Articles, one at a time.
Every piece cites primary sources and adds something you can act on: a worked example, a tested command, a documented limitation. Drafts that only summarize what already exists stay unpublished.
Your Prompt Tooling Has a Deadline Your Monitoring Can't See
Two vendors are switching off prompt and eval tooling. We probed the live APIs, found no warning headers, and built a scanner that dates every hit.
Read →
OpenTelemetry GenAI Tracing for Multi-Step Agents in Production
Instrument multi-step LLM agents with OpenTelemetry GenAI semantic conventions so a failure becomes a readable trace. Includes a local demo, no paid API keys.
Read →
MCP Server Supply-Chain Vetting: A Quarantine Pipeline for Tools
A vetting and quarantine pipeline for third-party MCP servers: scan install scripts, verify provenance, and test candidates in a secret-free sandbox first.
Read →
MCP Tool Poisoning Defense: A Description and Provenance Auditor
MCP tool poisoning hides instructions in tool descriptions to hijack agents. Build a local description-hygiene and provenance auditor for security reviews.
Read →
OpenAI Spend Limits Return 429: Your Retry Logic Will Make It Worse
OpenAI hard spend limits reject requests with a 429. Lab evidence on why default client retries treat it as a temporary blip, and the one-line fix.
Read →
Your AI Gateway Blocked the Request. Did You Still Pay?
We measured what OpenRouter returns when a request is refused, and whether refused calls still cost money. Ten blocked calls moved the bill by $0.00.
Read →
OpenAI Output Moderation: What the Scores Actually Catch
We asked an AI assistant for a threatening collection letter. It wrote one. Output moderation scored it 0.00008 and let it through. Here is what that means.
Read →
GPT-5.6 Prompt Caching: The 24-Hour Cache Is Gone
GPT-5.6 added explicit prompt cache controls and quietly took the 24-hour cache away. We ran the API and measured what that swap costs.
Read →
When an AI Agent Bungles a Tool Call, Does It Fix Itself?
We ran an OpenAI API probe on three tool-misuse faults an order agent hits (schema drift, rate limits, fake-success replies) and recorded the recovery.
Read →
GPT-5.6 Programmatic Tool Calling: When It Cuts the Token Bill
We ran GPT-5.6's programmatic tool calling on a real API. On one task it cut tokens 92%. On another it cost 2.2x more. Here's the rule that decides which.
Read →
Framer Review 2026: AI Website Builder Guide
Framer review for 2026: AI site generation, CMS limits, current pricing, code components, and how it compares to Webflow, Squarespace, Wix, and WordPress.
Read →
Claude Programmatic Tool Calling: Measured Token Savings
Claude can run your tools from sandbox code so big results never hit its context. We measured what that saves on a real API: 98% fewer tokens on one task.
Read →