Product updates, engineering deep-dives, and guides for building with WordPress + AI.

Direct costs are small and easy to look up. The number that decides it is the comparison: a 400-page meta backfill is ~15 hours by hand and under an hour with an agent.

Four of five failures are human, not model. What each layer prevents, the six-step recovery playbook, and how to handle the client conversation.

Translation quality isn't the failure — the plumbing is. Architecture, exclusions before languages, hreflang, linked drafts, glossary enforcement, and what humans keep.

The value is the audit, not the delete: what accumulated, where it came from, how confident the removal is — then a safe order, a verified backup, and a check afterwards.

The most common accessibility failure and the most boring to fix. A context-aware prompt, per-type handling, batch review — and a monthly habit so it stays fixed.

Between useless bulk actions and unforgiving SQL: express the set in plain English, review old → new, apply in batches — with a snapshot as the undo.

The pasting is the slow part. Drafts created in WordPress with real internal links and meta set, batch prompts for series, and the refresh prompt that beats new posts.

An agent authenticates as a user and inherits its ceiling. Application passwords, scopes, the four gates before your database, and the leak playbook.

Core has no model, no chatbot, no agent — but it does have the plumbing AI needs. What's actually in core, what plugins add, and why core probably won't ship an agent.

The connection and core tools are usually free; integrations, validated builder output and team features are where paid starts. Four questions that decide which you need.

WP-CLI assumes a trusted operator; MCP assumes the caller might be wrong. Where each wins, and the infrastructure-vs-content split that works in practice.

Webhooks are deterministic; agents exercise judgement. Which jobs belong to each, how the costs differ, and the one signal that a Zap should become an agent job.

MCP doesn't replace REST — it calls it. What the extra layer adds (discovery, semantic typing, governance), and the rule for choosing between them.

A buying guide, not a ranking: seven criteria with a test for each — typed tools, real plugin integrations, approval gate, audit log, scopes, builder-native output, client-agnostic.

Cursor plus MCP turns an editor into a site console: template and data in one prompt, review files for bulk work, rules as procedures, one repo per client.

Site tools next to code tools. The config, the workflows that genuinely belong in an editor, and the one mistake to avoid (committing credentials).

Gemini speaks MCP too. The config block, the verification prompt, the first jobs worth running — and the differences that actually matter in practice.

Three Claude surfaces, three configs, one server. How to connect each, how to verify the connection is real, and which one to reach for per job.

Five minutes from ChatGPT to real WordPress writes: connector setup, verification prompt, the ChatGPT-specific gotchas, and the safety defaults to keep on.

Plain-English definitions for every term in the WordPress AI stack, plus the four confusing pairs — Claude Skills vs Pro Skills, REST vs MCP, plugin vs server, automation vs agent.

Sites are becoming addressable, not just editable. What stops being work, what becomes work, how pricing shifts — and the incremental path to get there.

An agent is a model with tools and boundaries. What it can genuinely do on WordPress, what it can't, the permission model that makes it safe, and the first three jobs to give it.

Three layers, not one: content AI, builder AI, agent AI. What each actually delivers in 2026, which claims to ignore, and the four-part stack that pays for itself.

MCP is USB for AI: your site exposes typed tools, any AI client can call them. What that means in WordPress, what it lets you do, and where the safety boundaries sit.

Hands, know-how, specialist machines: how Claude Skills, MCP tools and NibWP Pro Skills differ, how they compose, and which combination each WordPress job actually needs.

Ten SKILL.md starters for agency work — audit, fix, SEO backfill, page/content from brief, Woo import, link sweep, migration, handoff doc, house standards — with frontmatter and the NibWP tools each drives.

From empty folder to hardened, shareable skill: a care-plan audit SKILL.md written line by line — description that fires, read-only procedure, report template, allowed-tools, tests.

A complete SKILL.md that turns a brief into a native Bricks/Etch page through NibWP: outline first, learn the site, route through the Pro Skill, draft, verify, report.

Skills are know-how, MCP is hands. What a Claude Skill actually is, how it loads, where it runs, and how it operates a WordPress site through NibWP with approvals and an audit log.

A menu that adds to 100K, capacity math without fantasy, the operating week, and the three-layer system that makes one disciplined person enough.

More checks than hand delivery, an hour of human judgment, evidence in every report — the care plan that's honest on both sides of the invoice.

Presence, persuasion, transformation, operation — the price ladder decoded: real scopes, AI-era delivery hours, and the proof to demand at each tier.

The site is its own spec: rebuild from rendered pages onto a new token system, art-direct the twelve pages that matter, template the rest.

The store setup is easy; the product data was the swamp. Reviewed import batches drain it — 35 hand hours down to 14, one week door to door.

Model, templates, corpus, money loop — the five-system 20K project delivered in three weeks, with the recurring plan directories genuinely need.

Five stages, each ending in proof: inventory, move, rebuild, redirect map, watch window — the team-month migration as a person-fortnight.

Hour by hour: brief at 9:00, native sections by lunch, live by 17:00 with a report — the smallest, sharpest ticket on the agency rate card.

Tokens first, sections once, pages by composition — the conversion pipeline that turns a shelved 10K template into a library that prints margin.

Scope, day-by-day build with real prompts, the hours ledger and the margin math — the bread-and-butter agency project, delivered the NibWP way.

Price the outcome, not the hours: four billing models, the client script for “but AI did the work”, and the audit log as your accountability proof.

Real Bricks elements, global classes written the way Bricks stores them, validated nesting, query loops for repeating content — approved before saving.

Crawl the dead URLs, map old to new, approve one redirect plan, fix internal links at the source, verify — one session, not a lost week.

Claude Code, Claude Desktop, Cursor, Windsurf, ChatGPT — what each client is best at on a WordPress site, and the one-endpoint setup they share.

The real troubleshooting map: disabled Application Passwords, missing HTTPS, security-plugin 403s, domain locks, cache poisoning and missing tools.

Model the listings, generate the templates — cards, archives, wired search with map — and bulk-import the content, approvals throughout.

Tokens, validated components, pages composed from the library, dynamic loops — the full EtchWP site build, three small approvals at a time.

The honest map: what AI runs on Divi today (content, SEO, maintenance, Woo), the native-module limit, and the migration path for pages that matter.

The full arc — global styles, pages as real Kadence blocks, Elements for site chrome, content pass and pre-launch checklist, all behind approvals.

From a one-line brief to a real Elementor page — flexbox containers, kit-bound widgets, registry validation and an approval before anything is saved.

Move the senior process out of one head and into playbooks the AI must follow — same standard, every build, every teammate, every site.

The full map of typed MCP tools an AI agent can call on a WordPress site — content, SEO, security, migration, visual checks, integrations and builds.

Package maintenance under your brand and run delivery on autopilot — jobs and playbooks do the checklist, your team does the approvals.

A scheduled crawl on every site, one portfolio-wide sweep, approval-gated fixes, and a report that turns link rot into a billable care-plan line.

The five-minute setup for Claude Code, Claude Desktop, Cursor, ChatGPT and Windsurf — plus the real failure modes and how Status diagnoses them.

The answer we give clients: the realiztic failure modes, the five structural controls, and the audit log that turns trust into a record.

Screenshot, Figma, URL or HTML into real Elementor widgets — flexbox containers, global-kit bindings, validated against your live widget registry.

The plugin to connect AI to WordPress: Claude, Cursor and ChatGPT get 132 typed tools across 35+ integrations — building and managing your site with approvals and logs.

An AI WordPress workflow for teams: encode your build process once as a playbook and every request follows it — same standards, same output, junior or founder.

White-label AI WordPress services: NibWP runs back-office on your clients' sites — your brand on the reports, your rates on the invoice, the AI never in the room.

Bulk create WooCommerce products from a spreadsheet, feed or photo folder: real products with prices, images and categories — planned, approved, written to your store.

Screenshot to WordPress page in minutes: NibWP turns any screenshot or URL into native, editable pages in Elementor, Bricks, Kadence or Etch — validated, approved by you.

Safe AI for client WordPress sites: approval before every write, typed tools instead of arbitrary code, scoped credentials, audit logs — control you can show a client.

AI-powered WordPress care plans: the monthly checklist, audits and client report run on jobs and playbooks — you keep the approvals, the margin and the renewals.

Automated AI WordPress maintenance: scheduled jobs crawl links, scan security, collect updates and clean the database overnight — and ask before changing anything.

Manage multiple WordPress sites with AI: one connected assistant works across your whole portfolio — updates, audits, content and fixes, each site gated by approvals.

NibWP is the AI website builder for agencies: screenshots, Figma files and HTML become editable pages in your client's builder — validated, approved, on your own WordPress.

Scheduled jobs, playbook-driven passes and approval gates that run the monthly care-plan checklist for you — and write the client report at the end.

Six step-by-step walkthroughs that turn HTML, URLs, screenshots and even email markup into real, editable Kadence blocks — native attributes, no custom CSS.


Etch hit stable 1.0 in 2026 and developers are leaving page builders behind. Real code access, clean semantic HTML, native blocks, no lock-in — here is why, plus the honest caveats.

A built-in key-value store so agents keep project context across sessions.

One endpoint, a 1-click config, and you are querying your site from any editor.

Detecting repeating cards in source HTML and turning them into CPT + ACF + Bricks posts element — automatically.

Why we built a separate skill to generate the palette + type ramp + spacing scale rather than baking it into EtchWP.

A stable, schema-typed contract instead of brittle scripts that break on update.

Walking through the 5-step pipeline that turns a Tailwind dump into clean BEM + ACSS tokens.

How agents author persistent, auto-loaded PHP isolated from your theme and core.

Why agents kept ignoring our Pro skills, and the 3-layer fix that finally locked them in.

Generate a ready-to-paste config for 15+ AI clients in a single click.

How we turned WordPress into a Model Context Protocol server in under 6 months.