A plan looks fine on the page. What it actually asks of people — who waits on whom, who cannot say no, what gets finished and never accepted — you find out by living through it, once, at everybody's expense. The moves are the ones you already make: asking, agreeing, declining, countering, reporting it done, and saying yes, that was what I needed. Those are built in. What you design is how they connect — which act sets off which conversation, and what reports back to whom.
4-loop is a free browser tool for drawing the few conversations a piece of work turns on as promises between named people, and then running them. It is not a tracker: nothing here follows real work or keeps a record of it, and drawing all of it would be a worse tool rather than a fuller one. The acts arrive one at a time, and a work inbox shows whose day it is: what is on each person's plate, who they are waiting on, and the day each thing has to be done by. A process is designed from above and lived from below, and this is the view from below.
It will also write a network out as a training session, so a chatbot can facilitate it one moment at a time while one trainee takes a part — written in your browser, shown to you, and sent nowhere unless you paste it. There is a runner built for that text on claude.ai, and it receives nothing until you do.
Every commitment is one loop with four moments — the request, the agreement, the report that it is done, and the customer declaring themselves satisfied — and a loop is finished only when it gets back to where it started. The notation comes from the language/action perspective and the conversation for action described by Terry Winograd and Fernando Flores. What a run shows that a board of tasks cannot: who is waiting on whom, which promises were never actually asked for, which moments no person stands behind, and where a date comes from.
It is an instrument for designing systems rather than for running them: you draw how the coordination is meant to go, run it, and take what the design asks of people into the design and into the conversation. Everything around the work already records what happened and agents will record far more, but almost none of that machinery has a theory of what people are doing when they speak — it can tell you a message was sent, not that somebody was asked. This is where the account of what was supposed to happen lives. Nothing you draw leaves your browser, and an annotated example network opens in a click.