Earlier business notes
AI coding: a prototype still needs an owner
Questions to ask before an AI-assisted prototype becomes software people rely on: testing, access, data and maintenance.
By Rajiv Sankarlall · Updated
Getting a working screen from an AI coding tool can make an idea feel suddenly real. It gives you something to try and discuss. It also leaves questions that a successful demo may not answer: what happens with unexpected input, who can access the data and who will fix the software later?
This is a technology note for founders exploring prototypes. The coaching on this site focuses on burnout, values and purpose. AI development and AI employee installation are separate from that offer.
Define what the prototype is for
Before building, write down the question you want it to answer. Are you testing whether customers understand an idea, whether a workflow makes sense or whether an integration is feasible? Keep the prototype small enough to answer that question.
Use sample data where possible. A demonstration for a few people does not need the same access to information as a production system. Decide what the prototype should never do, such as send real messages or change customer records, and enforce those limits in the setup.
Test the parts that are easy to overlook
Try missing fields, repeated submissions, expired sessions and a user attempting to view another user's records. Check what happens when a connected service is unavailable. A screen that works once under ideal conditions gives you limited evidence.
If you do not have the technical knowledge to review those behaviors, involve someone who does before making the software available to people who depend on it. For security review, OWASP's GenAI Security Project covers risks associated with AI applications.
Decide who will maintain it
Keep the code, configuration and setup instructions in a place the responsible person can access. Know how to restore a previous version and who can respond when something breaks. Check ongoing costs and dependencies before turning a trial into a commitment.
You may decide the prototype has answered its question and should stop there. That is a useful outcome. You may also decide the idea merits a proper build. Make that choice from what you learned rather than the excitement of having shipped a demo.
For a related question about ownership, see why procedures still need people who own the work.