In the second installment of our mini-series with Matt Schwartz, he expands on practical, cautious ways agencies can adopt AI to gain real benefits while avoiding common pitfalls. Matt runs a WordPress agency in Atlanta (since 2011) and created CheckView, a QA product focused on forms and checkout. His focus is on pragmatic adoption: where AI helps most, where to be careful, and how to protect clients and businesses.

Give AI access to your agency’s brain

One low-friction win is connecting an AI chatbot to existing agency documentation — PM tools, wikis, CRMs. Instead of complex integrations, point the bot at your ClickUp, Asana, or internal wiki so it can confirm answers against up-to-date SOPs and project data. This reduces hallucinations and leverages the knowledge you already have. The approach is especially useful for internal support, recurring questions, and generating annotated outputs (answers with citations to source docs).

Model Context Protocol (MCP) and internal guardrails

MCP — Model Context Protocol — is an emerging, agent-focused way to bridge AI tools to external systems. Rather than giving every employee API keys for every system, agencies can build or adopt an internal MCP that acts as a secure proxy to all tools. This centralizes access, applies guardrails (for example: prevent deletion of sites), and simplifies usage. SaaS vendors are already offering hosted MCP-like solutions, but be mindful of who holds your credentials and the security model.

Vibe-coded internal tools: when they make sense

Agencies are creating lightweight, internal “vibe-coded” tools—small, purpose-built apps for reporting, profitability dashboards, and internal workflows. These are lower risk if kept internal, well-guarded, and used for non-public tasks. They’re a great place to get value from AI: combine QuickBooks, time-tracking, and ecommerce data for insights you previously couldn’t get easily.

Be cautious about replacing mature public-facing SaaS

Matt cautions against replacing robust, external SaaS like ManageWP or established website management tools with fragile, internally built alternatives just to save a small fee. Public-facing or mission-critical systems require thorough code review, error handling, testing, and ongoing maintenance — costs that often outweigh perceived savings.

Disposable one-off apps

There’s a trend toward single-use or short-lived apps built quickly by AI (for a migration, a seasonal campaign, or a one-off client need). These can be fine when risk and maintenance needs are low, but persistent tools (like QA systems or management dashboards) should be built with longevity, validation, and monitoring in mind.

QA, checklists, and automation

AI shines at automating repetitive checklist tasks and expanding test coverage. Tools like Claude Skills let you teach a process and then run it in context: onboarding, site launches, and routine QA checks. The sweet spot is binary or low-risk checks (is noindex on? are redirects configured?) that multiply into big time savings. High-risk items should stay human-verified. Use AI to augment testing, not replace human judgment.

Impact on the WordPress plugin ecosystem

AI lowers the barrier to creating small, single-purpose plugins or one-off code fixes. That has already started affecting sales for many small plugin authors and could erode the diversity of contributors in the ecosystem. Larger plugin shops are consolidating around core product moats, while many smaller plugins struggle. Matt and Nathan worry about the community implications: fewer small contributors can mean less vibrancy and fewer entry points for new developers.

Experimenting beyond WordPress

AI also makes it easier to explore unfamiliar stacks and platforms. Some agencies are experimenting with static or headless approaches for brochure sites or trying new frameworks because AI helps bridge knowledge gaps quickly. That experimentation can be beneficial, but agencies should pick technology based on project fit, not novelty alone.

Risks and cautions

Matt highlights key risks agencies must consider:
– Data security and leakage: many AI tools retain conversation data and might treat inputs as training material or public. Treat any sensitive client data with caution and prefer secure, private options.
– Vendor dependence and cost volatility: commoditized AI features may be free today and costly tomorrow. Design processes so you can reduce AI reliance if pricing or capabilities change.
– Lack of error handling and testing: quickly generated tools may miss edge cases, lack validation, or produce fragile code. Build logs, monitoring, and robust error handling.
– Overconfidence and frictionless interaction: low friction can lull teams into assuming infallibility. Maintain a human-in-the-loop for key decisions and critical systems.

Likely outcomes for agencies

Matt expects hiring practices to shift: some execution work will be automated, reducing demand for junior execution-only roles while raising demand for strategy, automation design, and oversight skills. Agencies that productize processes (vertical-specialized packages for plumbers, dentists, etc.) can deliver more consistent outcomes by combining domain knowledge with AI automation. The human role will increasingly be manager-of-AI: designing, validating, monitoring, and iterating on automated workflows. This will also create demand for better QA and monitoring tools.

Final advice

Dabble, document, and be deliberate. Start with low-risk automations, connect AI to the knowledge you already own, and avoid baking critical systems that cannot function without a specific AI vendor. Build guardrails, logs, and processes to handle errors. Above all, protect client data and the community values that make WordPress strong.

For more detail and links, see the show notes at wptavern.com/podcast. Matt’s work and tools (CheckView, Inspry) are available if you want to explore these ideas further.

This is a rewritten summary of the second part of Nathan Wrigley’s Jukebox podcast conversation with Matt Schwartz about how AI is reshaping WordPress agencies.

Who is Matt Schwartz

Matt runs a WordPress agency in Atlanta (since 2011) and builds QA tooling for WordPress forms and checkouts. He focuses on practical, cautious ways agencies can incorporate AI to gain real benefits while avoiding common pitfalls.

Episode focus

This episode builds on a previous discussion. It moves beyond general enthusiasm for AI to concrete strategies agencies can use: linking AI to internal documentation, employing Model Context Protocols (MCPs) to safely bridge systems, building internal tools and QA automation, watching plugin-market effects, experimenting beyond WordPress with low-risk stacks, and managing security, vendor, and reliability risks.

Key takeaways

1) Let AI access your agency brain
– Point AI chat agents at your project management systems, wikis, or documentation so answers are based on your real processes (ClickUp, Asana, internal SOPs).
– This reduces hallucinations and lets AI pull up-to-date, contextual answers that include citations or links back to source documents.
– It’s an easy, low-effort way to bring value from AI without heavy engineering.

2) Model Context Protocol (MCP) and internal guardrails
– MCP is an agent-friendly, open approach to connecting AI with external systems. Think of it as a gatekeeper that knows how to safely map actions to APIs.
– Agencies can build or subscribe to an MCP that aggregates connections (billing, CRM, sites, helpdesk) so employees don’t need individual API keys.
– MCPs also enable role-based guardrails (for example, preventing destructive actions like deleting a site) and centralize control.
– Some SaaS providers already offer hosted MCP-like services; evaluate trust and security before outsourcing credentials.

3) Vibe-coded (one-off) agency tools: pros and cons
– Agencies are experimenting with small, AI-generated internal apps: reporting dashboards, profitability views, time-tracking summaries — generally lower-risk and useful.
– Be cautious about replacing established SaaS (site management dashboards, public-facing tools) with quickly generated internal tools: maintenance cost, errors, and missing error-handling can make that a false economy.
– There’s a rise of disposable, single-purpose tools built for a one-time need. That can be efficient for low-risk tasks, but persistent systems need proper QA, maintenance, and code reviews.

4) QA, checklists, and automation
– AI is useful for producing and running repeatable checklists (e.g., launch and onboarding flows). Claude Skills and similar features let you teach an agent a process and run it reliably.
– Use AI to automate routine, low-risk checks (binary validations like “is noindex set?”), freeing humans to focus on higher-risk judgment items.
– Always decide up front which steps are automated and which require human review; embed that in your SOPs and AI vision document.

5) WordPress plugin market implications
– AI lowers the barrier to producing small utility plugins or one-off solutions, which can depress sales for small plugin authors and reduce incentives for hobbyist contributors.
– Larger plugin makers may consolidate, focusing on defensible, complex products rather than niche one-off plugins.
– There’s a community concern: if small plugin authors and contributors drift away, we lose entry points that sustain the open-source ecosystem and the collaborative culture.

6) Experimenting beyond WordPress
– AI makes trying other stacks or static-site approaches faster and less risky. Agencies are testing alternatives for brochure sites or simple projects (e.g., static frameworks, cloud platforms).
– For some clients, a lighter stack plus AI-driven maintenance/chat may be preferable. But complex sites still benefit from WordPress’s ecosystem and extensibility.

7) Risks and cautions
– Data security: entering keys or client data into AI tools can expose sensitive information. Some vendors treat inputs in ways that make them accessible or reproducible; always check terms and retention policies.
– Overdependence on vendors: tools and pricing can change. Services available cheaply today may become expensive or restricted tomorrow; design processes so you can scale back or switch providers.
– Error handling: AI-generated tools often lack thorough validation and logging. Without careful error handling, edge cases can break things in unexpected ways.
– Human complacency: the low friction of AI can lull teams into assuming outputs are correct. Maintain a human-in-the-loop for critical decisions.

8) Likely outcomes for agencies
– Hiring and roles: routine execution may be automated, reducing demand for junior designers/developers doing repetitive work. Agencies will likely emphasize productised services, strategy, and higher-value roles.
– Productisation: AI can help agencies package repeatable processes into offerings tailored to niches (e.g., plumber-specific site builds), improving margins and predictability.
– Monitoring and QA demand: as AI automates more, investment will increase in monitoring, testing, and tools that validate and observe automated flows.

Practical advice

– Start small: connect AI to internal docs and PM tools first.
– Document an AI vision: decide which tasks should be automated and which require human oversight.
– Secure credentials: use an MCP or similar controlled bridge rather than exposing keys broadly.
– Don’t replace robust SaaS or public tools with quick, untested one-offs unless you plan maintenance and code review.
– Prepare for vendor cost changes: avoid making AI essential in a way that would cripple operations if pricing or features change.

Final thoughts

AI offers real productivity gains for WordPress agencies but needs thoughtful adoption. The best path is pragmatic: pilot internal uses, build guardrails, focus automation on low-risk, high-frequency tasks, and maintain human review where it matters. Agencies that document strategy, secure systems, and invest in monitoring will be best positioned to benefit as tools and costs evolve.

Where to find Matt

Matt participates in agency communities (Admin Bar Facebook group), LinkedIn, and runs CheckView and Inspry. He encourages agencies to experiment cautiously and to document their AI approach before scaling up.

This is the second half of a two-part conversation with Matt Schwartz, a longtime WordPress agency owner and creator of CheckView, about practical ways agencies can adopt AI — and the risks to watch for.

Why think about AI now

AI isn’t just a novelty; it can automate repetitive work, surface missed insights, and let agencies focus on higher-value strategy. Matt’s message is pragmatic: experiment, but be deliberate. Document your goals and decide where humans must stay in the loop.

Give AI access to your agency “brain”

One easy win is connecting your AI chat assistant to existing knowledge sources — project management tools, wikis, SOPs, CRM notes. When the AI can reference your living documentation it reduces hallucinations and gives staff consistent answers. Practical result: faster onboarding, fewer forgotten details, and better internal support without heavy engineering effort.

Model Context Protocols (MCPs) and guardrails

MCPs (Model Context Protocols) are emerging as a cleaner way for agents to access multiple systems. Instead of every employee creating separate connectors or exposing API keys, an agency-level MCP acts like a gatekeeper: it aggregates systems, enforces permissions, and presents a consistent interface to the AI. This reduces friction and risk, but still requires strict guardrails. Limit destructive actions (deletes, production changes), add logging, and consider third-party MCP providers carefully — you’re handing them sensitive credentials.

Vibe-coded internal tools vs. public systems

Many agencies are building quick, internal “vibe-coded” tools — dashboards, profit reports, or one-off automations — which are lower risk because they stay behind the team firewall. These can yield big operational wins: consolidated reporting across QuickBooks, time tracking, and e-commerce data; faster insights on profitability; and automation of repeatable tasks.

But caution is warranted for public-facing systems. Replacing mature SaaS or management dashboards with quickly generated, agency-built apps can be risky unless you have rigorous code review, error handling, and ongoing maintenance. A cheap one-off can become a liability if it must run reliably long-term.

QA, checklists, and Claude Skills

AI shines at repeatable checks. Use it to build and run launch checklists, automated QA for binary checks (e.g., is noindex enabled?), and to free humans for nuance and review. Claude Skills (or similar workflow-based AI features) let you teach an agent a process and define which steps need human verification. The sweet spot: automate the mundane, but require human review for high-risk items.

WordPress plugin ecosystem impact

Smaller, narrowly focused plugins are already seeing sales pressure because agencies can spin up tailored solutions quickly with AI. That dynamic may shrink the pool of small plugin authors, which worries both hosts of grassroots contributors and the broader open-source community. The likely outcome: an increased bar for plugins — more robust, differentiated features will be needed to justify monetization — and potential consolidation among larger plugin vendors.

Experimenting beyond WordPress

AI lowers the cost of learning new stacks. Agencies can more easily prototype on modern frameworks (static sites, Astro, Cloudflare setups) for certain projects. That has two effects: more fit-for-purpose choices for simple brochure sites, and a push to evaluate when WordPress remains the best tool. Open APIs and documentation give WordPress an advantage, but agencies may diversify their offerings where it makes sense.

Risks and cautions

Key risks Matt highlights:
– Data security and exposure: chat tools may claim broad rights over inputs and differ in how they treat data. Treat anything you send carefully, particularly client secrets and PII.
– Over-dependence on vendors: built processes tied to specific AI plans can become expensive if providers change pricing or features.
– Error handling and testing: AI-generated tools can miss edge cases. Without proper logging, validation, and manual review, you can create fragile systems.
– Community erosion: if tooling and quick AI fixes supplant plugin development and community engagement, the open-source ecosystem could weaken over time.

Agencies should anticipate price increases, prepare to pivot if vendor terms change, and design systems so they can function without uninterrupted AI access.

Likely outcomes for agencies

Matt expects several shifts:
– Hiring changes: routine execution work may require fewer junior staff as automation handles more tasks, while demand for strategy and oversight roles increases. However, if AI costs rise, that could swing back toward more human execution.
– Productised services: AI makes it easier to package niche vertical offerings (e.g., plumber-specific site processes) into repeatable products.
– New tooling emphasis: more monitoring, QA, and observability tools will be needed to supervise AI-driven automations and sites.

Practical takeaways

– Start small: connect AI to existing docs and PM systems for immediate value.
– Build an AI vision document: define tasks AI handles automatically versus where humans must validate.
– Consider an MCP if you’re technical and need secure multi-system access, but enforce strong guardrails.
– Use AI for internal reporting and checklist automation, but be cautious building public-facing systems without strict review.
– Monitor vendor reliance and plan for pricing changes.
– Be mindful of community health: support open-source participation where possible.

Where to find Matt

Matt is active in agency communities like the Admin Bar, on LinkedIn, and via his sites (CheckView and Inspry). He encourages agencies to experiment thoughtfully, document choices, and prioritize security and error handling.

If you run a WordPress agency or freelance business, the practical advice here can help you identify low-risk AI wins and plan for the inevitable trade-offs as the landscape evolves. For more detail and links, see the show notes at wptavern.com/podcast.