#  AI-built Prototypes 

 



 ##  

  expand\_more  

 
  

 

## An Emerging Demand

It's a genuinely exciting moment for how ideas get built at Harvard. What used to take weeks, documenting a concept, briefing a designer, iterating through rounds of mockups, can now go from idea to a working, tangible prototype in an afternoon.

LLM tools have compressed that cycle dramatically, and that benefits everyone: departments can express their vision faster and more concretely, and by the time HWP gets involved, we're often working from something tangible instead of a written description.



 

##  I built a website or app using AI tools. Can I host it with HWP? 

AI-generated prototypes are not the same as HarvardSites websites. HWP cannot simply upload and host exported AI-generated HTML, CSS, JavaScript, or app code.

However, we can help determine the best path forward. There are two general paths forward.



 

## Path 1: Rebuild it in HarvardSites

**Recommended when the prototype is primarily a website, landing page, or content experience.**

   ![Path 1 process](/sites/g/files/omnuum13026/files/styles/hwp_1_1__720x720_scale/public/2026-09/Option1_6.png?itok=8ND8w4eE) 

 

Your AI-generated prototype becomes a reference for understanding the goals, audience, content, structure, and visual direction of the project. We then translate those ideas into an accessible, maintainable HarvardSites site using available components and layout patterns.

- We review the prototype to understand its goals, audience, content, structure, and features.
- We identify the HarvardSites components and layout patterns that can support the experience.
- We rebuild the site using Harvard’s supported platform, design system, accessibility standards, and security practices.
- The final site may not be a pixel-for-pixel reproduction. The goal is to preserve the important ideas while making the result maintainable, accessible, and aligned with Harvard standards.
- We help coordinate with the appropriate School or Unit request process, which varies by area.

**This is the path HWP generally recommends when HarvardSites can support the site’s functionality and content needs.**



 

## Path 2: Host it elsewhere and link to it

**Appropriate when the prototype must remain intact or functions as a custom application.**

   ![Path 2 process](/sites/g/files/omnuum13026/files/styles/hwp_1_1__720x720_scale/public/2026-09/Option2_1.png?itok=TgM7hBYw) 

 

In this path, the AI-generated site is hosted by the School, Unit, or another approved provider. The HarvardSites site can link to it using a standard component, such as a Call to Action Link or a 50/50 Card component.

- The prototype remains intact and is hosted outside HarvardSites.
- The School or Unit is responsible for selecting, approving, hosting, securing, and maintaining the external site.
- HWP can help place a link to the external site on a HarvardSites page using an available component.
- HWP does not host, secure, maintain, or provide ongoing support for the external application.
- The external site must still meet applicable Harvard accessibility, privacy, security, content, and governance requirements.



 

##  Before anything is published 

In both cases, your School or Unit governance group makes the call on which approach fits. Regardless of where it's hosted, any content that represents Harvard must align with Harvard's [Digital Accessibility Policy ](https://accessibility.huit.harvard.edu/digital-accessibility-policy)and Privacy and Security [policies and standards](https://privsec.harvard.edu/policies-and-standards). Hosting something outside HarvardSites does not exempt it from University policies.

Regardless of which path applies, any AI-generated content, images, files, or design recommendations must be reviewed by the site owner before publication. Harvard's [generative AI guidelines](https://www.huit.harvard.edu/ai/guidelines) note that AI-generated content can be inaccurate, misleading, fabricated, or contain copyrighted material, and that you are responsible for content you publish or share.

The guidelines also state that confidential data, including Level 2 and above data, should not be entered into publicly available generative AI tools unless the tool has been assessed and approved for that use by Harvard.