How Frog Stone Media Runs Its Own Businesses on AI

Who it helps

Small-business owners who want AI to take routine work off their plate while still working the way their business works. Frog Stone Media runs its own publications and tools on AI, using the same setup it builds for clients: a separate AI working environment for each business, with a shared operations layer underneath that keeps every site checked, maintained and safe to update. This page is the working example behind the AI architect service.

What the setup does

AI that follows how each business works

Each project gets its own AI working environment, set up with that business’s rules, its writing voice and its connections to the tools it uses. AI working on a grocery price site follows grocery rules. AI working on a car publication follows that publication’s editorial standards. Each environment is kept separate, so one business’s habits stay out of another business’s work.

Shared operations behind every site

• A daily maintenance check runs across all the sites, and a weekly digest lists anything that needs attention.
• Scheduled jobs run on self-hosted machines, so routine work runs without anyone having to start it.
• The connections that let AI use outside tools are written in-house or audited before they touch any account. The MCP servers page covers how those connections are built and checked.

Safer updates

Websites are built from outside code packages, and a tampered package is one of the ways a site gets compromised. Every deploy uses three safeguards: dependency versions are pinned, install scripts are turned off, and new package versions wait out a cooling-off period before they are adopted. Together these are designed to keep a bad package from reaching a live site.

Content checked by tools and by a person

Content runs through automated voice and fact checks. Then a person reads the whole piece before it publishes. The automated checks catch the patterns they were built for, and a person checks accuracy, context and business fit.

Build, review, approve

Claude Code builds. OpenAI Codex reviews. Human verified and approved. A person reviews the work and how it fits the business before it goes live. The full process is on the methodology page.

What it does for the business

• Smoother operations. Routine checks and jobs run on a schedule, and anything that needs a decision lands in one weekly digest.
• Revenue growth. Work that protects income, like keeping affiliate links current and content fresh, gets done on a regular cycle instead of when someone finds time.
• Happier customers. Content stays accurate and in the business’s own voice, because a person reads it before readers do.

Examples from Frog Stone Media’s own work

• SacGroceries. AI reads store flyers and customer photo submissions to pick up prices, and a person approves each change before anything publishes.
• The Weekly Driver. AI is used only in the reliability database and recalls. It helps classify federal complaint and recall text and drafts buyer-guide summaries, and a person approves each summary.
• MK Library. AI is used only to manage affiliate links, suggest internal links, and spot articles that have gone stale and need an update.

What this means for your business

The same setup gets built around your business: an AI working environment with your rules, your voice and your tool connections, shared checks running behind it, and a person approving the work before it goes live. Scoping starts with a free email exchange, with no sales call. See how the AI architect service works, or send a note about your business.

For your technical team

• Each repository gets its own Claude Code environment: project rules, skills, subagents and hooks, set up for that project.
• A shared workspace layer runs a daily maintenance cron with a weekly digest, plus periodic Actions pinning and configuration audits.
• Self-hosted GitHub Actions runners on always-on Macs handle scheduled and browser-bound jobs.
• MCP servers are written in-house, or forked and audited before use. Credentials load from project-local environment files.
• Supply chain: committed lockfiles, npm ci --ignore-scripts in CI, a cooldown on new versions through npm-check-updates, and GitHub Actions pinned to commit SHAs.
• Editorial gates: automated voice and pattern scanners, followed by a required hand pass.
• Review: OpenAI Codex reviews the changes Claude Code writes, and a person approves before merge and deploy.