If we build an internal tool with AI, does the finished application need to use AI every time someone uses it?

Question

Tyler Smith · Accepted Answer

No. Use AI to accelerate development without assuming the finished application needs an AI model in every workflow. Many valuable internal tools are straightforward forms, approval queues, calculators, and status views.

Separate two decisions: how the software gets built, and how it behaves in daily operation. Lovable and Replit can help turn a plain-language description into an application. That does not mean a purchase approval or capacity calculation should depend on a model generating a fresh answer.

Choose ordinary application logic when the rule is explicit:
- Calculate totals using defined formulas.
- Route approvals using documented thresholds.
- Validate required fields before submission.
- Limit access using configured permissions.

Consider AI when the input is unstructured and interpretation is useful, such as summarizing a project update, extracting proposed fields from a document, or drafting a response for review. Make uncertainty visible and preserve the original material so a person can check the result.

This follows the EOS principle of simplifying systems into actionable components. The goal is a tool people can understand and use consistently, not the largest possible collection of AI features.

For Batteries Included, the useful distinction is AI-assisted building versus AI-dependent operation. Decide deliberately where each belongs.

The recommendation is to build the simplest reliable workflow first. Add a runtime AI feature only when it solves a specific problem better than a form, rule, or search function. Evaluate its errors, response time, operating cost, and fallback behavior before making the workflow depend on it.