The best thing on the Internet right now. youtu.be/3TNpOD6bov8?...

LGTM (Looks Good to Me) - Claude Opus 5.5 music video

YouTube video by Degenerative Pixels

youtu.be

I have never heard anyone talk about this topic, even at #monktoberfest, but it is hitting home: #tallpoppysyndrome in tech, by @brookejamieson.bsky.social

There's an endpoint that scores whether you can get the Spot VMs you want, 0 to 1. Most teams have never called it. But it's advice, not a reservation. I watched a zone fall 0.90 to 0.10 in 4 hours. Part 2 of 9: caseywest.com/obtainabilit...

Obtainability, Part 2: The API That Tells You What You Can Get

There's a beta Compute Engine endpoint that answers "can I get twenty Spot VMs of a given machine type in this region right now?" with a number between zero and one, and most teams on Spot have never ...

caseywest.com

Spot teaches you to survive losing a machine. It doesn't teach you what to do about the one you never get. My advisor said 0.90. Three regions gave me zero L4s for 36 hours. Part 1 of 9 on the half of scarcity nobody built for: caseywest.com/obtainabilit...

Obtainability, Part 1: The Other Half of Scarcity

You went to Spot VMs for the discount and did the homework: checkpoint code, a drain handler, a preemption survived. Then a job sat in Pending for days across three regions and never got a single L4. ...

caseywest.com

My AI-generated code percentage is embarrassingly high, and it's the wrong number to chase. It's this era's lines-of-code: pulled from a diff, easy to game, silent on whether anyone can explain the system in six months. caseywest.com/the-code-generation-percentage-your-org-shouldnt-be-measuring

The Code-Generation Percentage Your Org Shouldn't Be Measuring

"Percent of code AI-generated" is this era's lines-of-code: easy to compute from a diff, trivially gamed, and silent on whether anyone can explain the system in six months. Here is what the field data...

caseywest.com

Three Kubernetes projects added brakes to their own elasticity. Kueue froze a quota field. LeaderWorkerSet limited its autoscaler. Agent Sandbox slowed its refill loop. Same window, same kind of fix. I think I know why. caseywest.com/three-kubernetes-projects-shipped-a-brake-pedal-this-month

Three Kubernetes Projects Shipped a Brake Pedal This Month. Why?

In one week of late-August 2026, Agent Sandbox, Kueue, and LeaderWorkerSet each shipped a limiter on their own elasticity: a warm pool that refills more slowly, a field that used to be editable and is...

caseywest.com

Your OKF bundle passes validation. It can still be wrong: a new source contradicts a page from last year, and the spec never notices. A timestamp records a fact, it doesn't run a check. So I built the system to manage your LLM-Wiki. okfctl: caseywest.com/a-field-is-n...

A Field Is Not a Process

A timestamp records when a wiki page was written. Nothing in the format runs to keep it true. Twenty-one years after I helped build a commercial wiki, the LLM-wiki idea is back, and it dropped the one...

caseywest.com

When the accelerator you want is gone, don't just retry. Rank the ones you'd accept, fall to the next, and climb back when the best returns. That's a fallback ladder—and on GKE, one ComputeClass object. https://caseywest.com/compute-scarcity-is-permanent-build-a-ladder

Idle AI agents don't burn cycles. They hoard RAM. Google's GKE Agent Sandbox benchmark: the isolation lever is worth 44%. Suspending idle agents and reclaiming their memory is worth better than 3x on its own. The sandbox is the smaller half. https://caseywest.com/your-agents-are-too-awake

I usually work with engineers. Last Wednesday I got to open the first stop of Build with Gemini, in NYC, with the business track. Finance, HR, ops. No code, and they still shipped an agent that waits on a human before spending money. cloud.google.com/events/build...

A Pending pod that "should" fit—free chips everywhere, none adjacent. I built that into a browser puzzle: real H100, A100, TPU slices, five levels, no signup. Running clusters like this and explaining them is the job I'm hiring for. caseywest.com/schedule-the...

Schedule the Accelerators: A GPU/TPU Puzzle

I'm hiring a Senior Developer Relations Engineer for GKE and AI Infrastructure. Instead of describing the work, I built it: a small, playable cluster scheduler that teaches the real problem behind run...

caseywest.com

A team asks for Claude + web search in prod, this quarter. The architect reads three cells—tier, toggle, contract—and says yes before lunch. A lookup, not a risk assessment. Series finale. New? Start at Part 1: caseywest.com/governed-gro...

Governed Growth, Part 4: Multi-Model, Governed, and Boring

The finale of the series: the mature end-state the first three parts build toward. When every server-side capability's trust boundary is a decision on record—the deployment tier you chose, the toggle ...

caseywest.com

No single off switch for "every server-side tool." 5 controls across 4 layers—partner web search denied by default. Part 3 is the runbook I've shipped: which toggle, at which scope, in gcloud + Terraform. Finale next. https://caseywest.com/governed-growth-part-3-default-deny-for-model-capabilities

👀 Sneak peek behind the scenes of Google Agent Skills! In my new article, I share how we build, test, and scale AI skills at Google: • Standardized layout • Remote MCP tools for IAM • Automated CI/CD checks • 2x2 evaluation matrix Full article link in replies 👇 #GoogleCloud #AIAgents #AgentSkills

The thirty-day retention everyone quotes for Grounding with Google Search went stale last February. It's up to 3 days of debug logs now. The old thirty days moved to Maps grounding—same provider, opposite posture: caseywest.com/governed-gro...

Governed Growth, Part 2: The Retention Window That Quietly Shrank

On February 23, 2026, Google quietly made the retention terms for Grounding with Google Search better: the window dropped from thirty days to up to three, and what's kept narrowed from your prompts an...

caseywest.com

Your agent can't really search the web. It guesses from training, or scrapes a snippet. gemini-search-mcp gives it one tool: a real Google search, run through Gemini, answered with the sources it used. One install, one static binary: https://github.com/cwest/gemini-search-mcp

Picked the small Gemini model for my search tool. Two newer models shipped, so I re-ran the eval—four models, one run. The small one still wins. The biggest came last on everything, at ~140x the cost. caseywest.com/building-gem...

Building gemini-search-mcp: Grounding, Citations, and Choosing a Gemini Model

The builder's companion to Governed Growth—the same reference repo from the implementer's chair. Why I built a one-tool MCP server to hand an assistant a reliable, governed Google search, how Gemini g...

caseywest.com

"We approved the model" and "we approved the model's web search" are two different decisions. Most platforms treat them as one—so one line in a PR opens a data path nobody chose. Part 1 of a 5-piece run. Follow along: caseywest.com/governed-gro...

Governed Growth, Part 1: Better Together, Governed Apart

Opening a platform to more teams and more models—first-party and partner, such as Gemini and Claude—quietly assumes those capabilities share one trust boundary. They don't. Multi-model expansion is a ...

caseywest.com

Casey analyzes and deconstructs the viral case of a coffee shop that was run by AI agents as an experiment. How can we avoid similar mistakes?

Casey West@caseywest.bsky.social · 4mo ago

They swapped the model mid-experiment. The café failed the same way. So the model wasn't the problem. It shipped with no spend cap and no approval gate—the seatbelts you install on day one. The proof and the five fixes: www.linkedin.com/pulse/ai-did...

They swapped the model mid-experiment. The café failed the same way. So the model wasn't the problem. It shipped with no spend cap and no approval gate—the seatbelts you install on day one. The proof and the five fixes: www.linkedin.com/pulse/ai-did...

The AI Didn't Fail. The Deployment Did.

The Infographics Show, one of the biggest explainer channels on YouTube with an audience in the millions, just put out a video titled "Google Gemini Ran a Real Business for a Month and It Almost Bankr...

linkedin.com

My `gh` calls started failing: rate limited. Weird. I'd barely made any. Turns out GitHub meters REST and GraphQL on two separate budgets, and agents drain GraphQL while REST sits full. Route across both, double your headroom. caseywest.com/github-gives...

GitHub Gives You Two Rate Limits. You're Only Using One.

GitHub's REST and GraphQL APIs have separate hourly rate limits—two budgets, not one. Most automation hammers a single bucket while the other sits full. Here's how to see both, why agents drain GraphQ...

caseywest.com

Claude Fable 5 shipped today. By lunch it was running in my own GCP project, on data terms I'd actually read. One undocumented 403 stands in the way. Walkthrough + trade-offs, stated plainly: www.linkedin.com/pulse/claude...

Claude Fable 5 dropped today. Here's how to try it without gambling your company's data.

Anthropic released Claude Fable 5 this morning. By this afternoon it was the default model in my tooling, running inside my own Google Cloud project, under data terms I'd actually read.

linkedin.com

Presenting The Agentic Manifesto: a new model for the era of autonomy. The old SDLC can't govern emergent AI. We need a new playbook. We call it the 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗗𝗲𝗹𝗶𝘃𝗲𝗿𝘆 𝗟𝗶𝗳𝗲𝗰𝘆𝗰𝗹𝗲 (ADLC). Read the manifesto: https://agenticmanifesto.ai

I analyzed a year of peer "thank you" notes, expecting fluff. I was completely wrong. Over 90% weren't just compliments; they were solid, admissible evidence of high performance. Next time you thank a colleague, don't just be nice. Be specific. Give them evidence they can use for their career.

My `archon-gemini-cli-extension` is out! 🚀 It turns 𝗚𝗲𝗺𝗶𝗻𝗶 𝗖𝗟𝗜 into 𝗮 𝘀𝘁𝗮𝘁𝗲𝗳𝘂𝗹 𝗽𝗿𝗼𝗷𝗲𝗰𝘁 𝗺𝗮𝗻𝗮𝗴𝗲𝗿 𝘃𝗶𝗮 𝘆𝗼𝘂𝗿 𝗔𝗿𝗰𝗵𝗼𝗻 𝘀𝗲𝗿𝘃𝗲𝗿. Use it to: → 𝗖𝗼-𝗮𝘂𝘁𝗵𝗼𝗿 𝗣𝗥𝗗𝘀 → 𝗘𝘅𝗲𝗰𝘂𝘁𝗲 𝗱𝗲𝘃 𝗽𝗹𝗮𝗻𝘀 → 𝗠𝗮𝗻𝗮𝗴𝗲 𝗽𝗿𝗼𝗷𝗲𝗰𝘁 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 #AIdev github.com/cwest/archon...

GitHub - cwest/archon-gemini-cli-extension: An Archon extension for the Gemini CLI

An Archon extension for the Gemini CLI. Contribute to cwest/archon-gemini-cli-extension development by creating an account on GitHub.

github.com

@kweinmeister.bsky.social's 𝗽𝗿𝗶𝗺𝗲𝗿 𝗼𝗻 𝗽𝗼𝗹𝘆𝗴𝗹𝗼𝘁 𝗮𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲𝘀 𝗶𝘀 𝗿𝗲𝗾𝘂𝗶𝗿𝗲𝗱 𝗿𝗲𝗮𝗱𝗶𝗻𝗴. 𝗛𝗲 𝘁𝗲𝗲𝘀 𝘂𝗽 𝗺𝘆 𝗽𝗼𝘀𝘁 𝗳𝗼𝗿 𝗮 𝗥𝘂𝘀𝘁 𝗶𝗺𝗽𝗹𝗲𝗺𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻 𝘁𝗼 𝘀𝗼𝗹𝘃𝗲 𝗣𝘆𝘁𝗵𝗼𝗻'𝘀 𝗰𝗿𝗶𝘁𝗶𝗰𝗮𝗹 𝗖𝗣𝗨 𝗯𝗼𝘁𝘁𝗹𝗲𝗻𝗲𝗰𝗸𝘀. ⚡ #𝗥𝘂𝘀𝘁 medium.com/google-cloud...

On the road, folks ask where to learn AI. Here's my answer: Google Skills, your all-in-one platform for AI learning and so much more. New? Get 150 free credits!💪 #AI cloud.google.com/resources/of...

Google Skills: Skill Up in AI

Learn more about Google Skills, receive learning credits, and more.

cloud.google.com

I wrote this a decade ago to help managers build strong, independent teams. It's called 𝗧𝗵𝗲 𝗛𝗲𝗹𝗽 𝗧𝗶𝗺𝗲𝗿: a technique I’ve used with a lot of teams to help them balance independence with collaboration. A tool for mentorship, not just efficiency. 🧠⚙️ #EngLead caseywest.com/the-help-timer

The Help Timer

Wait an hour to ask for help. No more, no less.

caseywest.com

𝑁𝑒𝑤 𝐴𝐼 𝑎𝑔𝑒𝑛𝑡 𝑓𝑟𝑎𝑚𝑒𝑤𝑜𝑟𝑘 𝗔𝗿𝗰𝗵𝗼𝗻 is 𝘳𝘢𝘥! 💥 Just had its first release. I've been trying it with 𝗚𝗲𝗺𝗶𝗻𝗶 𝗖𝗟𝗜 for 𝗚𝗲𝗺𝗶𝗻𝗶 𝟮.𝟱 models, local dev, & smart RAG. Impressive and ready for you to explore! #AI #AgenticAI #Gemini github.com/coleam00/Arc...

GitHub - coleam00/Archon: Beta release of Archon OS - the knowledge and task management backbone for AI coding assistants.

Beta release of Archon OS - the knowledge and task management backbone for AI coding assistants. - coleam00/Archon

github.com

Come to my half day, hands on workshop "From Prompt to Action: Building Production-Ready AI Agents" in Istanbul on November 7th! Sign up to leave with the skills to build, deploy, and scale dynamic, production-grade AI solutions. rsvp.withgoogle.com/events/devel...