Digest #223: AWS bills a user $140B, a kubelet leak in Kubernetes 1.36, Slack's Shipyard EC2 platform and pods vs AI agents

Plus the four horsemen behind thousands of Postgres outages, how Uber solved the identity crisis for AI agents, and catching full table scans in SQLite before they catch you.

Someone opened their AWS console this week to a billing alert projecting $140 billion for the month, on an account that normally spends $2-3 on S3. Underneath the meme-worthy screenshot, the real theme was infrastructure hitting its limits: a kubelet memory leak in Kubernetes 1.36, four recurring failure patterns behind thousands of Postgres outages, and Slack rebuilding its entire EC2 fleet management from scratch. And with half this week’s tutorials about giving AI agents identity, memory, and APIs, the CNCF is already asking whether a pod is even the right place to run one.

AI is hitting your infra faster than you can govern it

AI is reshaping infrastructure, but the teams winning aren't adopting fastest, they're governing best. Spacelift surveyed 406 infrastructure and platform engineering leaders and found 93% have already had an AI-caused incident, while only 30% have a formal AI policy. The report reveals what separates the Pioneers, from platform engineering to the AI-specific signals most teams still aren't tracking.

Download the report to see what the most advanced teams do differently.

If you have feedback to share or are interested in sponsoring this newsletter, feel free to reach out via LinkedIn or simply reply to this email.

Newsworthy stories

• Just got an AWS billing alert projecting my monthly cost at $140B

• Fixing a kubelet memory leak in Kubernetes 1.36

• Inside the unpatched Argo CD RCE that leads to K8s cluster takeover

• Is a pod the right deployment unit for an AI agent?

• How we built Slack’s next-generation EC2 platform

• The four horsemen behind thousands of Postgres outages

• The database is not the data model

Tutorials of the week

• Just use Postgres for job queues

• Cloudflare as a data platform?

• How to encrypt Terraform state files

• Implement a unified memory from scratch via knowledge graphs

• Using Go for mobile apps

• Automating cross-repo documentation with GitHub Agentic Workflows

• Automated incident remediation with AWS DevOps Agent and Kiro CLI

• Detecting full table scans with SQLite

• Solving the identity crisis for AI agents

• Designing APIs for agents

Videos of the week

Projects of the week

• Portero shows every listening port on your machine, which process owns it, and lets you kill it or block the port from a built-in firewall.

• Greysight is an open-source dashboard for Snowflake cost observability that automatically suspends idle warehouses.

• Goshot generates polished code and terminal screenshots with syntax highlighting and automatic secret redaction.

• deja-vu is a memory index for coding agents that searches past session logs so they can recall previous solutions.

• pgsavvy is a terminal UI for PostgreSQL that works like “the lazygit for Postgres.”

• Khoj is a self-hostable AI second brain for semantic search and chat over your own documents.

• LobeHub is a platform for managing a team of AI agents as coordinated, schedulable units of work.

Meme of the week

I advise infra, devtools, and AI startups on AI product strategy, cloud security/FinOps, and tech M&A. If that's you, just reply to this email.