Prototypes are not that helpful anymore

Open original

Elliot Morris argues that prototypes used to force engineers to communicate clearly, surface real problems, prove competence, and engage with reality, but LLM-built prototypes break most of those benefits. He thinks the useful future prototype is a sharp, single-thesis experiment rather than a big, sprawling demo.

  • Prototypes once delivered four bundled benefits: clear communication, early problem discovery, a visible competence signal, and forced contact with reality. LLM prototypes keep only the first of these reliably.
  • LLM output tends to drift toward a generic "average" answer, so the builder can end up communicating something they didn't intend without noticing.
  • Experienced engineers make rougher, smaller prototypes because they test one narrow thesis. Morris suggests the key question is now "what specifically does this prove, and why should I care?"

Elliot Morris's core argument is that prototypes used to be one of the best cheap tools an engineering org had, and LLM-generated prototypes have eroded most of what made them valuable. He lays out four benefits of a traditional prototype: it let an engineer communicate an idea clearly, it exposed the technical and political problems a direction would hit, it proved to the org that the engineer understood the domain (and gave them a way to win projects and build a reputation), and it forced contact with reality before people made confident claims. He calls it a "wash your hands before surgery" practice, a rare case where the cost was low and the payoff was almost universal. The point he keeps returning to is that prototypes worked because building one was hard enough that doing it implied real competence, so everyone trusted the artifact. The prototype itself was never really the point. Morris goes through each benefit against the AI-prototype case and finds them mostly weakened. For communication, LLM prototypes are fine, even better than napkin sketches for showing interaction, but they drift toward a bland "LLM-average" and quietly make decisions for you, so the skill now lies in constraining the model rather than prompting it to do more. For surfacing problems, you can survey many options shallowly, but you lose the organizational friction that used to show whether an engineer could get a server provisioned or get past three layers of management, and a junior person can't judge how serious a problem is in their own context. The competence signal is mostly gone, since "they built this" no longer implies much, and the reality check is weakened because tools that hide reasoning and push auto mode let people skip the contact with reality that used to catch their mistakes. He notes that some junior engineers are now confidently asserting things that fall apart once they meet the organization, and that colleagues rarely say so out loud. The closing material offers a few pointers and a theory about why experienced engineers' prototypes tend to look rougher and smaller. Morris suggests that seniors have less time and less to prove, so they skip the polish. He also argues that people who have been getting results from prototypes are, in effect, spending down the legitimacy the practice built up, and that this will run out. The takeaway for a reader is that a prototype's value now depends almost entirely on the person behind it, and that the signals orgs once read for free have to be rebuilt some other way.

Ask about this article