How I Use Two AI Agents to Build SaaS Without the Bloat

AI is great at writing code quickly — the problem is it often writes too much of it. A simple feature can quietly turn into extra interfaces, dependencies, services, and abstractions for problems you may never actually have.

To avoid that, I use two AI agents when building my SaaS.

1. The Architect

The Architect creates the plan. I give it a feature, bug, marketing task, or structural change, and it reviews the database, routes, UI, dependencies, and edge cases before suggesting what needs to be built.

The plan is usually detailed — often more detailed than it needs to be. The Architect designs the complete system, but as a solo founder, I need the smallest system that solves the problem.

2. Ponytail

Before I start building, I send the plan to Ponytail — a cynical senior developer persona with one rule:

The best code is the code you never write.

Ponytail challenges the plan:

• Do we need this now?
• Can a native browser feature replace a new package?
• Can three interfaces become one function?
• Are we solving a real problem or preparing for an imaginary one?

Where the Architect might suggest a state-management library for a basic form, Ponytail replaces it with standard HTML form data. The Architect expands the plan; Ponytail cuts it back down.

Why this works

The Architect helps me see the whole problem. Ponytail keeps me from building more of it than I need. Together, the workflow looks like this:

1. Create a detailed plan.
2. Challenge every part of it.
3. Remove speculative work.
4. Build the smallest version that solves the problem.

It gives me the speed of AI without letting the codebase spiral into unnecessary complexity, and I use the same process for landing pages, marketing copy, and bug fixes.

Curious to see what I'm building with this approach? Check out DataCurlew.