Summary
IndyDevDan argues that raw model capability is no longer the main edge for engineering teams, and that focused agents with specialized tools, context, and system prompts beat a single generic agent on the same model. The video walks through what he calls small agent teams: a handful of specialists, each owning one kind of work, coordinated by an orchestrator agent.
The video starts with commentary on the new Fable 5.2 release. He points to the model’s system card, which shows a five-agent team outperforming both a single agent and a large batch of asynchronous agents, though at higher token cost. He also looks at Artificial Analysis data, noting Fable 5.2’s pricing of $8 per million input tokens and $40 per million output tokens, and highlights that the medium and low reasoning settings score 56 and 54 on the intelligence index, which he sees as more notable than max mode. He names Opus 5.5 and Kimi as other capable options.
The second half demonstrates a tiny software factory in which a lead agent delegates a task, such as sketching a client-server architecture for an image generation platform, to a team of agents. It shows team collaboration settings, a conversation prompt that defines how the team communicates, and why prompt and harness engineering matter more than the model at this level.
šŗ Source: IndyDevDan Ā· Published November 16, 2026
š·ļø Format: Hands On Build







