Internal tools are a classic engineering time sink. Every team needs one, nobody wants to maintain one, and the backlog of small requests never ends. Generating the first version with AI changes that equation, but only if you treat the generated app like real software. This post covers a practical workflow for engineers.
1. Write the Spec as Data, Not Screens
Start with the data model. Engineers know that screens follow from data, and AI builders work the same way.
| Entity | Fields | Relations |
|---|---|---|
| Asset | name, type, serial, status | belongs to Location |
| Location | name, address | has many Assets |
| Booking | start, end, purpose | belongs to Asset and User |
| User | name, email, role | has many Bookings |
Add the roles and what each role can read and write. Permissions written up front avoid painful refactors later.
2. Generate the Application
Paste the data model, roles and core workflows into an AI Web App Builder. It generates a full-stack app: database schema, authentication, role-based access, CRUD views, filters and dashboards. Review the generated schema first. If an entity or relation is wrong, fix the description and regenerate rather than patching screens.
3. Test Like You Would Any Codebase
- Role tests: sign in as each role and attempt actions that should be forbidden.
- Edge cases: overlapping bookings, deleted assets with active bookings, empty states.
- Data import: load a realistic dataset and check performance on list views and filters.
Keep a short test checklist in the repo or wiki and run it after every significant change.
4. Iterate in Small, Reviewable Changes
Request one change at a time, for example "prevent bookings that overlap an existing booking for the same asset". Small changes are easier to verify and roll back. Track each change in a changelog so the team knows what the tool does today.
5. Decide What Stays Generated
Many internal tools never need hand-written code. When a requirement does exceed what the builder handles well, such as a complex integration with a legacy system, own that part explicitly and document the boundary.
Checklist Before Rollout
- Schema reviewed and matches the spec.
- Permissions tested for every role.
- Seed data imported and spot-checked.
- Backup and export confirmed.
- Owner assigned for future change requests.
Conclusion
Generating internal web apps is not about skipping engineering discipline. It is about applying that discipline to the spec and the tests, and letting AI handle the repetitive scaffolding in between.