The latest AI Contributor meeting covered the global off switch for WordPress 7.0, whether the MCP adapter ships standalone or bundled, and image editing graduating to a core feature. Here is what each of those decisions means for your sites.
Offline evaluation for LLM agents: routing accuracy, LLM-as-judge scoring, and RAG faithfulness, plus the quality gate I use to block a deploy when the scores come in under threshold.
WordPress.org now runs an MCP server for the Plugin Directory. It lets AI editors like Cursor and Claude validate readme.txt, check submission status, and submit plugins without leaving the editor.
AI 0.6.0 turns experiments into a type of feature, adds iterative image editing in the Media Library, and renames hooks. C2PA content provenance is queued for 0.7.0.
How to linearize a nonlinear objective with piecewise linear approximations, keep the segments adjacent using SOS type 2, and hand the result to an ordinary LP/MIP solver like Gurobi.
WordPress.com has added 19 write abilities to its MCP integration, so an agent can create posts, pages, media and terms. Drafts by default, explicit approval before every change, and your own role still caps what it can reach.
A readmission model with 94% accuracy made things worse because it found a correlation, not a cause. Where Pearl’s ladder of causation comes in, and how DoWhy’s four steps keep you honest.
Same ten numbers, two different variances. NumPy defaults to the population formula and Pandas to the sample one. Here is what ddof does and how to set it so your pipeline and your prototype finally agree.
Mean imputation and unclipped outliers are what break credit risk models. This covers splitting first to avoid leakage, IQR clipping with training-set bounds, and what to do when the missingness itself is informative.
An agent turned a missing database index into a 500-line middleware layer. What I do differently now: plan mode before any code, core WordPress functions first, and a read through the tool calls rather than just the final file.