Anthropic released in-window prototyping 1, making prototyping faster and more universal. Users can now chat to describe their idea and Claude will create a prototype in the browser window. These HTML/Javascript prototypes allow the user to iterate with Claude and, once it’s working the way they want, copy the Claude-generated code.
Rapid prototyping has been around for a long time, but the term was mostly used in manufacturing and culminated in today’s 3D printing techniques. In software, the concept took hold primarily in the design of user interfaces, at a stage before devs were involved 2. This was a helpful step in confirming functionality before starting the more expensive step of software development.
Claude’s prototypes are doing both, allowing the user to try the functionality while also writing the code. In my use, I found Claude was also doing well at thinking through use cases. I had it create a prototype for an SMS-to-ticket app. Claude identified workflow ambiguities, like at what point messages from the same number should create a new ticket.
In the pre-AI world, we wrote requirements and specifications to keep track of success criteria. This kept business and software engineering on the same page. With Claude able to keep track of functionality and generate the code, this documentation seems unnecessary. Truth be told, in fast-moving engineering environments, it was already rare for engineering teams to refer to requirements or specifications. The only problem pre-AI was reading through the code to figure out what the code would do in different scenarios.
A new model of software development is:
- Business request
- Prototype ready in staging environment
- Business review (UAT, essentially)
- Revisions deployed to staging
- Final review, deploy to prod
In this new model, like before, the truth of the software’s behavior is recorded only in the code.
1 Prototype AI-Powered Apps with Claude artifacts
2 Rapid prototyping in Agile development
