Utiliser une équipe d’agent developpeurs → à tester
hey @everyone
We’re open sourcing our newest project: The Coding Ability
This is novel approach using a human in the loop, multiple agents & recursive programming all work cohesively to craft a substantial piece of software.
Each agent is an expert in it’s own domain (Product Owner, Systems Architect, Developer), which allows this system to:
-
Break down the user’s request into requirements
-
Design the system and redesign based on feedback from the user (human in the loop in action!)
-
Code the system, while revisiting and updating sections that were incorrectly written during the coding process
-
Package, deliver and deploy the system to the user
Rather than using a single agent to write code, we have provided you with an entire tech team.
In particular, we’d like to point you to our novel approach 
Recursive Programming
The Code Ability can theoretically program a system of any size and complexity. This is because of the recursive way its algorithm works. It breaks the program into small, manageable chunks that a large language model can write without mistakes.
These chunks are like standalone functions that don’t need outside context. If a function is too complex, it gets broken down further.
Example
- We ask the system to create the entire program.
- It splits this into five smaller sub-programs and connects them within one, simple main function.
- The system writes the function names first, and these get implemented in the next loop. Any unimplemented functions are handled by a large language model with no context of the overall program to avoid confusion.
- Each of these five sub-programs is implemented individually. If one of these is still too complicated, it will be broken down further.
This is how the Code Ability handles complex coding tasks.