An LLM can retrieve the right document and still quote the wrong price. This is the detection and healing layer I built to catch that: five failure patterns, a faithfulness score, deterministic patches, and a SQLite drift monitor that caught a real bug in staging.
AI has absorbed the task-level work, and juniors who sell themselves as code writers are getting ghosted. After 14 years in WordPress development, here is what I look for instead: work you actually owned, and a repo that shows a human debugged it.
Ghost is an agent-first Postgres platform. Notes from using its MCP server and database forking to let AI agents test schema changes and index strategies without touching a shared staging database.
Prompt caching only pays off when the prefix stays byte-identical. I go through the mistakes that break it, a semantic cache layer for WordPress, loading tool definitions on demand, and what to keep or drop between turns.
Anthropic’s subscription ban pushed me off Claude for OpenClaw. Here is the OpenRouter setup I use with Kimi-K2.5, what it costs, where it runs slower, and the GDPR problem with hosted weights.
Part two of the Fernao build log: the ICS calendar feed needed five minutes to generate a schedule, so I moved to the Google Calendar API and got it down to twenty seconds. The new Task Breaker module splits oversized tasks into 20-minute subtasks.
Studio Code Beta is a CLI coding agent for WordPress, built on Claude Code. It runs WP-CLI, spins up local sites, screenshots its own work, and validates block markup against the editor before inserting it.
Vibes-checking AI output is the new bottleneck. What actually moved my Claude Code performance: terminal permissions for the agent, a test requested with every feature, and a no-stop rule until they pass.
Prefill is compute-bound, decode is memory-bound, and running both on one GPU wastes most of it. Splitting them into separate pools, the KV-cache transfer tax that comes with it, and when it pays off.
Retrieval gets worse as the memory pool grows, while the confidence score keeps climbing. Here is what causes RAG memory decay and the four fixes I put into production systems: topic routing, semantic deduplication, relevance based eviction, and lexical reranking.