Archive

Evaluation-Driven Development Rests on Judges Whose Reliability Is Rarely Measured

Evaluation-Driven Development Rests on Judges Whose Reliability Is Rarely Measured

THIS WEEK: The method has a formal process model, a documented practice, and a tooling market that consolidated sharply in 2026. What most teams skip is the step that determines whether any of it means anything.

Lina Sibbel
Lina Sibbel
ODCS v3.1.0 Settles the Data Contract Format Without Settling Enforcement

ODCS v3.1.0 Settles the Data Contract Format Without Settling Enforcement

THIS WEEK: The specification fight ended in December 2025 and the project graduated the following summer. What still has no default answer is what runs the file once it is written.

Lina Sibbel
Lina Sibbel
Kafka and Materialize Are Being Assigned Jobs They Do Not Perform

Kafka and Materialize Are Being Assigned Jobs They Do Not Perform

THIS WEEK: The multimodal lakehouse is a real architectural shift with real money behind it. But the reference designs circulating to implement it give two components jobs the documentation says they do not do.

Lina Sibbel
Lina Sibbel
The Warehouse Wars are Over. Welcome to the Catalog Wars.

The Warehouse Wars are Over. Welcome to the Catalog Wars.

THIS WEEK: The format war is won, but your next vendor lock-in is quietly hiding in the metadata layer.

Lina Sibbel
Lina Sibbel
Where AI pipelines actually go to die.

Where AI pipelines actually go to die.

THIS WEEK: Part 3 of 3. Ollama, LiteLLM and Langfuse: serving, cost and observability, the half no tutorial covers.

Lina Sibbel
Lina Sibbel
Stop writing prompt spaghetti.

Stop writing prompt spaghetti.

THIS WEEK: Part 2 of 3. DSPy, Instructor and Outlines: turning output that usually works into output that cannot break.

Lina Sibbel
Lina Sibbel
4 open-source tools that fix RAG before the model ever runs

4 open-source tools that fix RAG before the model ever runs

THIS WEEK: Crawl4AI, Marker, Chonkie and Qdrant, Part one of three, on building an AI pipeline you can actually trust.

Why Kimi K3 could be worse than the DeepSeek moment

Why Kimi K3 could be worse than the DeepSeek moment

THIS WEEK: Why the calm reaction to Kimi K3 is the wrong tell, and what a frontier open-weight model in the wild could actually cost your Enterprise.

AI Didn't Eat Data Engineering. It Split It in Two.

AI Didn't Eat Data Engineering. It Split It in Two.

THIS WEEK: A longer read on the state of data engineering careers in 2026.

Where your AI agent runs, decides how your pipelines get built

Where your AI agent runs, decides how your pipelines get built

THIS WEEK: Claude Code runs in your shell, OpenAI's Codex in the cloud. That gap, not the benchmark, decides which model is for you.

The Warehouse Wars Are Over. The Real Fight Is for Your Agents.

The Warehouse Wars Are Over. The Real Fight Is for Your Agents.

THIS WEEK: The five leading cloud data platforms are converging on the same design, and it changes what a data engineer should be optimising for.

The Kill Switch in Your Data Stack

The Kill Switch in Your Data Stack

THIS WEEK: Washington just proved a frontier model can be switched off for half the planet overnight. The exposure isn't legal, it's architectural, and it's already in your pipelines.

FirstBack
12345678
Next Last