Prototype

A prototype is a first, deliberately unfinished version of something one wants to build. This can be a device, an app, a website, or an engine. The purpose is not sale, but learning. One wants to find out whether an idea works at all before much money and time flow into it. That’s why a prototype is allowed to be ugly, run slowly, and in many places not be able to do anything yet. What matters is only that it answers exactly the one question it was built for.

Finding mistakes early instead of expensively

A mistake costs more the later it is discovered. Changing a sketch takes an hour. Changing an already produced component affects tools, warehouses, and supply contracts. In industry, one roughly calculates that the cost of a change multiplies with each project phase. A prototype shifts failure forward, to where it is cheap.

The second reason is feedback from real people. What sounds logical at a desk can be completely impractical in the hand. Only once someone actually uses a thing does it become clear which steps are annoying and which buttons nobody can find. Such insights cannot be replaced by thinking.

Prototypes are also central for investors. A startup that only shows a presentation is selling a claim. A startup with a running prototype shows proof. That’s why financial news often features phrases like “first functional prototype presented.” For investors, this is a signal that a risk has become smaller.

From paper model to running sample

Prototypes come in very different levels of maturity. At the very beginning there is often a paper model: drawn screens that one lays in front of someone and asks where they would tap. Then comes a click model, where the interface looks real but nothing happens in the background. Only later does a sample emerge that actually computes or moves.

A common trick is faked technology. With a voice assistant, a person in the next room can type the answers while the test subject believes they are talking to a machine. That sounds like deception, but it’s clean research: one tests the idea without having already built the elaborate technology. In technical language this is called a Wizard-of-Oz test, named after the man behind the curtain.

It is also typical for a prototype to be deliberately built poorly. The code contains shortcuts, fixed values instead of real calculations, no safety checks. This is called throwaway code, because it gets replaced later. This is exactly where a common danger lies: if a prototype is well received by management, it’s often supposed to go straight into sales. The stopgap then becomes a product that remains unstable for years.

Prototypes in AI products and in the news

In the AI world, you encounter this term constantly. New chatbots and image generators first appear as a preview or beta, i.e., as a publicly tested prototype. The providers openly say that errors can occur, and evaluate usage as a test. One advantage of modern language models is that a rough prototype can be built with them in a few hours, for which one used to need months.

In the automotive industry and with chips, on the other hand, a prototype is a milestone with a date and a price. A first test chip quickly costs millions, because dedicated photomasks are needed for it. News about a successful first test run therefore regularly move stock prices.

You should distinguish the prototype from two similar terms. A proof of concept only checks whether something is technically possible at all, and is usually discarded afterward. A minimum viable product, on the other hand, is already a real, sold version with few features. The prototype lies in between: more than a test, but not yet a product.