Investigate a failure in the telemetry pipeline
Distinguish a healthy service from missing evidence. Follow collection, buffering and export failures before trusting an empty dashboard.
Articles
Guides, news and explanations about AI and reliable systems.
Not sure where to begin? Choose a short reading path, or browse at your own pace.
Keep useful telemetry without counting every job as a new metric
See how a job ID multiplies metric series, why a small trace sample can miss failures, and which records should answer each investigation.
T3 Code explained: one workspace for your coding agents
T3 Code brings your coding agents, project conversations and diffs into one workspace. Learn how it works, try a small repair and understand what stays on the host.
Agents and tool calling: follow one complete exchange
Trace a synthetic status lookup from proposed function call through application validation, execution and returned evidence.
Embeddings and vector search: finding the right passage
Encode a small document collection, inspect similarity results and distinguish a relevant passage from one that applies to your system.
OpenCode 2.0 explained: what it does and how to use it
OpenCode is an open-source AI coding assistant that works with your project’s files and tools. Here’s what that means, what changed in 2.x, and how to try a small repair without handing it your whole environment.
Help the model find the exception that changes the answer
Follow a certificate-renewal question through retrieval, context assembly and citation checks, using a small Python packet you can inspect.
Tokens and context windows: what your model actually receives
Inspect the assembled request, count text tokens, reserve output space and shorten context without losing the condition that changes the answer.
Give Gemini 4 Argon a migration you can actually check
Understand Argon’s model, context and API costs, then compare its handling of a bounded migration or investigation with the tool you already use.
What are model weights, and how does training change them?
Change two learned numbers, save them and use them again. The small model explains what training changes and why a better fit still needs testing on new examples.
Checking AI incident-summary citations with Python
A quote can be exact and still come from the wrong release. A small Python checker catches that mismatch; a second example shows why the right quote can still support the wrong restart advice.
Microsoft UserRepGen makes inputs for a simulated user
A support assistant can answer well and still ignore a restriction revealed later. UserRepGen prepares the user description for testing that kind of conversation.