Observing Codex MCP Tool Calls with Langfuse
A practical local setup for tracing Codex MCP requests, tool execution, and evaluation without sending application data to the cloud.
Read storyA practical local setup for tracing Codex MCP requests, tool execution, and evaluation without sending application data to the cloud.
Read storyA plain-English walkthrough of how Markdown knowledge pages become searchable AI context using embeddings, pgvector, HNSW, and evaluation checks.
Read storyHow vertical feature slices and dependency rules can keep product workflows easier to change.
Read storyHow runtime schemas can document boundaries, normalize inputs, and keep API and UI contracts consistent.
Read storyWhy localization is a product boundary that affects routes, validation, dates, permissions, and support.
Read storyHow drag-and-drop UI can support ordering workflows without hiding keyboard access or persistence rules.
Read storyHow I think about model tools as narrow product interfaces instead of hidden access to an application.
Read storyWhy copied UI primitives can become a practical product system instead of a black-box dependency.
Read storyWhy a static-first Astro site is a good fit for curated engineering writing and professional notes.
Read storyWhy some product workflows need dense tables, dirty-state tracking, and careful batch saves.
Read storyWhy AI tools need schemas, permissions, confirmation, and audit trails before they change product data.
Read storyHow I split workflow tests between fast service checks and a few browser paths that matter.
Read storyHow runtime validation and readable docs help keep public API behavior from drifting.
Read storyWhy I used a database-backed job queue for slow product workflows, and what had to be visible to users.
Read storyWhy durable product state belongs in the database, and where Prisma should stop leaking into the app.
Read storyHow route loaders and actions can make user tasks explicit without putting every rule in the route.
Read story