An instructor at Next '26 shrugged and said Cloud Run is fine. He's mostly right. Here are the five things that actually justify reaching for Kubernetes instead, and why the two are closer to the same thing than the debate admits.
Dataplex's Knowledge Catalog turns data products into first-class, discoverable, requestable objects. A full field tour — finding, describing, sharing, and securing certified data products — plus the gotchas the docs skip.
I built a remote MCP server for my blog, then realized a bearer token was all I needed. Here's the honest trigger for when an MCP server is actually worth it, and how the build and auth really work.
For years I shipped Apache Beam pipelines as one giant file in a Dataflow Flex Template because custom containers never worked for me. The fix I had missed: a Flex Template is two images, a launcher and a worker, each with its own entrypoint.
The first agent I worked on had evals, but they only ran in CI/CD: a pre-flight checklist with no in-flight instruments. Here is the case for live, continuous evals as part of your observability: what to measure, what a decent score looks like, and how to tune them over time.
The right AI tool depends on who's asking, and every tool only works if it's grounded. How I split agentic tooling by persona for a fictional bank: DaaS golden paths for customers, a knowledge-graph-grounded analytics assistant for analysts, and a shared RAG knowledge base.
A walkthrough of the RAG-powered chat widget on this site - how it indexes blog posts, how grounding actually works, and why you don't need expensive managed services to build something solid.
You know you've created value when you've built a product that earns people's trust. A story from a loss-prevention project at AutoZone, and a manufacturing COO who ran the same playbook on a shop floor.
