WordPress now has WP-Bench, a benchmark that tests AI models on Core APIs, security patterns and block development. It runs the generated code in a real WordPress runtime instead of grading answers on paper.
Hardcoded API keys hold up until the first model deprecation. Here is how I use AWS Bedrock as the gateway instead: Boto3 calls, inference profiles, IAM permissions, and a triage assistant that runs on Lambda rather than inside the WordPress request cycle.
Longer context windows cost server memory, not just compute. Infini-attention replaces the growing KV cache with a fixed-size compressive memory matrix updated by a Delta Rule, which cut memory use by 114x in Google’s tests.
Prompting stops scaling as soon as a task has real edge cases. Here is what the Stanford ACE research found about its generator, reflector and curator loop, where the gains show up, where they do not, and how to keep the same kind of playbook in WordPress.
Vanilla vector search breaks down as soon as your data is relational. Notes on GraphRAG, enforcing a data contract on LLM output in PHP, and the production RAG lessons worth stealing from the Towards Data Science write-ups.
Query optimization and neighbor expansion can add 40 to 50% to latency and cost for marginal quality gains. What the numbers say about when each add-on is worth shipping in a RAG pipeline, and when a naive baseline is fine.
Most LLM agents ship with no visibility between the input and the answer. How the NeMo Agent Toolkit traces every tool call and scores trajectory rather than answer accuracy alone, using Arize Phoenix and W&B Weave.
The difference between data drift and concept drift, the two tests I reach for first (K-S and PSI), and how I wire the check into a WP-CLI command so a drifting model triggers an alert instead of a bad quarter.
Agents pick the tool with the longer description, not the right one. What that length bias looks like in a WooCommerce integration, how to keep internal arguments out of the prompt, and why a proxy beats editing someone else’s MCP server.
Three habits that stop a coding agent inventing hooks and tables: an AGENTS.md file it has to keep updated, real documentation and schema files in the context, and a fresh thread when the task changes.