Welcome to Southbridge! This is a task that will see you use coding agents, understand them and work with them.
The Task
Build an interface where you can submit prompts and view outputs from at least three different coding agents. Claude Code, OpenCode, Cline, Codex and Gemini CLI are all solid candidates.
We want to be able to submit multi-modal input to the agents, see actions being performed, inspect tool calls, view file diffs as they are edited.
All of the agents listed above (except for Cline) can be used as headless servers (some through the agent client protocol, others through their own SDKs) to connect to a frontend.
This task is designed to be completed in 3 or 4 hours. If you can't complete in that time, don't worry! It's designed to be hard. Partial results are accepted, as long as you acknowledge where you fell short.
Stack
Use NextJS and Bun for this application. Additional points for:
- Packaging the app into an executable using Bun
- Using new Bun features like the embedded terminal and sandbox
Outputs
Please provide:
- A repository with the code, and instructions on how to execute it, with layered commits with your thoughts and issues noted.
- A process document with all of the prompts used, AI assistance, problems with AIs you noticed, etc. written and collected manually. This is the most important document - submissions will not be accepted without this document, unless they were done without AI assistance. Please do not AI generate this document.
Inspiration
You can use the AI SDK playground as inspiration. There's also a sample UI (please don't use this as a direct mockup, this is simply for illustration) in the page below:
AI Use Policy
Southbridge is an AI deep-tech company that is also AI-enabled. Our internal team is extremely well versed in AI tools across the stack.
You are allowed to use AI assistance for solving this problem. As much as you want.
What is not allowed is dishonesty. All AI-use must be disclosed, explicitly mentioned, and flagged as such. You are responsible for any code handed in.
However, at the end of the day, what is being evaluated is your performance - either as a developer, a reviewer of code from an AI, or a software architect.
Below is the Southbridge AI policy that we apply to our entire team.