THE PUBLICATION ARCHIVE
AI Strategy
An Agent Evaluation Score Needs an Operational Decision Behind It
Evaluation becomes useful when a result changes a release decision. Before optimizing a score, define which user outcomes must hold and what evidence would justify deploying a changed…
Google demonstrates a continuous evaluation loop with Agents CLI
Google Developers has published a walkthrough of the Agents CLI evaluation workflow. It covers generating traces, grading results, comparing runs, analyzing failures, and using evaluations to test changes.
Model Routing Needs a Budget and a Quality Contract
Automated routing moves a decision from the developer to the platform. That can reduce friction, but the organization still needs to define the quality, cost, and accountability expected…
GitHub Copilot Fusion becomes generally available for model orchestration
GitHub says Copilot Fusion is generally available, combining model selection with multi-model orchestration. Developers select a system and effort level rather than choosing a model for every step.
Treat Inference Cache Isolation as a Security Requirement
Tenant isolation must remain correct when an inference request fails. This incident is a reason to ask how a serving stack handles rejected input and reused memory, rather…
Cloudflare discloses remediation of cross-request Workers AI cache exposure
Cloudflare has published an incident account involving cross-request KV cache exposure in its GLM-5.3 deployment on Workers AI. The company describes a configuration mismatch combined with an SGLang…
An AI Pilot Needs an Exit Criteria
A promising demo can conceal a costly operating commitment. Before an AI pilot begins, decide what evidence would justify expanding it—and what result means stopping.