Skip to content

Topic

AI-generated code

Source code written or modified by AI coding assistants and autonomous agents, now a growing share of commits merged into production software.

Current stories

securityOne report1 publisher

Security leaders polled for ArmorCode say their trouble starts after a scanner flags a flaw

Just over half of 200 senior leaders polled for ArmorCode say they will struggle to simplify software security if they keep working the way they do. AI is adding to the pile from both ends, since developers write code faster and scanners find more weaknesses.

Reality

Evidence40
Adoption
Not scored
Hype gap+15
Incentives70
Confidence45
buildConfirmed4 publishers

Meta's agent plan produced 220 percent more code changes and 36 percent more shipped work

Project OT would have cut some teams by 60 percent. Meta's own numbers showed incidents up 40 percent and the time spent on them up 70, and Zuckerberg killed the second wave.

Perspective Coverage

4 publishers
Builder
Builder 27%
Operator
Operator 49%
Investor
Investor 24%

Reality

Evidence60
Adoption45
Hype gap+25
Incentives50
Confidence58
productOne report1 publisher

Vicarius puts a human reviewer and a deterministic judge between its AI-written fixes and production

ScriptAI drafts detection and remediation logic for software flaws that have no vendor patch behind them. Vicarius says the draft, review and verify cycle finishes in under an hour, against weeks by hand. It is live for vRx customers today.

Publishers:siliconangle.com

Reality

Evidence32
Adoption12
Hype gap+35
Incentives80
Confidence40
buildOne report1 publisher

An unquoted glob in a generated pack.sh makes LC_COLLATE part of the build

A dev.to walkthrough traces a checksum failure that only appears in CI to the shell expanding fragments/*.json in whatever collation order the job inherited, and proposes committing the file list so the order ships with the script.

Publishers:dev.to

Reality

Evidence52
Adoption
Not scored
Hype gap0
Incentives22
Confidence60
buildOne report1 publisher

Seal's only pre-release design review came from the agent that wrote much of its code

Jason Page has submitted his encrypted inheritance app to Apple with public source, a default release schedule of 90 days of silence plus 35 cancellable days, and a published list of what has not been tested.

Publishers:runtimewire.com

Reality

Evidence55
Adoption6
Hype gap−20
Incentives55
Confidence58
buildOne report1 publisher

A sleeping laptop expired a 30-second HTTP budget by half an hour

A queue worker built its remaining-seconds budget from time.time(), and after the host woke from sleep it logged remaining=-1842.7. Two days of debugging went to sockets, Docker and NTP before the clocks got compared.

Publishers:dev.to

Reality

Evidence58
Adoption
Not scored
Hype gap0
Incentives60
Confidence55
buildOne report1 publisher

Review time on Salesforce's largest pull requests plateaued and then declined

Salesforce attributes the flattening to overload eroding its second-pair-of-eyes guarantee, while a dev.to analysis argues part of it is reviewers declining to spend time on agent pull requests they can already tell will never merge.

Publishers:dev.to

Reality

Evidence38
Adoption30
Hype gap+12
Incentives52
Confidence42
buildOne report1 publisher

curl's confirmed-vulnerability rate fell from above 15 percent to under 5 in 2025

The case that generated code fails without the usual warning signs comes from two practitioner accounts, while the only measured series is curl's inbound bug reports, and Daniel Stenberg stopped taking those in January 2026.

Publishers:dev.to

Reality

Evidence38
Adoption58
Hype gap+34
Incentives62
Confidence46
buildOne report1 publisher

Watching a backup client from outside its own logs catches the variant that saved nothing

Before judging agent-written code, a dev.to author seeded a small C backup client with eight known behaviors and observed it with system-call traces, a loopback capture and a digest check to see which observer caught what.

Publishers:dev.to

Reality

Evidence48
Adoption
Not scored
Hype gap−12
Incentives20
Confidence55
buildOne report1 publisher

Abnormal AI runs agent-driven code in an ephemeral MicroVM on the inline email path

AWS's post on the deployment gives the sandbox defaults that decide whether it fits an inline path, among them a 15-minute session time-to-live configurable to eight hours and a choice between VPC-only and public internet access.

Publishers:aws.amazon.com

Reality

Evidence46
Adoption62
Hype gap+25
Incentives82
Confidence52
productOne report1 publisher

Testing teams adopted AI for writing tests 3.5 times as often as for judging risk

PractiTest puts AI use across testing organizations at 76.8%, with test creation at 69.6% and risk identification at 19.9%. Sonar has 42% of committed code AI-assisted and heading to 65%, all of it flowing toward the same reviewers.

Publishers:devops.com

Reality

Evidence57
Adoption64
Hype gap+13
Incentives72
Confidence55
leadershipOne report1 publisher

Evil Martians makes the API contract the artifact humans review in agent-written frontends

The consultancy's ten-check list treats generated code as something to prevent and detect rather than read line by line, which trades review labour for a dependency on whoever keeps the backend's OpenAPI spec current.

Publishers:evilmartians.com

Reality

Evidence46
Adoption20
Hype gap+14
Incentives58
Confidence54
buildOne report1 publisher

Path rules sort 50 of 93 changed files to the bottom of the review

A team that interviewed more than 100 tech leads about AI-generated pull requests shipped per-viewer criticality tiers. Reviewers like the ordering and report awkwardness at declaring which files they skip.

Publishers:dev.to

Reality

Evidence28
Adoption20
Hype gap+12
Incentives85
Confidence50
buildOne report1 publisher

Running detekt instead of detektMain leaves every type-resolution rule silently inert

A dev.to post argues that assistant-written Kotlin breaks between files rather than inside them. Its own worked example, a nine-dependency ViewModel with an empty catch block, is already covered by detekt defaults nobody has to enable.

Publishers:dev.to

Reality

Evidence55
Adoption
Not scored
Hype gap+18
Incentives
Not scored
Confidence57
leadershipOne report1 publisher

A Meta chatbot changed the email on any account for whoever asked

Authorization was absent rather than broken, which is the defect class a reviewer catches and a test suite signs off on. The newsletter reporting it also finds individual output up and team output flat.

Reality

Evidence34
Adoption41
Hype gap+22
Incentives63
Confidence45
buildOne report1 publisher

Point at one line and ask the submitter to explain it

Code review has always been sampling, and the signals it sampled stood in for a mind that had read the code. A dev.to post argues that when diffs arrive without one, what needs rewriting is the process, not the effort.

Publishers:dev.to

Reality

Evidence32
Adoption
Not scored
Hype gap+14
Incentives22
Confidence48