Cursor turns a four-hour refactor into a 20-minute one, but only with a plan. My four steps: spot the trigger, write a plan.md, let Claude execute with test scripts, then compare the output against the original.
After 14 years I moved most of my work out of the IDE and into the terminal: Claude Code in Warp, a Git worktree per agent so they stop colliding, and stored slash commands so I am not retyping the same prompt.
Six months in, WordPress Core AI has a stable shape: the Abilities API lands in 6.9, WP Client AI abstracts the LLM providers, and the MCP adapter turns a site into a server that agents can call.
Junk topics usually mean the pipeline is wrong, not the data. Here is how seeded KeyNMF and an LLM summarization step keep a topic model focused without burning the whole compute budget.
Ninety percent accuracy sounds fine until the wrong JOIN lands on a production database. This covers why Execution Accuracy beats Exact Match, what Spider 2.0 exposed about enterprise schemas, and how to ground an LLM on the schema you actually have.
Notes on porting statistical models into WordPress: where research gets diluted instead of distilled, how real-time aggregates in user meta leak data through race conditions, the atomic SQL fix, and why a bigger LLM is the wrong answer to arithmetic.
A knowledge graph has three layers: an ontology, controlled vocabularies and observational data. This post walks through what each one does, and shows the WordPress version of the shift from flat meta_query lookups to taxonomy-based entity relationships.
Retrieval-augmented generation is past the demo stage. Notes on chunk size as an experimental variable, why HNSW precision degrades as the index grows, the latency cost of extra retrieval layers, and a transient wrapper for caching retrieval in WordPress.
Rolling-window forecasts treat every SKU as independent. This post models a WooCommerce catalog as a graph instead, with GraphSAGE over shared plants and attributes, the PHP that extracts SKU neighbors, and where heterogeneous transformers go further.
A multi-agent Autogen loop running Mixtral 8x7B locally on an M3 MacBook rewrote a Rust matrix multiply until it hit NEON SIMD and Rayon, taking it from 760ms to 359ms. Notes on how the loop was set up and where the generated code went wrong.