AgenticSRE is an independent journal about AI and the work of keeping systems reliable. Expect useful questions, close reading, and experiments whose evidence you can inspect.

The questions behind the screen

I am interested in the work between an alert and a decision: finding the relevant change, checking a promising explanation, deciding whether to act, and leaving enough context for the next person. AI makes that territory worth exploring. It also gives us new things to get wrong.

An incident summary can be readable and still omit the observation that changes the next step. A suggested fix can be technically possible while exceeding the operator's authority. A tool can save minutes generating an answer and consume them again in review. Those are the details I want this publication to stay with.

This is for engineers exploring how AI changes operations, and for practicing SREs and platform engineers who want something they can question, adapt, or test. You do not need to have chosen a model, bought an agent platform, or accepted anyone's prediction about the future of the job.

Explore the observatory

The Journal holds the longer arguments. Start with Before you give an agent the keys to production, a worked fictional incident about evidence, approval, and the boundary between a proposal and an action.

Signal collects original sources with a reason to read each one. The aim is to connect a paper, engineering article, or documentation page to a concrete operational question. A source appearing here means it is worth examining; it does not make every claim in it settled.

The Lab lets you inspect fictional evidence and review a scripted assistant's recommendation. It also holds proposed experiments with methods and success criteria. The Observatory connects the recurring questions: permission, evidence, operator judgment, and evaluation.

You can enter through whichever door is useful. The atmosphere is part of the invitation; the writing should earn the rest of your attention.

Who is Rook?

Rook is my pseudonymous author identity, represented by the investigator-mechanic in this world. Relay is the machine familiar beside Rook: a companion character and a way to navigate. Neither character is presented as an autonomous system operating real infrastructure.

This is a personal editorial project. Opinions will have an author; technical claims should have inspectable support. When an argument changes, I want the reason for changing it to be visible.

What is ready, and what is still open

This welcome and the production-permission essay are the first published pieces. Other essays are marked Editorial draft while they develop. Lab exercises are fictional, and the experiment plans are proposals, not reports of completed evaluations. There are no measured agent-performance results to claim yet.

Follow the RSS feed for published essays and reading editions. Drafts stay out of that feed. There is no promised publishing cadence yet; I would rather add a useful piece when it is ready. The next question is already on the desk: what evidence would make an assistant's recommendation worth trusting?