Ebola Patient’s Undetected Travel Across Three NationsKremlin says Trump welcomed Russia's role in Iran settlement talksVirginia police respond to house fire, rescue trapped man, bodycam footage releasedPresidential Message on Leif Erikson Day ReleasedCharlamagne Tha God discusses Netflix partnership, midterm elections, and mental health on Bloomberg podcastTrump proposes nuclear deal pathway for Iran amid continued military pressureLegal drama "Cupertino" premieres on CBSHeadteachers' unions consider strike ballots over concerns with Ofsted inspection regimeAustralian advocates call for subsidised weight‑loss drugs for women with PMOSMexico probes viral video alleging cartel-linked mercenaries in UkraineCiti predicts Treasury Secretary Bessent may cut long‑bond auction sizes next month
Politifex
Search
Back to event
Ars TechnicaCybersecurityOpen SourceComputingRegulationAI coding assistantssoftware development productivitycode reviewsoftware output

AI coding agents generate more code, but not more software

By Kyle Orland · Oct 9, 2026, 2:43 PM CDT

Read full article at Ars Technica
Anyone who has even tangentially associated with computer programming knows that modern AI coding assistants and agents can be incredibly efficient at generating huge amounts of functional code . But coders making use of those tools also know better than to trust the accuracy of that code , meaning substantial effort needs to be spent reviewing any AI-generated output. A recent study of actual cod

Excerpt shown under fair-use limits. Full text remains with the original publisher.

People in this coverage

Explore their history and attributable record. Being mentioned does not imply endorsement.

Kyle Orland →James Stratton →

Layer 1 · Claims & fact checks

AI analysis

Layer 2 · Biblical perspective

Biblical interpretation
INSUFFICIENT CONTEXT
Read the biblical analysis

Layer 3 · Reporting analysis

AI analysis

FramingThe article frames AI coding assistants as limited by human processes, emphasizing a negative outcome.

Appeal to AuthorityCiting Harvard researchers is used to lend credibility to the study’s conclusions.

Technical JargonSpecific metrics are presented to convey thoroughness, though the article does not explain how these metrics translate to the claimed conclusions.

Context

AI analysis

Missing context

The article does not provide details on the study’s methodology (e.g., how work events were classified, statistical methods used), the definition of "software output," or any control groups for comparison. It also lacks information on the types of AI coding agents evaluated, the industries represented, and potential confounding factors such as project complexity or team size.

Important context

Human code review is identified as a key bottleneck that offsets any speed gains from AI‑generated code, suggesting that the overall software production pipeline, not just coding speed, determines productivity gains.

Opinion vs. reporting

AI analysis

The piece primarily reports findings from the study but includes interpretive language (e.g., "little evidence that firms increase software output or reduce employment") that frames the results in a critical light toward AI coding tools.