What the channel means by it The team at Every uses the agentic loop to describe a working style imported from coding into all of knowledge work. Dan captures it with the mindset of the programmer: “I don't do the work alone; I'm building a system that performs work.” [^cb24ef4d-a000-5560-8446-9ed871f58aa8@65] The loop is the repeating cycle that makes such a system productive: set the rails, let agents do the work, review the results, and improve them over time. Why it matters in their model comparisons In their review of GPT-5.6 (Sol), the loop is the yardstick for what makes a model usable day to day. What counts is not raw intelligence but how well the model supports fast cycles of drafting, reviewing, and revising — reviews so cheap it is easy to explore options [^cb24ef4d-a000-5560-8446-9ed871f58aa8@817] — and how well it keeps multiple parallel threads moving in one place. [^cb24ef4d-a000-5560-8446-9ed871f58aa8@730] The team's benchmark consolidating 46 CSV files showed this directly: 5.6 in ChatGPT Codex asked smart clarifying questions and produced a useful result with minimal intervention, while other models got stuck or produced something unusable. Spreading beyond code Dan sees the agentic loop expanding from programming into general knowledge work, with 5.6 in the ChatGPT apps as the first place non-technical people can genuinely run it — the working surface hides the code so anyone can set the rails and let the system perform the work. - The agentic loop is a repeating cycle: set the rails, let agents do the work, review, improve. - It comes from coding (“I'm building a system that performs work”) and is spreading into general knowledge work. - In the channel's model comparisons, a model's value is judged by how well it supports fast, cheap iterations inside that loop. - The team frames it as now accessible to non-technical users through the merged ChatGPT app.