I wrote an article about how the Convergence Profile took shape — the thinking, the dead ends, the moment I had five different products that were somehow all the same idea. That was the design story.
This is the other story. The one about actually building the thing. With tools I had mostly never touched, a codebase I did not write by hand, and a deployment pipeline that felt suspiciously easy. Let me walk you through what actually happened.
First, a confession
I am not a developer. I have an IT background — years of it — which means I know what a deployment pipeline is, I know what OAuth does, and I know that “it works on my machine” is not a shipping strategy. But I do not write code professionally. I have opinions about code. That turned out to be enough.
There is a gap between “I use AI tools” and “I build with AI tools.” Most people are in the first category. ChatGPT for drafts, Claude for thinking, Perplexity for research. That is my AI Stack, and it works great for knowledge work.
Building a product? Completely different game. I needed a build stack. Here is mine — six tools, each with a very specific job, and a few stories about what went wrong along the way.
The build stack
Claude Code
The one that wrote the code
Not Claude the chatbot. Claude Code — the command-line tool that writes, edits, and iterates on actual code. This is where the app got built. I would describe what I wanted, it would write it, I would test it, we would go back and forth.
The conversations were wonderfully mundane: “The guide selection grid breaks on mobile.” “The progress tracker does not update after the Root conversation completes.” “Add error handling for when the API key runs out of credits mid-conversation.”
What made it great: plain language in, working code out. What made it maddening: long sessions where it would fix one thing by cheerfully breaking something we had already solved three hours ago. I started keeping a CLAUDE.md file in the project — basically a note taped to the fridge that says “do NOT touch the authentication flow, we fixed that already.”
Cowork + Claude Desktop
The thinking room
Before a single line of code existed, everything was shaped here. The spec, the brand framework, the guide roster, the build plan — all of it came out of long conversations with Claude through the desktop app and later Cowork.
This is where the existential questions lived: “Is five lenses too many?” “Should the Tension Meter lead or follow?” “What frameworks have IP problems and will get us sued?”
The distinction matters: Cowork and Claude Desktop are where I think. Claude Code is where I build. Mixing those up is like trying to do architecture and plumbing at the same time. You will flood something.
Anthropic Console
The electricity bill
Every time someone talks to a guide in the Convergence Profile, that is a Claude API call. The Anthropic Console is where I manage the API key, monitor usage, and watch the costs.
It sounds boring. It is boring. It is also the thing I check every morning now, because a full Convergence Profile — five conversations plus synthesis — costs roughly 15 to 40 cents. That is nothing when three people test it. It becomes a conversation when someone shares the link on LinkedIn and 200 people try it on Tuesday.
The Console went from “admin tool I glance at occasionally” to “dashboard I stare at like a stock ticker.”
Vercel
The “wait, that is it?” moment
The app is a Next.js project. Vercel is where it lives. Push code, it deploys. Environment variables stay server-side. Free tier covers everything.
For someone who has spent years in IT dealing with server setups, firewall rules, and deployment pipelines designed by people who hate joy — this felt almost illegal. Push a commit, get a live URL. No tickets. No change advisory board. No waiting three weeks for someone to approve a port opening.
I kept refreshing the page waiting for the catch. There was no catch.
Resend
The one that sends email without making you cry
The feedback form needed to actually deliver emails. Resend handles transactional email — simple to set up, no SMTP wrestling, no enterprise marketing platform that requires a two-day onboarding.
If you have ever configured an enterprise email service, you know the special kind of despair that comes with DKIM records and SPF alignment. Resend skips all of that. It just works. Revolutionary concept.
Google Cloud Console
The familiar pain
Authentication. OAuth. Credentials management. Identity profiles. This is the one that felt most like traditional IT work — and I mean that in the way you remember a difficult ex. Familiar, complicated, and you are never fully sure it will not break at 2 AM.
If you have ever set up SSO or identity providers in an enterprise environment, congratulations: you already know how this feels. Same energy, different logo.
What the flow actually looks like
The AI Stack page on The Overview shows my thinking tools. But the build stack has its own rhythm:
Design (Cowork / Claude Desktop) — Figure out what to build. Write the spec. Argue with the AI about whether five lenses is too many. Lose that argument. Keep five lenses.
Build (Claude Code) — Write the app. Iterate fast. Fix what breaks. Fix what the fix broke. Fix what fixing the fix broke. Ship it anyway.
Connect (Resend + Google Cloud) — The infrastructure that makes the app talk to the real world. Email, authentication, the boring parts that matter the most.
Ship (Vercel) — Deploy. Get a URL. Text it to someone. Refresh obsessively.
Monitor (Anthropic Console) — Watch the costs. Check the usage. Pretend you are calm about it.
What surprised me
The code was the easy part. Setting up Next.js, writing React components, getting Tailwind to look right — Claude Code handles that well. The hours I lost were on the edges: environment variables not loading, API routes timing out, authentication redirects failing silently. Infrastructure problems, not code problems. That felt very familiar from my IT career, actually. The system always works until it meets reality.
You still need to be the architect. I could not have built this without Claude Code. But I also could not have built it by just saying “make me a personality tool.” Every useful prompt came from understanding the spec, the data flow, the edge cases. The AI writes the code. You need to know what to ask for — and more importantly, what to refuse.
Context management is a real skill. Both with Claude Code (keeping it aware of the full project) and with yourself (keeping track of which tool does what, which conversation had the key decision, where that one config value lives). CLAUDE.md went from afterthought to the most important file in the project.
Six tools is a stack. Not a team. Not an agency. Six tools and a clear spec. That is what it took to put a working product online — as someone who understands systems but does not write code professionally. That is not a replacement for developers. It is a new capability for people who understand problems and know how to describe them precisely.
The Convergence Profile is live. People are using it. And the stack that built it fits in one article with room for bad jokes.
Was this worth your time?