Inspiration
The attraction of visual IDE environments over the years is that they allow a high level, structured way to organize an application as it is being built. With AI builders we are switching between two extremes - an abstract array of natural language statements to describe/refine the app, and an option to switch back to the generated code in full. We are missing the middle ground.
What it does
AIIDE is an IDE for the AI-era. It doesn't produce code itself and it doesn't ask you for prompts - platforms like Bolt.new already do that well enough. AIIDE provides the UI and tools to organize application building at a level between natural language and code. It does this by using a tree/node structure of arbitrary structure and depth. Applications are designed at this level - at any complexity needed, starting from blank or from a template. The tree is then fed into an AI app builder platform like Bolt.new, which is now given structured guidance on what to build. Equally importantly, the AI app builder platform is instructed to keep the tree updated as changes are made in normal AI fashion. This turns the tree into a blueprint - a detailed spec of how the final built app works, but at a much more flexible level than just taking the produced code, which would be a highly opinionated version of that spec.
How we built it
AIIDE was built 100% in Bolt.new, deployed to Netlify, with a domain from Entri. In sequence:
- The tree structure mechanism and UI was built first
- Import/export was added
- Linked to Firebase for authentication and online storage
- Separate Applications and Templates sections were created
- A Content Hub was added for people to share and download official and community templates
- The website and info pages around the AIIDE were built
- A suitably mnemonic domain was created for people that need aid building AI apps - aiide.me !
Challenges we ran into
No major challenges encountered. Bolt.new proved a fascinating experience, turning out application sections with ease. Indeed sometimes, making minor adjustments such as a context menu appearing above the clicked item when it is near the bottom of the screen took more effort to build than whole parts of the application.
Accomplishments that we're proud of
The tree mechanism is very rich. Quick addition of tree node types and useability enhancements like immediate focus to a name/value field for quick configuration, tree-wide search with smart scroll-to-result and auto expansion of tree nodes if they were collapsed, intuitive icons and layout, smart detection of suitable node options depending on the node selected. The Content Hub is also a favourite - allowing people to easily share templates, complete with versioning (and version override detection), categories, short and long descriptions, and category/subcategory options with search.
What we learned
AI app builders like Bolt.new are revolutionary, but preparation and method to achieve solid results that are reliable enough to maintain and expand are a must. And AIIDE helps with that !
What's next for AIIDE
The major and obvious next step would be for an AIIDE-type of toolset to be merged with AI app builders directly in a single environment. This would provide immense value to the AI app builder paradigm - allowing much more reliable and faster implementation, branching, restructuring, and expansion of an app.
Built With
- bolt.new
- entri
- firestore
- neltify
Log in or sign up for Devpost to join the conversation.