Prompting-as-Prototyping
Validate a product idea by prompting a model in a local browser before building anything
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 3
- Confidence
- 87%
Instead of moving from PRD to design to build, use raw prompting as the prototyping medium. Because a general model can already approximate many behaviors, you can mock up a feature by prompting it, demo the result, and gather reactions in hours — deciding whether to build before committing engineering time. Nguyen calls prompting a new way of doing product development and prototyping for designers and PMs.
Origin
Karina Nguyen developed this while building the 100K-context file-upload feature at Anthropic and wrote a blog post on it; she cites it as one of the biggest ways AI has changed the product builder's job.
Core principles
- 01A general model is a working prototype you already have — prompt it rather than spec it.
- 02The demo is the artifact that wins buy-in, not the document.
- 03Small 'micro experiences' are often just a well-designed prompt away.
- 04Prototyping this way is now the fastest path from idea to something people can react to.
How to run it
- 1
Prompt the behavior in a local/browser sandbox
Take your product idea and prototype it purely by prompting a model locally — no training, no backend. Get it to approximate the target experience.
- 2
Demo it to real people and read the reaction
Show the prompted prototype to colleagues or target users. Strong pull (people asking for an API, wanting to keep using it) is your signal to build.
Pro tip Watch for unsolicited demand — when people spontaneously ask 'can I get this as an API?', the idea has validated itself.
- 3
Promote validated prototypes into real features
Only the prototypes that earn a reaction graduate into engineered features. The rest cost you only a prompt.
Pro tip Great candidates are micro-experiences — e.g. auto-generating conversation titles by sampling the user's last five conversations to match their style.
In the wild
When launching 100K context, Nguyen prototyped the file-upload experience by prompting the model in a local browser and demoed it. People loved it and immediately asked for a file-upload API.
→ The demo's reaction validated the feature and clicked the realization that prompting is a new product-development medium; it shipped and became a feature she says people use daily.
To generate conversation titles matching a user's style, the team took the user's five latest conversations, asked the model for the user's style, and generated titles in that style for new conversations — prototyped by prompting.
→ A small, well-loved micro-experience shipped in Claude.
Common mistakes
Writing a PRD and design before testing feasibility
Investing in specs and designs before you know the model can deliver the experience wastes cycles. A prompted prototype answers the feasibility and desirability questions in hours, so do it first.
Ignoring the strength of the demo reaction
The prototype's value is the signal it generates. Building something that got a lukewarm demo response — while skipping ideas people spontaneously asked to keep — means optimizing effort against the wrong evidence.
Is it for you?
Best for
Designers and PMs who want to test AI feature ideas quickly and build the case for investment with a live demo
Not ideal for
Features whose value depends on capabilities the current model genuinely lacks (deep expert knowledge, reliable long-horizon agency) where a prompt can't fairly approximate the experience
From the transcript
“I was just like prototyping this in the local local browsers. I did the demo and like people really really loved it”
“prompting is like a new way of like product development and like prototyping for designers and for like product managers”
“we we took like five latest conversation from the user and like ask the model like what's the style of the user”
From the episode
OpenAI researcher on why soft skills are the future of work
Karina Nguyen (Research at OpenAI, ex-Anthropic)