-
Give The Agent A Decision Ladder
Before letting an agent add files, dependencies, or abstractions, make it answer a short sequence: does this need to exist, can the platform do it, can the standard library do it, is a current dependency enough, can it be one line. Put that ladder in your project instructions or coding agent rules. This turns “be simple” from a preference into a repeatable gate.
-
Prefer Native Features First
Many AI coding tools reach for packages because packages show up often in training data. For small business apps, the browser, framework, database, and standard library already cover a lot. Ask the agent to check native capabilities first, such as built-in dialog behavior, form validation, date formatting, file APIs, or database constraints.
-
Require A Deferral Note
When the agent chooses the simpler path, have it leave a brief note explaining what it skipped and why. That note becomes a debt ledger, not a vague TODO. Later, if the simple version fails, you know exactly which library or abstraction was deferred and what trigger would justify adding it.
-
Constrained Agents Often Follow The Real Task Better
Less code is not only about lower token bills. A constrained agent spends less effort decorating the solution and more effort satisfying the core requirement. For builder workflows, this means prompts should define the outcome first, then explicitly penalize unnecessary structure, styling, and dependencies.
-
Plugins Help, But The Rule Matters More
A plugin can make this workflow persistent with audits and project-level enforcement. But you can get much of the benefit today with a direct instruction like: “Follow YAGNI. Use native platform and standard library features before adding dependencies. Explain any new abstraction.” The value is not magic, it is making restraint part of the build loop.
Why it matters
Small businesses do not need impressive code volume. They need working tools that are cheap to maintain, easy to inspect, and quick to change. A decision ladder keeps AI-assisted projects from turning simple internal workflows into fragile mini-platforms.