The challenge
Average Robot, an AI consultancy co-founded by our founder, Matt Perry, launched its website on a no-code AI builder. That was the right way to start and the wrong way to grow.
- Growth waited on the tool. Each new article needed a manual redeploy before search engines could read it.
- The business didn’t own its data. The back end sat in an account the business could never take over.
- The ambition outran the team. It wanted a website that publishes like a media team, a podcast it controls, and a content engine it could also run for a client, without building an engineering department.
What we did
- Strategy: moved the website, the code and the data onto a platform the business controls, so growth no longer waits on a tool.
- Website: grew it into a publication that search engines and AI assistants can read page by page: 69 articles, 27 guides and 21 comparison pages.
- Content engine: built a system that finds topics, drafts articles, checks them against the house rules, publishes them and shares them to social.
- Podcast: launched a show on a feed the business owns, with four episodes released.
- Platform: made the engine serve more than one business. It now serves two businesses, each kept separate.
- Delivery model: set up a team of 17 specialist AI agents for research, writing, data and engineering. Our senior lead sets the direction and approves what ships.
The result
- Speed: 528 reviewed changes shipped in the first 60 days on the new platform.
- Reach: 160+ live pages, each one readable by search engines and AI assistants.
- Ownership: the website, content, social and podcast run on a platform the business owns, with no engineering department.
Adding a channel or a client is now a brief, not a hiring plan.
What this means for you
If your team has AI tools but isn’t getting this kind of output, what’s missing is direction, not software. A Fast Start ships your first result within two weeks. A Program builds one growth capability with your team. A Custom Build delivers a platform like this one, in your accounts and owned by you.
Under the hood
- Stack: React and Vite on Cloudflare Pages; Supabase Postgres with about 90 serverless functions; GitHub Actions.
- Models: Anthropic Claude for writing, review and code, with OpenAI and Perplexity wired in.
- Built with: Claude Code, 17 specialist agents, 13 scheduled tasks, MCP connectors to the CRM, books, mail and calendar.
- Scale: 150,000+ lines of code across its whole history, 222 database migrations, 160+ live pages.
- Quality: 185 test files and 7 automated checks on every change, including one that keeps each client’s data separate.
- Who built it: one senior lead and AI. Since August 4, 2026, 1,142 commits came from our lead’s account, and no other person contributed code.
- Integrations: Google Analytics, Search Console, CRM, email, social scheduling, SEO data and WordPress.
- One lesson: when a model upgrade quietly changed a default and article drafts began failing, the fix was one line. The engine now flags any output it doesn’t recognize.
