Buck Dossey

ETL dev turned solution consultant turned account manager. Wrote C# and SQL in production. Now helping sellers build AI-powered GTM tools that don't fall apart.

[Buck Dossey @buck_dossey]

New Codex connector for ZoomInfo launched today. Start building -https://t.co/qMX7O55WQi

[Buck Dossey @buck_dossey]

Salespeople are vibe coding now. That's exciting. But most are building on unstable ground.

The most common mistake I see from real sellers in the field: trying to build everything at once. One giant prompt, one tangled app, and when something breaks you have no idea where to look.
The fix is first principles thinking and the single responsibility principle. Each component does one thing. You make sure it works. Then you connect them.

Step one: don't write your own prompt. Ask the AI to write it for you.
Before touching a single feature, tell the AI what you're trying to build and ask it to generate an architecture prompt that follows industry best practices for a single-user app.
Like this:

"I want to build a [describe your app] for personal use. I'm the only user. Please write me an initial architecture prompt that follows industry best practices for a single-user application, including how to structure the frontend, backend, and database as separate layers. I want to understand what each layer does and why."

That response becomes your blueprint. You're not guessing at structure. You're starting with something that reflects how real software gets built, scaled appropriately for what you actually need.

Step two: build and verify each layer independently.
Don't connect anything until each piece works on its own.
Build the database layer. Test it. Build the backend logic. Test it. Build the frontend. Test it. Then wire them together.
This sounds slower. It isn't. Debugging a tangled app takes ten times longer than verifying three clean components.

Why single-user matters.
Most architecture advice assumes you're building for thousands of users. Authentication systems, load balancers, complex permission models: none of that is necessary when you're the only person using the app.
Telling the AI upfront that this is a single-user tool keeps things appropriately simple, and simple is easier to maintain, easier to improve, and much less likely to break.

You don't need to become an engineer.
But if you're building tools to support your GTM motion, having a mental model for how your app is structured will save you hours of frustration.
Don't boil the ocean. Use first principles. Make each piece work before you connect it.
That's the whole skill.

[Buck Dossey @buck_dossey]

Genuinely, what an incredible time to be in go-to-market. AI is changing how revenue teams operate, faster than most people realize. I'm here to share what's actually working, with real examples from the front lines.
More soon.