LogoValérian de Thézan de Gaussan

It was only supposed to be a prototype

ValerianValerian
November 16, 2026
4 min read
Table of Contents

I once developed a feature for a client as a prototype. It was deliberately bare-bones: we wanted to test one thing, and one thing only.

Then word got around and more people tried it. Some started complaining that it was missing this capability or that option. They were judging it as a finished feature, while we’d built it to answer a specific question.

I understood the complaints. They hadn’t necessarily been part of the original conversation. They saw something available to use and expected it to do the job.

What interests me about that experience is how easily the question can change. We start with “does this idea work?” and end up discussing “when will you add the missing features?” We haven’t necessarily answered the first question, let alone decided to build the rest.

A missing feature isn’t always an oversight

A prototype leaves things out on purpose. If I want to find out whether a different way of searching helps users find documents, I might not build saved searches, sharing, or an administration screen.

Their absence doesn’t mean I’ve forgotten them. It means I don’t yet know whether the search itself is worth pursuing.

Requests for those features are useful feedback. Perhaps the idea is valuable but incomplete. Perhaps it only helps in a narrow situation. Perhaps users can’t meaningfully test it without one of the things we left out.

Those possibilities need different responses. Automatically turning every complaint into a task can commit us to building a product before we’ve decided whether it deserves to exist.

And that’s just feedback from actual users. I’m not even talking about the manager who doesn’t use the product but wants to sound clever in a meeting: “This other product has it. We should too.” Never mind that the other product solves a completely different problem. I’d want to hear whose problem the feature would solve before spending even five minutes on it.

With agents, it’s easy to skip that decision

This isn’t new, but AI makes it particularly easy to drift into.

When a request takes five minutes to implement, I can be tempted to just do it. Then another request comes in, and that one is easy too. Soon I’m maintaining something much broader than the experiment I intended to run.

Each addition can also make it harder to stop. People save information, share links, and adapt their work around the feature. Removing it now means asking them to undo that work.

The fact that I can afford to implement the next request doesn’t tell me whether I want to take responsibility for the resulting product.

Give the experiment an ending

The design sprint described in Sprint, by Jake Knapp, John Zeratsky, and Braden Kowitz, gives this a useful structure: choose a problem, build a focused prototype, and test it with customers within five days. The prototype is there to help answer a question, not to quietly become the first release.

I’d rather define the trial before sharing the prototype: what we’re trying to learn, who is trying it, and when we’ll review what happened. People should also know whether their data will be kept and whether the tool might disappear.

For the search example, that could mean asking a small group to use it for two weeks on real tasks, then reviewing where it helped and where they returned to their old method. We’d consider feature requests alongside those observations, rather than treating the request list as the next development plan.

At that point, we can decide to stop, run another focused experiment, or commit to developing a supported feature. If we continue, there’s a reason for doing so beyond “it’s already there.”

A “prototype” label alone won’t accomplish that. If we leave it available indefinitely and let people build their work around it, we’re encouraging a dependence that the label doesn’t excuse.

I still want to test ideas quickly. But I want to keep the option to conclude that an idea isn’t worth pursuing. Otherwise, every afternoon spent experimenting risks becoming another application I’ve accidentally agreed to maintain.

This is the fourth of four follow-up articles to Cheap to build does not mean disposable.