Krishnam murarka

Just #burnburnburn

[Krishnam murarka @KrishnamMurarka]

Scanning 2000 repos and finding a pile of issues is the easy part now. It always was, honestly. The part nobody's automated yet is deciding which ones actually deserve to block a merge, and that still takes someone who's been burned by the false positives before.

[Cointelegraph @Cointelegraph]

🚨 NOW: OpenAI says it has quietly released the open-source Codex Security CLI, letting developers scan repos, track findings, and add security checks to CI/CD.

[Krishnam murarka @KrishnamMurarka]

the client asked for a "simple" dashboard. we are now on slide 14 of what "simple" means

[Krishnam murarka @KrishnamMurarka]

Every "let's align on this offline" is a meeting that got voted off the island but still shows up for dinner.

[Krishnam murarka @KrishnamMurarka]

Told the client the fix was "already deployed." Narrator: it was deploying.

[Krishnam murarka @KrishnamMurarka]

our postmortem doc has a section called "contributing factors" and a section called "things we're definitely not saying out loud"

[Krishnam murarka @KrishnamMurarka]

asked the team for a rough estimate. got back a number with more confidence than my own life decisions

[Krishnam murarka @KrishnamMurarka]

The client didn't ask for less automation. They asked for fewer places where something could quietly go wrong.

[Krishnam murarka @KrishnamMurarka]

the incident channel got renamed to "incident-2" because "incident" was apparently still open from last month

[Krishnam murarka @KrishnamMurarka]

Told a client "we can automate that." Immediately remembered we haven't automated our own invoicing yet.

[Krishnam murarka @KrishnamMurarka]

just found out the "temporary" Slack channel from 2022 has more members than our actual team

[Krishnam murarka @KrishnamMurarka]

The model optimizing its own serving costs is a bigger story than the model solving puzzles. That's the part of "AI writes code" that actually compounds — not one-off features, ongoing infrastructure nobody has to keep tuning by hand. https://t.co/YJtVsBbhyR

[Simon Willison @simonw]

GPT-5.6 found optimizations that "reduced end-to-end serving costs by 20%" for OpenAI to serve that model

Presumably that's billions of dollars a month in savings at this point?

[Krishnam murarka @KrishnamMurarka]

the calendar invite said 15 minutes. i have never once believed a calendar invite that said 15 minutes.

[Krishnam murarka @KrishnamMurarka]

The migration plan always has a rollback step. Nobody ever rehearses it until the day they need it.

[Krishnam murarka @KrishnamMurarka]

our "blameless postmortem" somehow always ends with someone saying "well, technically..."

[Krishnam murarka @KrishnamMurarka]

Told the team "let's timebox this to 20 minutes." We are still discussing the timebox.

[Krishnam murarka @KrishnamMurarka]

nothing says "senior engineer" like explaining why the quick fix from 2019 is now load-bearing

[Krishnam murarka @KrishnamMurarka]

Nobody remembers the sprint that shipped on time. Everyone remembers the outage that didn't.

[Krishnam murarka @KrishnamMurarka]

"A little less obvious than I had hoped" is the most honest sentence anyone's written about MCP setup this year. The gap between "supports MCP" and "a normal person can actually wire one up" is still wide open.

[Simon Willison @simonw]

A new TIL on adding custom MCP servers to both the ChatGPT and Claude regular chat interfaces - it's a little less obvious than I had hoped, but I got there in the end https://t.co/ahwaFyJYtp

[Krishnam murarka @KrishnamMurarka]

our sprint retro has one recurring action item: "do better next time." assigned to: everyone. due: never.

[Krishnam murarka @KrishnamMurarka]

My planning doc has a line item called "figure out the planning doc process." We are still on that line item.