The AI Team

Four agents, visible handoffs, one campaign

The problem

“Multi-agent” is the most over-claimed phrase in AI. Most demos are one model talking to itself with different labels, and every agent agrees with the last one.

What I built

One brief in. A researcher works out who actually buys the product, hands findings to a strategist who builds the campaign idea, who hands it to a copywriter who writes the assets, which a reviewer then scores honestly and improves. Each handoff is visible in the interface as it happens.

How it works

• Each role is a separate call with its own system prompt and only the previous agent's output as context — so the handoff is real, not decorative.
• The pipeline is driven client-side, which means the user watches the work move between agents instead of staring at a spinner.
• The reviewer is prompted to be adversarial. On a live test it scored its own team 6/10, named two specific weaknesses and rewrote their best headline.

Stack

What it proves

Honest limitations