---
title: 'Develop with artificial intelligence and verify the results'
url: 'https://grassroot.it/en/artikel/ki-entwicklung-pruefen'
markdown: 'https://grassroot.it/en/artikel/ki-entwicklung-pruefen.md'
lang: en
date: '2026-11-11'
description: 'Artificial intelligence (AI) can help formulate requirements, draft programs and explain code. For Grassroot IT, this provides another way to turn specialist knowledge into usab…'
taxonomy:
  category:
    - 'Current developments'
  tag:
    - Development
---

Current developments 11.11.2026

# Develop with artificial intelligence and verify the results

Artificial intelligence (AI) can help formulate requirements, draft programs and explain code. For Grassroot IT, this provides another way to turn specialist knowledge into usable applica…

 ![Develop with artificial intelligence and verify the results — symbolic illustration](https://grassroot.it/user/pages/03.artikel/04.ki-entwicklung-pruefen/cover.png) 

Artificial intelligence (AI) can help formulate requirements, draft programs and explain code. For Grassroot IT, this provides another way to turn specialist knowledge into usable applications. The organisation remains responsible for purpose, data and business results.

## Start with a testable task

Describe inputs, expected outputs and limits. Rather than requesting a general analysis, specify which records to include, how missing values should be handled and the expected total for a small example. This provides a benchmark for testing.

Derive expectations from the business problem. If the same system generates the code and every expected value, both can share the same error. Known correct results and justified boundary cases provide an independent reference.

## Understand and check small changes

Build the draft in manageable steps. After each, check the new processing and whether previous results remain correct. For a macro in Microsoft Excel or an application in Microsoft Access, file and database access should be understood before real operational data is used.

Code explanations support understanding but do not replace behavioural testing. Include unsuitable inputs, repeated runs and interrupted processing. Label a prototype with its known limitations until the required testing and approval process is complete.

## Consider data and operations early

Use external AI services only with data approved for that processing. Credentials, confidential business information and personal data require the appropriate safeguards and approval. Clarify rights to included components and dependence on particular services.

An application that calls an AI service during operation raises further issues: variable results, provider-dependent cost and availability, and possible human review for important decisions. Carry these questions through the [application lifecycle](https://grassroot.it/en/konzept).

## Further reading

The National Institute of Standards and Technology (NIST) describes secure development practices in the [Secure Software Development Framework (SSDF), version 1.1](https://csrc.nist.gov/pubs/sp/800/218/final). This article applies the principle of verifiable work steps to departmental development.

---

## Navigation

- Parent: [Articles](https://grassroot.it/en/artikel.md)
- Previous: [Operate applications reliably](https://grassroot.it/en/artikel/anwendungen-verlaesslich-betreiben.md)
- Next: [BAIT and DORA: the position towards the end of 2026](https://grassroot.it/en/artikel/bait-und-dora.md)