What Went Quiet

In January of 2026 a neuroscientist named Jared Cooney Horvath sat in front of the United States Senate Commerce Committee and said one sentence that nobody on the panel had a good answer for. Our kids are less cognitively capable than we were at their age.

For most of the last century, on every standardized measure anyone bothered to track, each successive generation outperformed the one before it — the sharpest gains in fluid, abstract reasoning. The trend had a name — the Flynn effect — and it ran like clockwork from the 1930s through the early twenty-first century. Generation Z broke it. The break shows up around 2010 in U.S. adult test data and in the OECD's PISA scores — which measure schooling, not IQ.

2010 is close to when the Common Core left cursive out and Chromebook pilots started rolling into American classrooms.

This is the receipt we've been writing for forty years and refusing to read. The chain runs cognitive offloading produces cognitive debt produces cognitive stunting. Cognitive debt is the MIT Media Lab's term: in June 2025 the Lab put electrodes on the heads and measured up to a 55 percent connectivity drop when the AI was in the chair. The Brookings premortem traces the same offloading loop across ≈400 studies and 50 countries.

The senior IT operator did not cause this. The senior IT operator did inherit it. The new hire walking into the building this Monday will. The next new hire after that one is in even worse shape.

If Part 1 named the trap, Part 2 names the line that was crossed. The line was the first time in nearly a century the next generation tested worse. The Senate has the testimony. The schools have the data. The pencil has the answer it always had.

You may ask how I know. I know because of a kitchen table.

I was three years old when my grandfather first sat me in front of a computer. He held two master's degrees — one in math, one in computer science. He was not anti-machine. He had every machine. He turned them on in front of me before I could read. What he would not do was let me actually use the machine until I had worked the same problem out the long way — on paper, in pencil, with the steps shown. The computer was always within arm's reach. The pencil was always in my hand. Reach for the machine, and he would put a sheet of paper down between me and the screen and tell me to do it the other way first.

I fought him about it. I was a kid. I did not understand. The machine was right there. The machine was faster. The machine was already showing the answer. The pencil was slow and the steps were hard and the kitchen table was a battleground for years. He never raised his voice. He just kept putting the paper in front of me. He understood, in a way I did not understand for the next thirty years, that the friction was building something the machine could not.

I did not figure out what the friction was building until I picked the pencil back up. Two things had gone quiet on me without my noticing.

The first was the inner voice — the running commentary that narrates the work as you are doing it, asks the next question, names the thing that does not fit, says ahhh it makes sense.

The second was the inner picture — the ability to hold a system in my head, rotate it, walk around it, see where the part I was looking at fit into the whole. Both of them had gotten quieter year by year as I used more digital tools. I did not notice the silence. The silence happened slowly.

When I sat back down with a notebook and a pencil, both of them came back. Not all at once. Not all the way. But they came back. And like any muscle you must continue to use it or lose it. My grandfather had been defending them at the kitchen table the whole time.

The atrophy closes in three stages, and most operators are at least one stage in before they can name it.

Nobody notices the silence close in. The silence is the absence of something that used to be running in the background. By the time you notice it, you have already lost what was making the noise.

Part 1 told you to open your calendar and ask whether you could rebuild last week's work if the AI vanished. This is the smaller, sharper version. Open last week's shipped work and ask one question of each item — did the thinking behind this come out of my head, or out of the tool's?

Not the typing. The thinking. The judgment underneath the work — the diagnosis, the design choice, the ranking, the call. Did it form in you, or did you read it back off the screen and ship it?

• A Supporter goes through last week's closed tickets. Whose diagnosis was it, yours or the chat's?
• A Builder goes through last week's runbook edits. Whose reasoning is in the steps?
• An Architect goes through the ADRs signed. Whose design is on the page above your name?
• An Orchestrator goes through the summaries sent. Whose ranking decided what the room saw?
• A Strategist goes through the decisions made. Whose assumptions are holding them up?
• A Process Owner goes through the gates enforced. Whose judgment is in the approve-or-reject?

Mark each item mine or the tool's. Be honest, because nobody is watching this count but you. Most supporters, operators, builders, architects, orchestrators, strategists, and process owners are surprised how far the marks lean toward the tool — and more surprised that on some items they cannot tell which one it was. The ones you cannot tell are the worst ones. Those are the items where the line between you and the tool already blurred.

→ Count the marks. The lean toward the tool is the diagnostic. The items you cannot label at all are the diagnosis getting worse.

You are up against three things — the same three things you were up against in Part 1, working in the opposite direction.

The clock. Part 1 said the clock pushes you toward the AI because the AI is fast. The same clock pushes you away from the pencil for the same reason. Paper is slower. Drawing a diagram by hand takes minutes the queue is not going to give you. Outlining a runbook section by hand before typing it doubles the apparent work. The clock has not changed. The clock will not get more forgiving. The pencil costs time the clock has already spent.

The room. Part 1 said the room votes for the AI because the room is interested in this quarter's numbers. The same room is suspicious of the notebook. The peer wonders why you are not in the doc. The manager wonders if you are not being productive. The junior wonders if this is a generational thing. The room is not against the pencil. The room just does not have a column on the dashboard for what the pencil is doing.

Yourself. This is the hardest one. The AI flatters you. The pencil does not. The first page in the notebook when you pick it back up is rusty. The diagram comes out crude. The inner voice does not come on at full volume the way you remember. The page is a humiliating data point about how much has atrophied. The AI will never give you a data point like that. The pencil will give you one every time you sit down with it, until the muscle is back.

The clock will not give you the time. The room will not vote for the work. Your ego will recommend you stop the experiment after the first ugly page. Walk past all three, on a schedule, anyway. Do not overdo it. A muscle can be over-worked and will take more time to recover then daily or weekly budgets.

First, something that needs to be clear. The pencil is for the learning loop and for low-priority work with room to think. The crisis loop runs on every tool on the desk, AI included. When the P1 hits, you fight the fire with everything you have — you do not stop to hand-draw a diagram during an outage. The pencil is for the calm hours, and every archetype has them.

Take one item the audit marked as the tool's. Just one. Next week, that one starts on paper — in your hand, in your words, before the tool sees it. That is the entire practice. You are not banning the AI; you are making sure the thinking exists in you first. Here is how that one item breaks down at every altitude.

Whatever your job, it is made of exactly three things you can put on paper: a system you are responsible for, a decision you make over and over, and a concept you rely on but have never had to explain. The contents change. The structure does not.

And if the page stops you cold — it will, the first few times — do not answer the big instruction. Answer the small question. A command like draw your system freezes a cold muscle. A question like what did you touch last? never does.

The System. Draw the thing you are responsible for, from memory, then check it.

• Supportera user's environment. What was the last ticket you closed? Draw that machine, trace out to what it could not reach.
• Builderthe system or workflow under a runbook. What pages you most? Center it, draw what it leans on.
• Architecta design you approved. Which boundary did you argue hardest for? Draw inside it, then outside.
• Orchestratorthe org map or process you run across teams. Which handoff breaks most? Draw the two teams around it.
• Strategistthe landscape of bets in flight. Which bet has the most money on it? Write the assumptions under it.
• Process Ownera process you own, one gate at a time. Which gate do people complain about most?

Start there — state before, state after, handoff, decision record. If you cannot show a decision record on why a gate is there, what the alternatives were, when it should be re-tested, and what risks were accepted, you have found the cargo cult living inside your own process. One gate a week, not forty in an afternoon.

The Decision. Write a recurring decision by hand, no reference, then check it.

• Supportera diagnostic tree. What symptom do you see most? Write the first question you ask after it.
• Builderroll-back logic. What was the last change you rolled back? Write what tipped you.
• Architectbuild-buy-defer. What was the last component you chose? Write the two you turned down.
• Orchestratorescalate-or-hold. What was the last thing you escalated? Write what made you pull the trigger.
• Strategistfund-or-kill. What was the last initiative you killed? Write the signal that killed it.
• Process Ownerapprove-or-reject at a gate. What was the last change you rejected? Write the criteria it failed.

The miss is the inner voice, mapped onto the page so you can see where it has gone quiet.

The Concept. Teach one thing you use constantly and have never written down, by hand, as if to a new hire.

• Supporterwhat DNS does, or why a reset will not fix a token. What did you explain badly to a user this week?
• Builderidempotency, or why a health check has to mean something. What did a junior ask that you fumbled?
• Architectcoupling, or a consistency trade-off. What do you keep re-explaining in reviews?
• Orchestratorwhat your SLA actually commits to. What policy does your team keep asking you to justify?
• Strategistwhat a headline metric actually measures. What number is in every board deck? Open it up.
• Process Ownerwhy a control exists. What rule do your stakeholders push back on most?

The page tells you what you actually own. Most find it is 60 percent. The other 40 is a study list — and a study list is the cure for the feeling.

System, Decision, Concept. Run them in the calm hours. By the third month the inner voice is loud enough to catch what the AI gets wrong, and the inner picture is sharp enough to hold your system without the dashboard. By the third year the model in your head is the one the room comes to when the building catches fire, whether your room is a NOC or a boardroom.

The three above are the starter set. The full cadence, the inspection protocol, and the longer-form practices live outside this essay. CTRL ALT PRESS runs them as a workshop. The essay is free. The workshop is for operators who want the rest of the kitchen table.

Let me say the part most publications will not say at the end of a piece like this. The next three to five years in IT operations are going to thin the field, and the line they thin along is already visible in the data we walked through.

The tool is going to keep getting faster. The clock is going to keep getting shorter. The room is going to keep rewarding throughput over judgment, because throughput shows up on the dashboard and judgment does not. None of those forces are going to reverse. Betting against them is a fool's bet, and I'm not into making fool's bets.

So here is the bet I will make, with my name on it, in public, in the open.

Today, before you close this tab — pull one blank page and draw one system you own, from memory. Check it. The gap you find is the head start, if you start now.

Build and operate AI-augmented systems without magical thinking.

Chris, logging off.