Atiba de Souza.Start a conversation
WritingAI

Three AI Stories This Week. One Bill. It Was Never the Tool.

By Atiba de Souza

An editorial cover image for a piece on Three AI Stories This Week. One Bill. It Was Never the Tool. The cover reads: AI Three AI Stories This Week. One Bill. It Was Never the Tool.

The win arrived with a bill attached

My team was stuck in the worst way. Not bad enough to throw out, not good enough to publish with confidence.

The social content for Rx Growth Hub, a pharmacy-industry company, was not reaching the quality bar on Claude. It was grading around 5, 5.5. Then Astra came out, the team switched, and the grades rose to close to 7.

That is the win. Then the second number walked in behind it, from the same change, in the same week. The same job, for one pharmacy, went from an hour and a half to almost four hours.

Two and a half times slower, for the better score.

So I asked the team the only question that mattered. How do you define better?

The answer is not a feeling of alignment. It is a sentence your team can say without you in the room: better for what, and here is what we are trading to get it. Quality gain and throughput cost, measured in the same conversation. If you only have the quality number, you ship a pipeline that cannot scale across pharmacies. If you only have the speed number, you keep content nobody wants to publish.

I wrote that one up in full (read it here), because the answer gets made in a room without you, and you will live with it for a year.

Then two more stories landed in the same seven days. Nothing teased a screen-free AI device for next month (that one is here). And a comment section sat there with a cap on it that nobody admitted setting (that one is here).

A model switch, a hardware launch, and a comment thread. On the surface, nothing in common.

They are the same story. This is not a recap of three things that happened. It is one position, and the three articles are the evidence.

The tool got better. The tool was never the story.

Here is the position. Everything below is this line with proof hung on it.

In all three stories, the capability improved and the win came from somewhere else: a person made their reasoning explicit, in a form somebody else could run without them.

The model switch: the model got better, and what decided whether that was good news was a conversation about what better means.

The device: the hardware is new, and the marketing that will work on it is not new. It is a year planned as one thing, so every post points up at something larger than itself.

The comment section: the tool is not the cap. You are. The cap is you answering the easy, the flattering, the famous, and calling it triage.

Not one of those is a model problem. All three are judgment problems that got exposed the moment the tool stopped being the bottleneck.

Thinking transfer: moving the reasoning behind a decision out of your head and into a form another person can run without you. Not the task instructions. The judgment underneath them.

No

Yes

New capability lands

The tool stops being the bottleneck

Someone must decide what better means

Is the reasoning made explicit?

You stay the human API

Same decision next quarter

A sentence the team can run without you

It compounds

InconsistentConsistent
The model gets picked on quality score alone or speed alone"Better for what" gets named together, then the pick gets made
The leader stays in the work while the task runsHands off until it is done, no exceptions
The question is "What do we post this week?"The question is "What does this week point up at?"
The comment section is rationed to the easy, flattering, or famousEveryone who showed up gets a compliment, a specific response, and an open question
A win skips the reflection because it workedA win gets the reflection most, because you need to know why it worked

The shared mechanic is a question

Look at what carries the weight in all three. A question. Four of them after the task. Does this point up at anything. What made you stop scrolling on this one.

Eight weeks ago I went to Claude Code with an operating problem, not a software one. I wanted the agency to work three weeks a month and take the fourth off, and I wanted to know what would have to be true for that to hold.

I expected a plan. What I got was questions. Week after week, about my philosophies, my frameworks, how I thought. It never produced the plan.

It produced a brain of me. When the project manager side started sending a morning brief of what actually needed doing across every project, my mind went quiet for the first time, because it no longer had to hold any of it.

I asked for an answer. What changed my business was being asked.

The questions did not delay the value. They were the value. If it had handed me a plan in one sitting, I would have gotten a plan. Because it kept asking, I got a brain. Hold that, because it is the same mechanism running under all three frameworks below. The question is not the delay before the relationship. The question is where the relationship comes from.

First, second, least: the three frameworks

They are not equal, and treating them as equal is how people work hard and move nothing.

Most important: the year planned as one thing

The Campaign-Driven Enterprise. A year planned as one thing, laddered to your own framework, so every piece an audience sees points up at something larger than itself.

The visible structure is three layers. Four major events a year, three quarterly and one larger annual, with a monthly event underneath. Content maps down from the same map. The annual layer carries your whole process and drives to one larger annual event and the signature offer. The quarterly layer carries one pillar of that process. The monthly layer carries one sub-pillar underneath that quarter's pillar.

But the layers only mean something because of the map underneath them. The ascension map is the named path a person walks, from the monthly event, to the quarterly event, to the annual event, to the offer. It is not a funnel drawing. It is a sequence of beliefs, where each rung assumes the person arrives already knowing what the rung below taught.

You build it backward. Start at the offer and ask what a person must already believe to say yes. Then ask what they must have seen to believe that. Keep walking backward until you reach someone who does not know you exist. Then turn it around, and that becomes the order you teach in.

Here is the test for whether you have one. Take a piece you published last month and ask which rung it moves a person from, and to. If you cannot name both, it was a rung in nobody's ladder.

And you measure movement, not output. Assets shipped is a number that always goes up and never tells you anything. Measure where people entered, which rung they left from, and where they stall. If people keep arriving at the monthly event and never show up at a quarterly one, that is not a content problem. It is a map problem.

Now watch why the framework is the thing that does the work.

We screen for a role by sending every applicant the same questions to answer on video before any live conversation, so the team can hear the person before meeting them. The videos came back nearly identical. Same structure. Same phrasing. Same examples. The candidates had all fed our questions to AI. The tell was one written answer that still had the AI's closing line pasted at the bottom, offering to craft more follow-up answers.

The lesson is not that people cheat now. It is that we had built a process to hear individuals, and a process with no constraint measures nothing but the tool. We changed one thing: we put a timer on the answer, so the question is seen for the first time when it must be answered.

Your content calendar has the same disease. If you can generate it in an afternoon from a generic prompt, so can everyone in your category, and it will come back with the same seven tips and the same closing line. Your framework is the timer. A piece that has to ladder to your pillar and carry your process cannot come back generic, because generic does not fit inside it.

The cost of this one is optionality, and optionality feels like safety. Once you name the pillar, every good idea that does not ladder to it becomes a no, and those no's arrive as real invitations with real dates from real people who will be mildly offended when you decline. The map has a second bill nobody prices. Drawing it is an audit of everything you have already shipped. Put last year's content on it and count what actually moves a person from one rung to the next. The bill is that you have to kill the things that work but lead nowhere, not the things that fail. The series that performs but points at nothing. The newsletter that teaches a belief three rungs above where your audience stands. That bill arrives before any upside does.

Second: the reflection meeting after the work

The CASE Method, the framework behind The Delegation Trap.

The misread first, because it is the one I see most. CASE is not a task-assignment acronym. It is not a set of forward instructions you hand over at the start of work. C-A-S-E sounds like something you hand someone. It is a meeting that happens after the task is done. That ordering is the whole thing.

You delegate the task, stay completely hands off while it runs, and hold the meeting only after it is complete. Timing plus the hands-off rule is what separates judgment from supervision. Get those wrong and it turns into supervision, and supervision produces compliance, not judgment. You do not hand someone the wheel and then reach over at every intersection. You hand over the wheel, let them drive, and talk about the drive afterward.

Four questions, and their job is to excavate how your team thinks and bridge the gap to how you think. The one leaders skip is E. E is not "Easier." E is "Easier Than Expected," a strengths-surfacing question. Ask what turned out easier than they expected. If the only answer is the score, you have not found a strength. You have found a number. Push until they name the part of the work that got lighter, because that is the part they now own and you can stop holding.

That is the difference between delegating tasks and delegating thinking. Traditional delegation transfers output. This transfers how to process information, recognize patterns, and decide. It is the difference between a team that pings you all day for exceptions and a team that thinks like owners.

The cost of this one, said plainly. Staying hands off while the work runs, knowing something will come back worse than you would have done it. That is the tuition. The quieter cost is that a reflection meeting asks you to explain how you think out loud, to someone who will ask why, and the first time you try you will hear yourself be vague. That is uncomfortable in a way that correcting someone's output is not. Correcting is cheap. Articulating is the work. And the third cost: they might choose differently, and their reasoning holds. If it holds, say so out loud and adopt it, or name plainly what standard you are adding. What you cannot do is overrule without naming the standard, because that teaches them the standard is "ask the boss," and you are back to being the human API.

Least important, and most seductive: the comment volley

The Volley System, CRQ. You are the email provider and your audience is the user, and you have set a cap on attention, not storage. You answer the comments that are easy, flattering, or from somebody with a big following, and you call it triage. It is rationing.

Here is why I believe opening the door does not drown you. I am telling it as an analogy, not as proof, and I will not dress it up as a study of comment sections.

In the early 2000s, when the cost of physical disk space was falling through the floor, Yahoo Mail did the thing that sounded insane. They decided to hand out email with unlimited storage. Google thought they had lost their minds. Everybody did. People will abuse it. People will dump everything they own into it.

Then the metrics came back.

Storage usage went down. Remove the ceiling and people used less of the room, not more.

Sit with that, because it should not be true. A cap makes you hoard. Give someone a limit and they will creep right up to it, because the limit itself is the threat. Take the limit away and the threat goes with it. What people wanted was never the storage. What they wanted was the feeling of having as much as they want.

A note on the source, because I would rather you trust me than agree with me. I tell this as Yahoo Mail, and the well-known version of unlimited mail storage belongs to Gmail in 2004. The company is not the lesson. The direction of the result is. Hold the direction.

That is a belief, not a proof about comment sections. It is the shape of a thing I believe is true, offered so you have a reason to run one small test. And here is the belief. When you open the door and answer everyone, what arrives is not a flood. What arrives is a conversation. Scarcity produces hoarding, people shouting into the void to get a scrap of your attention. Abundance produces dialogue.

The Volley System (CRQ): responding to every comment with a compliment, a specific response, and an open question, so the exchange cannot end on your side. C is the wrapper. R is the proof you read the thing. Q is the move that makes it a conversation.

C is a genuine compliment, something you noticed because you were paying attention. R is a response customized to the post, which means you read your own caption first. Q is an open-ended question, the move that matters. "Did you like it" is a dead ball. "What made you stop scrolling on this one" is a volley.

If you only do one move, do Q. A compliment and a response with no question is a fan letter. Nice, forgettable, over.

And here is the decision rule that turns "answer everyone" from a slogan into something you can run. Four branches.

  1. A direct question with an answer? Answer it. Do not volley.
  2. Something broken and your fault? Fix it first, volley after the fix, if at all. A question aimed at an angry person before you have solved their problem is dodging with extra steps.
  3. About to write a survey? "What do you think?" is not open ended. Why, how, what if are the openers. Everything else is fishing.
  4. Otherwise: C, R, Q, and end on why, how, or what if.

The cost here shows up on day three. You ask a good question and get nothing back, and you feel like the person at the party laughing at their own joke. That is not a verdict. It is the gap between a volley leaving your hand and the ball coming back. It also costs you the pose. Answering everyone means you cannot stand behind the velvet rope, and some of the status you enjoy as "too busy for comments" is manufactured by the cap. And it will feel beneath you for about a week. You built a business, and now you are asking a stranger what surprised them. Do it anyway. When it feels beneath you, notice that the feeling is the cap talking, not the work.

The story that should not have worked

At a recent US AI event, Darvin asked me whether a coffee shop owner who does not know what AI is could even be a target for the event.

My answer was a man who had already come. His entire experience with AI, in his own words, was signing up for the course. Another attendee that year turned up with a brand new computer and asked how to turn it on.

Judged on where he started, that man was the least likely person in the room to get anything out of two days. The obvious target market was people already using AI.

A hundred thousand dollars this year. Said two days later, by a man whose entire experience with AI, in his own words, was signing up for the course.

Now stay with the part everybody skips. That number did not come from the thing we built in the room. It came from the rewiring of how he thinks about solving problems.

Which is the same finding as the other two stories, arriving from a different direction. The plan is not the value. The room is not the value. The number shows up when a person's thinking changes shape. And the furthest-back person in the room is often the one it happens to fastest, because there is less to unlearn.

Two moves from that. Do not screen beginners out of a teaching offer. And when you collect the proof, capture where the person started, not just where they landed. "I made a hundred thousand dollars" is a claim. "My experience with AI was signing up for this course, and two days later I could see a hundred thousand dollars" is evidence. Without the before, the after is just bragging.

Rank the parts, then pick a starting line

Most important: the year planned as one thing. Second: the reflection meeting after the work. Least important, and most seductive: the comment volley.

So you start with the annual pillar. Not the comments. The voice in your head telling you to start with the comments because it is cheapest is the exact seduction I warned you about. Fix the annual layer first, your whole process, one larger annual event, one signature offer. Everything else reorganizes once it holds. Then hand one task over and hold the reflection meeting. Then, and only then, the comment volley, because the volley is where you practice the one muscle all three need: the willingness to end on a question instead of an answer.

When not to do any of this

A framework handed over without conditions is a trap, so here are the conditions.

If you are pre-offer, skip the ladder. Go get one customer and earn the right to plan a year. If your business is not repeatable yet, an annual event amplifies confusion instead of creating demand.

If you already know the answer you want to hear, do not run the reflection meeting. That is a performance review wearing a reflection meeting's jacket, and people can tell.

If you cannot pay the silence, do not start the comment test. Day three is quiet, and a half-run test tells you nothing.

And if something needs your hands this month, do not install a year-long system as a way to avoid it.

The rewiring underneath all three

Name it, because it is the thing that stalls adoption, and nobody names it for you.

You are moving from reacting

<!-- cluster-links:begin -->

Related reading

<!-- cluster-links:end -->