Summary
As AI agents grow more capable, the biggest bottleneck in many engineering workflows is the human sitting in the loop. This video from IndyDevDan argues that top agentic engineers are moving from writing prompts, to supervising agents, to building systems that run on their own.
The central framework has three stages: in the loop, near the loop, and out of the loop, summarized as prompt, click, cron. In the loop, an engineer is in the terminal writing every prompt. Near the loop, agents are triggered by events such as webhooks, cron jobs, or log classification, and the human steps in only to give feedback. Out of the loop, systems run fully without supervision. The video argues that 80% or more of current engineering work could run without a person present given the right skills, tools, and architecture.
The presenter demonstrates several near-loop systems he runs today, including a durable news agent that surfaces and ranks information, and a UI replicator that repeatedly builds, screenshots, and compares output against a mock design until it hits a similarity threshold. He also covers the core four of agent design: context, model, prompt, and tools. Engineers interested in meta-skills, orchestration, and building systems that build systems will find a practical roadmap here.
📺 Source: IndyDevDan · Published October 26, 2026
🏷️ Format: Opinion Editorial
