Every item below shares a thread: ownership shifted, and the system built to track it didn’t move with it. A repo’s knowledge layer only stays legible when the file format matches who’s actually reading it, agents, not a rendering pipeline. A pull-request queue built to ration human attention on human-written diffs now waits 5.3x longer on code a machine wrote first. A team automated 75% of a role’s work and lost the person the moment nobody redefined what was left to own. Ownership doesn’t wait for the org chart, the file format, or the review process to catch up.

Here’s what’s worth your attention.

Worth your attention

Engineering

  • MD vs MDX: The Decision Most AI-Agent Repos Get Wrong: developer-written Markdown instructions improve agent task success by about 4% and cut agent-generated bugs 35 to 55%. Swap in MDX or LLM-generated instructions instead, and success drops while inference cost rises over 20%. The format isn’t cosmetic. It’s who the file is actually for.

Leadership

  • We Automated 75% of Reporting. Three People’s Jobs Changed Overnight.: “You didn’t just automate the process. You automated my job.” Three analysts’ weekly fire drill became a daily refresh. The two who got handed genuinely new ownership, of a churn model, of the stakeholder relationship, kept their engagement. The one left holding monitoring didn’t.

Strategy

  • Whoever Owns the Agent Owns the Repo: bot-authored pull requests merge at 32.7%, less than half the 84.4% human-authored rate, and wait 5.3x longer for a first review. GitHub still holds the code. It stopped holding the loop the moment an agent, not a human at a keyboard, became the one deciding what happens next.

One number worth sitting with

5.3x. That’s how much longer an AI-written pull request waits for a first human review than one a person opened themselves, per LinearB’s 2026 benchmark across 8.1 million pull requests. The code changed authorship. The queue built to ration attention on it didn’t. That gap is where ownership actually lives right now: not in whichever system holds the repo, but in whichever one is fast enough to keep up with who’s really doing the work.

Reply and tell me where a system in your stack is still tracking who used to do the work, not who’s doing it now. I read every response.