This is a recap of Jukebox podcast episode #216 with Matt Schwartz, continuing a two-part conversation about practical ways WordPress agencies can adopt AI — and what to watch out for. Matt runs a WordPress agency in Atlanta (since 2011) and built CheckView, a QA product for forms and checkouts. The episode focuses on realistic wins, workflows, and the risks agencies must manage.
Giving AI access to your agency’s brain
Matt recommends connecting your AI chatbot to your existing knowledge base: PM tools (ClickUp, Asana), wikis, SOPs, and client docs. When the AI can check your current project data before answering, it reduces hallucinations and yields actionable, document-linked responses. This is an easy, low-cost way to leverage what you already have and make answers reproducible across the team.
Model Context Protocol (MCP) and internal guardrails
MCP (Model Context Protocol) is an open approach to bridge AI agents to your internal systems without exposing individual API keys. Matt describes building an internal MCP as a centralized proxy: employees query the MCP and it fetches data from multiple services securely. That simplifies access, centralizes guardrails (what can and cannot be changed), and reduces the risk of exposing credentials. He cautions that MCPs are powerful and should enforce strict permissions (for example, disallow destructive actions) and be considered by more technical shops or via vetted SaaS MCP providers.
Vibe-coded agency tools: internal vs public
“Vibe-coded” refers to quickly built, AI-assisted internal tools. These can be valuable for reporting, profitability dashboards, and internal automation — low-risk uses where mistakes are containable. But be cautious about replacing established public tools (site management dashboards, client-facing automation) with rapidly generated code because AI lacks deep context and error-handling, and mistakes can be catastrophic. One-off disposable tools can be useful for short-term tasks; persistent or critical systems require careful review, QA and maintenance plans.
QA, checklists, and Claude Skills
AI shines at repetitive checklist work. Use it to generate SOPs, create launch checklists, and run routine checks so humans focus on high-risk items. Claude Skills (or similar agent “skills”) allow you to teach an agent a process (for site launches, onboarding, etc.) and run it with human prompts and checkpoints. Matt recommends designing an “AI vision document” that defines which steps are automated and which require human sign-off (e.g., verify no-index flag manually). Automating low-risk checks increases coverage and frees time for deeper human review.
Impact on the WordPress plugin market and community
AI lowers the barrier to producing small utility plugins or one-off code snippets, which may reduce sales for small plugin authors. That could push smaller contributors out of the ecosystem, shifting the market toward larger plugin companies and potentially weakening the community that makes WordPress vibrant. Matt and Nathan both urge awareness: the community value of small projects and the onramps they provide to contributors matter and could be at risk if disposable AI-generated solutions replace those entry points.
Experimenting beyond WordPress
Because AI helps with unfamiliar stacks, agencies can experiment more easily with non-WordPress platforms for certain projects (static sites, new frameworks). Some agencies are exploring alternatives for brochure sites or experimenting with headless/static options when those better fit a client’s needs. WordPress’s open APIs and documentation remain a strong advantage, but AI enables quicker exploration of adjacent tools and architectures.
Risks and cautions
– Data security: Beware submitting secrets or client data into third-party chat tools. Some services treat inputs as part of their data; always check terms and use private/self-hosted options where required.
– Overdependence on vendors: If a vendor changes pricing, plan, or features (e.g., removing capabilities from a low-cost tier), your costs and workflows can shift dramatically. Design processes that can tolerate changes in AI service pricing or availability.
– Error handling and testing: Quickly generated tools often lack robust error handling, logging, and validation. Build logs, tests, and monitoring into any production system an AI helps create.
– Human-in-the-loop: Keep humans responsible for high-risk decisions and mission-critical checks; AI should augment, not replace, oversight.
– Hidden inputs and privacy: Be mindful of what gets captured by AI conversations — sensitive numbers, contract details, or client pricing can leak or be reused by the model.
Likely outcomes for agencies
– Hiring shifts: Some execution-level roles may be reduced as agencies automate repeatable tasks, while demand for higher-level strategists, AI managers, and monitoring roles grows.
– Productised services: AI enables more refined, niche productised offerings (e.g., plumber-specific website packages) by codifying repeatable processes and accelerating setup.
– New tooling focus: Investment will move toward automation, monitoring, and QA/observability tools that ensure AI-driven processes remain correct and reliable. Humans will become managers of AI-driven workflows rather than performing every execution step.
Practical takeaways
– Start small: Connect AI to your PM/wiki to unlock immediate benefits.
– Define vision and guardrails: Document what AI is allowed to do and which steps require human checks.
– Use MCPs or a centralized bridge where appropriate to reduce credential exposure.
– Treat AI-generated code and tools like any other software: require code review, logging, testing, and monitoring.
– Be mindful of community impact: support contributors and consider how AI changes the economics of small plugins and projects.
Where to find Matt and resources
Matt is active in the Admin Bar Facebook group, on LinkedIn, and runs CheckView (checkview.io). The episode show notes on wptavern.com/podcast include links and a longer document Matt shared with detailed examples and resources.
Bottom line: AI can deliver quick, real wins for agencies — better documentation use, automation of routine QA, improved reporting, and more efficient productised services — but only when paired with clear guardrails, human oversight, secure integrations, and an eye toward long-term costs and community consequences.