GuideObservability
Shadow-testing an AI log classifier before migration
Run a candidate log classifier beside the current route, preserve complete outcomes, and promote only the slices whose mistakes and review cost are understood.
SRE
Reliability engineering, from service objectives to operational judgment.
GuideObservability
Measuring streaming LLM latency from first text to completed work
A client-side SSE probe separates first answer text from normal completion. Fixed arrivals and outcome counts show whether a serving change delivers more useful answers.
ExplainerAI in production
What Kimi K3’s million-token context offers SRE teams
Kimi K3 can keep a larger incident record in view. Its practical value depends on evidence handling, agent integration and the cost of useful investigation.
NewsAI in production
Data Agent Kit brings Bigtable queries into coding agents
Bigtable joins Google’s Data Agent Kit. An illustrative rollout investigation shows why retained cell versions need an application-level evidence check.
GuideIncident response
An SRE incident handoff in ChatGPT Space
A synthetic order-service incident follows a shared handoff through page edits, new findings and shift change, including the review and access work involved.
ExplainerAI in production
Why ninfer-ext’s smaller EXL3 model can run more slowly
ninfer-ext trades a smaller EXL3 file for more decoding work in its published tests. The saved memory matters when it lets a request fit that would otherwise fail.
ExplainerAI in production
Pi.dev: a coding assistant you can adapt to your work
Follow three incident files through Pi’s agent loop to a report another engineer can verify, then decide which customization is worth maintaining.
NewsAI in production
DeepSeek Harness connects a desktop agent to your models and tools
The desktop preview packages a configurable coding agent. Its plugins determine which model answers, which files it can read, and where commands run.
GuideObservability
OpenTelemetry tail sampling through a checkout failure
Tail sampling selects recorded traces by error, latency and other evidence. A checkout timeout shows how routing, waiting time and memory affect the trace a responder can retrieve.
ExplainerAI in production
Jev returns typed answers for routing and scoring tasks
Jev answers supplied questions with choices, scores and probabilities. A token-rotation ticket shows how those structured outputs can fit a support router.
NewsAI in production
Why GLM-5.3’s cyber assessments report different scores
Anthropic’s 50 successful attempts and CAISI’s 61.1% score measure different things. Their task definitions explain what each assessment says about GLM-5.3.
GuideAI in production
CodeRabbit: how to use AI code review
CodeRabbit brings AI feedback into pull requests, editors and the CLI. A retry loop that exceeds its caller’s deadline shows the kind of finding that can repay review time.