In partnership with

Every Tuesday - Deep dives, architecture lessons, and real engineering stories.

Every Saturday - High Value - the best DevOps, SRE, Cloud, AI, tools, tutorials, and projects from the week.

AI is the new search engine. Start showing up.

Your buyers still search. They just do it in ChatGPT, Claude, Gemini and Perplexity now, and they act on the answer they get.

Those answers have rankings too. A few companies get named. The rest don't show up.

ChatFeatured is the AEO platform that does the work to get you named:

  • See where you rank against competitors on every major AI engine.

  • Find the questions where AI names them and not you.

  • Get the content that changes the answer written and published for you.

WikiSleep grew brand mentions 113% while competing head-on with Calm and Headspace. 

If you already invest in search, this is the next place your buyers look.

Welcome to this week's Uptime Sync. We start with Revolut's write-up of its 20 October outage. A Google Cloud zone in London lost power, and the servers that decide which database replica becomes primary couldn't reach a majority, because one of them had already gone missing without an alert. It took about 11 hours for every service to come back. Then OpenAI explains how it rebuilt Git for its monorepo, cutting median push latency nearly 5x, and how Codex agents in its Performance Factory ran more than 1,700 slowdown investigations in about three months. We also cover how Uber rate limits hundreds of millions of requests a second, why Cloudflare picked QuePaxa, a consensus algorithm with no leader, how LinkedIn runs Kubernetes on more than 500,000 bare-metal servers, how Figma manages tens of thousands of sandboxes, how Zerodha built fast blob storage on SQLite, and a benchmark showing DigitalOcean's new "Advanced" Postgres is slower than the plan it replaces.

On the tutorial front: compare ufw, cloud firewalls and Tailscale for keeping a Droplet off the public internet, and limit SSH agent forwarding with ProxyJump and filtered agents. We also cover upgrading PostgreSQL from 16 to 17 with the Percona Operator, promoting Kubeflow models through a gated pull request with Argo CD, scriptable GeoDNS answers on Bunny DNS, and a two-node anycast PowerDNS cluster announced over BGP.

This week's projects include Pgboxdb for giving each Git branch or agent its own test database, and Wrkflw for running GitHub Actions workflows locally before you push. Devpush deploys Compose apps on your own server on every push, DNSControl previews DNS changes before they reach your provider, and Cost-xray shows which tool calls and prompts drive your AI agent's API bill.

Newsworthy Reads

Tutorials of the Week

Projects of the Week

  • Pgboxdb - Run 'npx pgboxdb@latest', copy your dev schema schema-only into a template, then create a separate branch database per agent or Git branch so parallel test runs stop sharing one database.

  • Wrkflw - Run wrkflw validate and then wrkflw run .github/workflows/ci.yml to catch workflow errors and test a CI change locally before pushing to the hosted runner.

  • Devpush - Run a Compose-based service on a spare Linux server and wire a GitHub repository so each push deploys with HTTPS, logs and backups, without adopting a hosted PaaS.

  • DNSControl - Run the Docker preview command against a dnsconfig.js for a test zone to see which provider changes DNSControl would make before pushing anything.

  • Cost-xray - Run the installer, use your agent for a session, then open cx to see which tool calls and cached prompt blocks drive API spend and remove MCP servers that are injected but never called.

👋 Find me on Twitter | Linkedin | Instagram | Connect 1:1

Thank you for supporting this newsletter. Consider sharing this post with your friends.
Y’all are the best.

Reply

Avatar

or to participate