Fix the Bottleneck

Version 0.1.0. Free. Works with major AI agents. Source: github.com/diaryofamademan/skills

What it does

Every chain that produces finished output is limited, at any moment, by one constraint, and output cannot exceed what that constraint passes, no matter how hard every other stage works. An hour lost at the constraint is an hour lost for the whole system; an hour saved anywhere else is a mirage.

Most improvement effort ignores this. It spreads evenly across the system, or follows whoever complains loudest, while fixes land at stages that were never setting the rate.

This skill turns Eliyahu Goldratt's Theory of Constraints into a procedure an agent or an operator can run and check. State the goal as throughput: the finished unit, its current rate, its target.

Map the chain and measure every stage. Name the one constraint the evidence points at, where the queue grows in front of it and the stages behind it starve.

Get more from the constraint as it stands, align everything else to its pace, and only then spend on capacity. When throughput moves, the constraint has moved with it, so the loop runs again, and the rules written for the old constraint get audited before they become the new one.

A fix counts only when system throughput moves.

The skill ships with its proof: it was field-tested before publishing on three live systems, an agent fleet, a revenue funnel, and a health plan, and each test changed the method. Read the agent fleet case study on GitHub

This skill's own launch became a worked example. The fourteen versions its launch copy went through, and the principle each one died on, are the case study inside Stop the Scroll. Read Stop the Scroll

Use it when:

  • Your system is running and its output is below target: a delivery pipeline, a CI/CD chain, a support operation, an agent fleet, where work is plentiful and finished output is not.
  • Fixes keep landing and their own metrics improve, but total output does not move, and a queue nobody can explain keeps growing in front of one stage while every stage reports itself busy.
  • You are about to spend on capacity: hire, buy, add machines, add agents, and the real question is whether the constraint is actually the thing being bought.

What is inside:

  • The seven step procedure: state the goal as throughput, map the chain and measure it, identify the constraint, exploit it, subordinate everything else to it, elevate it, and repeat, held in order by thirteen guardrails, including measure before naming, one constraint at a time, and exploit and subordinate before any spend.
  • The output contract every run returns: the goal with its rates and window, the chain map and the stage equation that has to multiply out to the measured output, the constraint verdict with its kind (resource, policy, hero, or external) and the evidence behind it, the exploit, subordinate, and elevate lists in order, the re-measure plan with its done rule, and the inertia watch.
  • The reference files: Goldratt's five focusing steps in depth, the flow math (Little's Law, the utilization curve, and the stage equation), throughput accounting for trade-offs, the four constraint kinds and how each is treated, and a worked case study of a full run on an AI agent fleet.

Get the skill

Put your email in and Fix the Bottleneck arrives right away: the install link and a one page quick start. You also join Running on Agents, my free newsletter, and get every update the day it drops.

Your email is used to send the skill you asked for and the free newsletter. The list runs on Beehiiv.

You can unsubscribe in one click in any email. Your address is not sold.

If you apply for the audit, that information is used only to review your application, unless you also tick the newsletter box on that form.