Map Workflows and Diagnose Fragility

In the last chapter you learned to name the pieces: the tools, the workflow that connects them, the system around them, and the point where AI acts. Naming is the setup. Now comes the method itself. In this chapter you will learn to draw a workflow as a simple map, then read that map for its weak spots, the places most likely to fail. This is the core move of analysis, and everything later in the course builds on it.

Map a Workflow Step by Step

To map a workflow, you walk it from its starting point to its finished result and write down what happens at each step. Six things are worth capturing, and once you have them, the whole path is visible at a glance. You do not need to memorize a rigid form; you need to name each piece honestly, in order.

Start by listing the components: every named part the workflow touches, meaning each tool and each person involved. For the food-drive thank-you workflow from earlier, the components are the sign-up form, the shared spreadsheet, the group chat, the AI assistant, and the volunteers who run them.

Then, for each step in order, name the actor: the person or AI tool actually doing the work at that step. A step always has exactly one actor, and being strict about this is what later lets you say precisely where a problem started.

Next, capture the input and the output of each step: what goes in, and what comes out. Between them sits the transformation, the change the step makes to turn its input into its output. Copying a form entry into a spreadsheet takes “one submitted form” (input) and produces “one new spreadsheet row” (output), and the transformation is the act of copying.

Now mark each handoff: the moment work passes from one actor or tool to the next. Handoffs are where information travels between pieces, so they are worth naming precisely. The point where a volunteer takes the form’s entry and moves it into the spreadsheet is a handoff, and so is the point where the spreadsheet list is passed to the AI assistant.

Note every decision point: a place where someone chooses between paths, such as approve or reject, send or hold, redo or move on. In the food drive, “do these totals look right before we thank anyone?” is a decision point.

Finally, record the dependencies. A dependency is when one step cannot start until an earlier step finishes or hands over its output. The AI assistant cannot draft accurate notes until the spreadsheet list is complete, so the drafting step depends on the copying step. Dependencies tell you the order that actually matters, not just the order things happen to occur.

Spot the Fragility Patterns

With the map drawn, you read it for fragility points: the steps most likely to fail or to produce a wrong result. You do not have to guess at random, because fragility tends to appear in a handful of familiar patterns. Learn these and you will know where to look first.

• A single point of failure is a step, tool, or person that the whole workflow leans on, so that if it fails, everything downstream stops with it ;

• Manual re-keying is any place where a person copies information by hand from one tool into another, which is a frequent source of typos and dropped entries ;

• Stale data is information that was correct when it was captured but has since gone out of date, so a later step quietly works from the wrong numbers ;

• Unclear ownership is a step that no one is clearly responsible for, so it gets skipped, delayed, or done twice ;

• An unchecked AI step is an AI output that moves on with no human-in-the-loop, letting a fluent-but-wrong result pass as if it were verified.

For each fragility point you suspect, add a quick evidence check. Ask whether you have direct evidence, such as a visible error, a mismatch between two tools, or a missing confirmation, or only a hunch that something feels risky. A suspected weak spot backed by a real mismatch deserves attention before one you are only guessing at.

Watch a Full Map Come Together

Let us put both moves together on the food-drive thank-you workflow, the same effort whose final donor list came out wrong in the course introduction. First the map, step by step.

• Step 1. Actor: the donor. Input: their details. Output: one submitted form entry. Transformation: filling in the form ;

• Step 2. Actor: a volunteer. Input: the form entry. Output: one new spreadsheet row. Transformation: copying it across by hand. Handoff: form to spreadsheet ;

• Step 3. Actor: the volunteers together. Input: the spreadsheet totals. Output: a “totals look right” confirmation. This is a decision point ;

• Step 4. Actor: the AI assistant. Input: the donor list plus the instruction to write a warm note. Output: a set of drafted notes. Handoff: spreadsheet to AI assistant. This step depends on Step 2 being complete and correct ;

• Step 5. Actor: a volunteer acting as the human-in-the-loop. Input: the drafted notes. Output: approved notes ready to send. Transformation: reading and correcting.

Now read the map for fragility. Step 2 is manual re-keying, so a mistyped or skipped entry would give the whole workflow a wrong donor list, which explains the missing names. Step 3 has unclear ownership: if no single person is responsible for confirming the totals, the check gets skipped, which explains the total that did not add up. Step 4 becomes an unchecked AI step the moment Step 5 is rushed or dropped, which is exactly how a note thanked someone for a gift they never made. Notice that none of these required anyone to be careless. The map simply shows where the chain was always going to strain, and now you know where to look before it breaks.

Let’s Recap!

• Mapping a workflow means walking it start to finish and naming its components, actors, inputs and outputs, transformations, handoffs, decision points, and dependencies ;

• A short ordered list, written start to finish with the actor and handoff named at each step, is a complete map that needs no special software ;

• Fragility points cluster into familiar patterns: single points of failure, manual re-keying, stale data, unclear ownership, and unchecked AI steps ;

• For each suspected weak spot, check whether you have direct evidence or only a hunch, and rank by likelihood and by the damage a failure would cause ;

• A good map does not just describe a workflow, it shows in advance where that workflow is most likely to break.

You now have the two-move method, mapping and then diagnosing fragility. The next chapter walks the whole method through one full example from start to finish, so you can see every move applied to a fresh, everyday effort before you try it yourself.