What a useful multi-agent AI project needs beyond prompts

Prompts are not enough

Many AI projects stop at a prompt box. A useful multi-agent product needs more structure: provider boundaries, secure credentials, workflow state, and a clear way to compare output.

Tarka Sabha is where Subhajit Pradhan explores that engineering layer. The product idea is an AI debate platform, but the important technical question is how to coordinate multiple agents without turning the codebase into provider-specific glue.

Provider boundaries matter

If provider logic leaks everywhere, every model change becomes a product refactor. A cleaner implementation keeps orchestration behind backend boundaries, so the interface can stay focused on the debate flow.

That also protects API credentials. A browser should not need direct access to provider keys. For Subhajit Pradhan, this is where full-stack judgment matters: product UX, backend orchestration, and security boundaries have to agree with each other.

Make the workflow visible

An AI debate platform should show more than final text. It should make the workflow understandable:

• what context was used
• which agents participated
• how responses were generated
• where comparison or history should live next

That turns an AI demo into a product with architecture, state, and room to scale.