No one mourns the calculator.

Wisdom that outlasts the algorithm.

──────────────────────────────

Last week I gave you what 51 of the world's most-trusted institutions decided about AI: machines get the work you can cheaply check, humans keep the work you can't.

Then the replies came, and nearly all of them asked the same question. Fine. You looked at Nature and BBC and the NY Times. But what can I do at my own desk on Monday?

Start here.

The truth is: you already do this. You have been doing this your whole life.

You write things down because memory fails. You use a calculator to check accuracy. If you have ever trusted a Wikipedia article without chasing a single footnote, you handed your thinking to a stranger and got on with your day. That is cognitive offloading. It is ancient, it is mostly good, and nobody mourns the calculator. No one calls the notebook a crutch.

So the interesting question was never whether to offload.

Run the GPS test. If you are forty or older, you learned to drive by memorizing routes, reading signs, and occasionally unfolding a map across a steering wheel. Now you type the destination into a screen and you arrive on time, every time, navigating around traffic. Real gains, no doubt. But can you do it without the phone?

Here is what makes this moment different from the notebook and the calculator. Every offload before this one took something off your plate that was obviously not you.

Arithmetic. Note storage. Route-finding.

You can’t delegate a decision to a notebook.

These are the first tools to produce judgment-shaped output. They are fluent. They sound like conclusions. Which means for the first time, the offload can look exactly like thinking. And you can walk away from the exchange knowing more and understanding less.

So: shortcut, or scaffold?

A shortcut carries you past the work. A scaffold holds you up while you do it. Same tool, same prompt, different intent, and the gains quietly accrue.

So I made the choice explicit. One question, asked before I hand anything over: does the effort of this task outweigh the judgment it requires? I call it responsible cognitive offloading.

Start by sorting your work into four buckets:

Automate. Low effort, low judgment. Scheduling, formatting, sorting. If it needs neither your grind nor your judgment, set it up once and stop carrying it.

Offload. High effort, low judgment. Summaries, transcription, research sweeps. Hand over the grind, keep a light review. You can harvest real hours here.

Co-create. High effort, high judgment. Strategy memos, keynotes, analysis. The machine drafts and pressure-tests, you steer and choose and own the result.

Reserve. Low effort, high judgment. The hire. The apology. The final call. Minutes of work, and every one of those minutes requires you.

Most people can sort their week in about ten minutes. The sorting is not the hard part.

The hard part is the question almost nobody asks afterward, which is the one I want to leave you with: how are you redeploying the capacity you just freed?

Offload the toil. Reserve the call.

──────────────────────────────

→ Half-Baked.
The corpus grew and the line mostly held. Last week the index stood at 51 institutions, 44 of them drawing the same permit line. As of this morning it is 55 and 46, and the fit rate has slipped from 86 percent to 84 percent. The most interesting addition is the Vatican, the first faith institution in the corpus, and it only half-draws the line: between a machine and a human being, it says, only the latter is truly a moral agent. Same instinct, arriving from theology instead of publishing ethics, landing at a different border. An emerging pattern is surviving new evidence it did not choose, and not all of the evidence agrees. One institution in there reverses it outright. It is a pattern, not a law.

→ Hot Take.
Your AI is not an oracle. It is a lens. An oracle you bring questions to and accept answers from. A lens you pick up and apply. The lens you choose decides the range and the sharpness and what gets filtered out of the picture. People who get disappointing results from these tools are almost always holding an oracle. Pick the lens on purpose, and stay in the frame.

→ Confession.
When I first encountered generative AI I decided it was a party trick. It wrote a custom poem, and I thought: cute. So what?!
Since then I have trained more than fifteen thousand people on this technology. I was flat wrong. I judged the demo instead of the trajectory. Worth remembering the next time something looks like a toy.

──────────────────────────────

Everyone has one. It is rarely the big thing. It is usually something small and quick where the judgment was the entire task.

You were tired, and the machine was right there. That inversion is worth catching because your credibility is on the line.

──────────────────────────────

The Monday Move.
Take your Monday task list and put one of four letters next to each item: A, O, C, R. Automate, Offload, Co-create, Reserve. Ten minutes, no tools, no app. You are not looking for a perfect sort. You are looking for the mismatches: the R you might have otherwise just handed over to AI. The O you have been grinding out by hand. Out of habit…or guilt.

The Asymmetric Move (this quarter).
Pick one thing you already offload today and name - in advance and in writing - what the freed capacity is for. Not "more time." A specific harder thing. Redeployment is the entire mechanism, and it is the step everyone skips, which is why so many people report using AI constantly and feeling no more valuable than they did a year ago. The tool gave them the hour. Nobody told them the hour was the point.

The Decade Move (this decade).
Crystallized intelligence, everything you already know, is exactly what these systems copy. Good, cheap and fast. Protect your fluid intelligence: the capacity to reason under new conditions. Fluid intelligence is built the slow way, through struggle and repetition and direct contact with problems that do not have a template. Intuition is not downloaded. It is earned.

──────────────────────────────

I gave a lightning talk on responsible cognitive offloading last October to the HealthAI Collective, hosted by Tintash, and it holds up. 31 minutes, and the last third is the best part, because the audience took it apart. One engineer made the case that you keep the framing and give the machine only the narrow task. Another argued that the invention of writing was a bigger cognitive shift than this one, and that we survived it. A third told me his junior developer said the quiet part out loud: when I code, I make mistakes and I learn. When AI codes, I learn nothing.

I have not published a polished framework page for this. That is coming, and this newsletter will get it first. In the meantime the talk is the fullest version that exists, unedited, arguments included.

→ Watch the talk: https://youtu.be/evVClQziWOA

→ Last week's institutional half of this: https://calibrated-authority.chrishuberreitz.com

Borrow freely. If you run the four-bucket sort on your own week, send me what surprised you.

──────────────────────────────

This newsletter is called Signals from the Curve because there are two kinds of forecasting: the curve and the cliff. The cliff says everything changes at once. The curve says it is already changing, you just have to know where to look.

Last week, 51 institutions drawing one line without ever meeting. This week, the same line at the scale of a single desk, which is the only scale where you control it.

One more thing. There’s something I did not resolve today. There’s a reason I insist on the word “responsible.” It is not decoration.

Offload everything and you eventually cannot make the calls you reserved. There is a real argument about what that costs and how fast. It lands here in just a few weeks.

If something here changed how you are thinking, hit reply. I will respond.

— Chris

Columbia Faculty · Chief of AI & Strategy at Essential Innovations · Founder, Attainable AI

chrishuberreitz.com · LinkedIn

Corrected 7/15: An earlier version of this issue said the Vatican lands on the same side of the line as a medical journal. It does not. The index scores it a partial draw, which is the more interesting result and is now what this issue says. Here’s the record.