This rewrite captures the key takeaways from a two-part conversation between Nathan Wrigley and Matt Schwartz about how WordPress agencies can adopt AI thoughtfully, where it helps most, and where to be cautious.
What AI can access: give AI your agency’s brain
– If your agency already uses a PM tool or wiki (ClickUp, Asana, internal docs), connect that knowledge to your AI assistant so it checks your real documentation before answering. That reduces hallucinations and lets the AI leverage SOPs, client notes, and CRM data without complex integrations.
– The practical payoff: faster, more accurate internal answers, automated bullets with sourced footnotes, and fewer repeated questions across the team.
Model Context Protocol (MCP) and internal guardrails
– MCP is an agent-friendly way to connect chatbots to multiple systems. Think of it as a single gatekeeper that abstracts APIs, so staff don’t need individual API keys or multiple connectors.
– An agency-level MCP can proxy requests to WordPress REST APIs, help desks, bookkeeping, and more. It centralises access and makes it easier to apply guardrails (for example, disallowing destructive commands like site deletion).
– Some SaaS providers already offer managed MCPs. That’s convenient but remember you’re giving a third party credentials and secrets — weigh convenience against security.
Vibe-coded internal tools: when it makes sense and when it doesn’t
– Low-risk internal tools are a sweet spot: reporting dashboards, profitability analysis using QuickBooks + time tracking, or merge views of WooCommerce subscriptions. These can uncover insights you wouldn’t have time to get otherwise.
– Risky moves include replacing mature public SaaS (ManageWP-like dashboards) or building public-facing website management tools with AI-generated code. The cost of maintenance and the likelihood of edge-case failures make this dangerous unless you have strong code reviews and QA.
– There’s a growing trend toward disposable or one-off apps for short-term needs. That can be sensible for low-risk, single-use tasks but avoid one-offs for persistent infrastructure.
QA, checklists and Claude Skills
– AI excels at automating repetitive checklist items and building SOP-driven processes. Tools like “Skills” (teach the model step-by-step processes) let you automate parts of onboarding, launches, and routine QA while keeping humans in the loop for high-risk items (e.g., checking noindex settings).
– Use AI to cover the low-risk bulk of testing, freeing humans to validate critical or ambiguous checks. Always define which steps require human review in an “AI vision” document.
Impact on the WordPress plugin ecosystem and community
– Smaller utility plugins that solve narrow problems are seeing pressure from AI-built alternatives. Some plugin developers report drops in sales as agencies implement quick AI solutions instead of buying plugins.
– Beyond economics, there’s a community risk: small plugin creators and contributors play a big role in the open-source ecosystem. If they disappear, WordPress culture and innovation could be diminished.
Experimenting beyond WordPress
– AI lowers the friction for trying unfamiliar stacks and platforms. Agencies can evaluate modern static or hybrid approaches (e.g., Astro, edge-based setups) more quickly and decide when WordPress is the right fit versus simpler brochure sites.
– That experimentation can open new business opportunities but should be chosen based on project needs, not just novelty.
Risks and cautions
– Data security and privacy: inputting client data into chat tools can expose sensitive information. Some platforms treat submitted content as logs or training data — check terms and use secure, enterprise-grade options when needed.
– Vendor dependence: many teams rely on a few AI suppliers. Plans change (features or pricing can disappear or move behind higher tiers), so avoid designing operations that cannot function if costs spike or capabilities shift.
– Error handling and logging: AI-generated tooling often lacks robust error handling. Build validation, logging, and alerts into any AI-driven system so edge cases don’t silently break core processes.
– Overconfidence and low friction: conversational interfaces make it easy to try things without deep understanding. That reduces friction but can lull teams into assuming correctness where it doesn’t exist.
Practical recommendations
– Create an AI vision document: define where AI will be used, what data it can access, what must be human-reviewed, and how success is measured.
– Start with low-risk automation: reporting, checklists, internal SOP assistants, and QA helpers.
– Keep humans in the loop for mission-critical and public-facing operations. Require manual reviews for high-risk steps.
– Centralise access with an MCP or managed connector carefully, and apply restrictive guardrails (no destructive permissions by default).
– Instrument all AI tools with logging, monitoring, and error handling.
– Plan for vendor cost increases: don’t assume current low prices are permanent.
Likely outcomes for agencies
– Hiring may change: routine execution becomes more automatable, so roles may shift toward strategy, productisation, monitoring, and AI management. Junior execution roles could shrink or be repurposed.
– Productisation rises: agencies can package niche, repeatable services more easily (e.g., plumber-specific site builds) using automated processes and AI assistance.
– New tooling focus: expect growth in QA, monitoring, and observability tools designed to watch AI-driven automations and ensure they behave over time.
Bottom line
AI offers big efficiency gains for WordPress agencies when applied thoughtfully: centralise knowledge, automate low-risk repetitive work, and keep humans responsible for critical judgments. At the same time, watch for security, vendor, and community impacts. Build guardrails, log thoroughly, and plan for change — with those precautions, agencies can harness AI to scale services, improve testing, and productise offerings without sacrificing reliability or the values of the WordPress ecosystem.