Most operators are renting their tools.They don't realize it yet.

When I signed up for Beehiiv, it looked like the right move. Clean app. Nice email composer. Recommended by other operators. Free at my scale. The kind of choice that feels obvious.

A few days in, I tried to do something interesting.

I wanted each new subscriber to receive a different first paragraph, generated by Claude based on the page they signed up from. I wanted my analytics layer to talk to my subscriber list. I wanted an AI agent to read each issue's performance every week and propose what to do next. None of these are exotic ideas. They are exactly the kind of thing AI-native operators should be building.

Beehiiv said no.

Not directly. It said no by hiding its API behind a paid tier I didn't qualify for. By keeping subscriber data in its walls instead of mine. By forcing me to compose every email inside its UI, which my AI agents cannot use. By being a platform.

The problem with platforms is not their features. It's that they own the relationship between my work and my data. Everything I built inside Beehiiv would have been Beehiiv's. The day I wanted to leave or do anything Beehiiv didn't anticipate, I would have discovered that "ownership" was always a feeling, not a fact.

So I left.

I migrated to Resend, the email-sending tool. My subscribers now live in my own database. My code lives in my own repo. Every piece is independently swappable. Nothing owns anything. And every time I add a new tool, those pieces compose with everything else, because they all speak the same language. Claude for AI. Vercel for compute. A generation API for visuals. An error monitor for reliability. Each one a single-purpose tool, each one composing with the rest.

A tool does one job and exposes a clean way for code to operate it. A platform bundles many jobs and only opens up the parts that do not threaten its lock-in.

The choice between them used to be an architectural detail buried in technical documentation. In the AI era, it is the choice that decides whether your business scales with agents or stalls at one human.

The smartest operators in 2026 are quietly stepping out of platforms and into the tools they own. They are done renting their work. They want to own it. They have noticed that AI agents can extend a stack of owned tools into a system. AI agents cannot extend a platform-based stack at all.

This is how CGA Creative is built.

The whole stack is owned tools. Vercel for compute. Supabase for data. Anthropic for AI. Resend for email. Stripe for payments. GitHub for code. Six pieces, each independently swappable, all composing into anything I want to ship.

This past week alone, the stack let me ship a public AI diagnostic, an automated newsletter pipeline (which delivered this email to you), an error-monitoring layer, five scheduled scanners, and a synthesis agent that emails me a daily brief on what's happening across the system. Every one of those was built by composing those tools. None of it would have been possible inside a platform.

This newsletter is the field notes from one week of building CGA in public. Each issue, three things:

The week's Sunday Lab tool, with the framework behind it. This week's tool is the Unique Lens Diagnostic, which I launched this morning. Next Sunday's is the Layered Interest Audit.

What I learned shipping it. What worked, what broke, what surprised me.

One operating principle I am willing to commit to publicly. This issue's was tools over platforms.

Forward this to anyone you think will recognize themselves in it.

Welcome.

Erich