Your Feedback Loop Is Pointed at the Wrong Folder
By Atiba de Souza
The folder nobody attached
A pharmacy owner of seventeen years walked into a live AI workshop I was running. She was opening a wellness clinic, and she had already done the hard part. She wrote her business "brain" document, the file that holds how the business thinks, what it stands for, how it decides.
She came to the workshop because the AI kept producing work that did not sound like her business. She was sure the problem was her prompting.
It was not her words. It was the wiring. Her projects had never been pointed at the brain file, so every project started from nothing, and then she asked the machine to sound like a business it had never been shown.
We attached the folder. We named the file in the instructions. Inside the last hour she built a vetting guide, a legal-coach skill, and a launch checklist.
Seventeen years of running a business. One hour to turn the output around. The fix was never in her prompting. It was in the wiring.
Stay with that, because it is not really an AI story. It is a feedback loop story, and it is probably yours.
She was running a loop: prompt, output that misses, rewrite the prompt, output that misses again. The loop repeated. It felt like learning. It was motion.
Her loop was not broken. It was pointed at the wrong folder. A loop can run perfectly and measure nothing.
What the CASE Method actually is
CASE Method: a scheduled, structured post-task reflection meeting. You delegate the task, stay completely hands off while they do it, then hold a four-question meeting after they finish, aimed at how they think, not at what they produced. It transfers judgment, not instructions.
CASE Method is not a way to assign work. It is not an acronym you hand someone on the way out the door. Everyone gets that part wrong, including people who have read about it, because the name sounds like a checklist and the method is a conversation.
Here is the shape of it. You delegate the task. You stay completely hands off while they do it. Then, after they complete it, you run a four-question meeting whose job is to excavate how they think and bridge the gap to how you think.
That is the whole point: delegate thinking, not tasks.
Traditional delegation transfers output. "Do it like this." CASE Method transfers input. How to process information, how to recognize patterns, how to make the call the way you would make it. That is the distance between a team that pings you all day for exceptions and a team that thinks like owners.
The work happens with your hands off. The learning happens after, with your mouth mostly shut.
The question that finds your blind spot
The four questions spell the method: C, A, S, E.
The first three set the table. The fourth is the one that does the thing you came here for. E is Easier Than Expected, the strengths-surfacing question. Not "E equals easier," which is the misread that turns a reflection tool into a compliment.
Look at what E actually asks. It asks what went better than the person expected. Which means somebody has to name what they expected. And when they answer, you find out where your map ran out.
You predicted the hard part would be X. It was not. You predicted they would need you at Y. They did not. Every place your prediction missed is the edge of your own map.
That is the blind spot. Not the output. Your model of the person and the work.
| The loop that repeats | The loop that learns |
|---|---|
| Questions arrive before the work, as instructions | Questions arrive after the work, as reflection |
| You hover and correct while it runs | You stay completely hands off until it is done |
| "Did you do it the way I said?" | "How did you decide that?" and "What was Easier Than Expected?" |
| Transfers output: the task, done your way | Transfers input: how to process, recognize patterns, decide |
| You learn whether the work got finished | You learn where your own prediction was wrong |
| They come back for every exception | They start thinking like owners |
The screening that measured nothing
Now watch the same failure show up where nobody expected it.
Every applicant for a role got the same interview questions to answer on video before any live conversation, so the team could hear the person before meeting them.
The videos came back nearly identical to each other. The candidates had fed the questions to AI. One written answer still had the AI's own closing line pasted on the bottom, offering to craft more follow-up answers.
Here is the part that should stop you. The stage was running exactly as designed. Applications in, videos out, comparable format, on schedule. And it was reading nobody. Let it run and the team quietly spends the whole screening comparing one AI's answers to another AI's answers while believing it is reading people.
Notice what the fix was not. It was not more diligence from the reviewers. It was a timer on the answer, so the question is seen for the first time when it must be answered. Redesign the loop so the input cannot be pre-processed.
That is what a CASE meeting is, by design. The questions come after the work, in conversation, about decisions only that person made. There is nothing to feed to a machine first. The timer is built in.
What this costs you, and when to skip it
Now the honest part, because you are not going to adopt this on a clean process diagram.
The fear: hands-off feels like abdication. You got where you are by knowing. Sitting still while someone does it their way, when you can see the better way from your chair, is not a scheduling change. It is rewiring your identity.
The cost: E asks what was easier than expected. Sometimes the answer is "most of it." That lands on you, in front of someone you manage. You asked the question. You get the mirror.
The calendar cost: this meeting happens after work that is already done and already fine. It will look like the least urgent hour in your week, and it will be the first thing you cut. Cut it and nothing visibly breaks, which is exactly why it disappears.
The payoff is not on the first one. The first meeting is clumsy. Someone gives you one-word answers. It pays out as the record builds.
Ranking, since not every part of this carries equal weight. The part that matters most is where you point the loop: at how they think, not at what they produced. The part that matters least is the name. Whether you call it CASE or run the questions off the back of a napkin is irrelevant.
Do not run this on work you cannot let fail, where hands-off puts a client or a person at risk. And do not run it on work with no judgment in it, where there is no thinking to excavate. This is for work where judgment is the deliverable.
Ask AI for the pattern, not the chores
Here is where AI earns its place, and it is almost never where people aim it.
An AI assistant clears the routine work off your desk and gives you back time. A second brain reaches the work that moves the needle. And here is the catch with the assistant: the tasks it takes over were mostly the ones you were never doing anyway. That is why the time it hands back always feels smaller than the promise.
Summarizing your meeting notes is the assistant. It saves a chore. It changes nothing.
Handing AI the record of your own reflection meetings and asking it to find the pattern is the second brain.
You might be thinking AI cannot tell you something about yourself that you do not already know. You are right. It can tell you something about your record, which is not the same thing and is far more useful.
So the record has to be real: their decisions, your predictions, the places the two diverged. And the ask is not "give me some feedback ideas." The ask is "across all of this, what do I keep predicting wrong about my team?" Ask for the variable you misjudge. Not a list of advice.
One warning, and it is the pharmacy owner's warning all over again. If the project is not actually pointed at your record, you are not getting a read on your business. You are getting a read on the internet. Same folder problem. Check the wiring before you rewrite the prompt.
Start Monday
Pick the next thing you are about to delegate. Delegate it. Keep your hands off it. Book the reflection meeting for after it is done, and put it on the calendar before you start, not after, because it will be the first thing you cut and nothing will visibly break when you do. That silence is the trap.
Run all four questions, in order. Then sit with the answer to E, even when the answer is about you and especially when it is. You asked for the mirror. Stay in front of it long enough to see.
Then once a month, take the record of those meetings and hand it to AI with one question: what do I keep predicting wrong? That is a loop with a mirror in it. Run it for a quarter and the blind spot stops being a blind spot. It becomes the place you are already looking.
Two failures, one shape. The pharmacy owner had the brain and never pointed a project at it. The screening stage had the questions and never built in a timer. Both loops ran perfectly and measured nothing.
Yours is one wiring decision away from being the same story, or from being the one that finally tells you the truth. Check where the loop points before you rewrite a single prompt. You were never missing the answer. You were missing the wiring.
<!-- cluster-links:begin -->