The Record You Keep Comes With You

A piece by Tawnya Means this week has had me thinking. She tells the story of Christopher Noe, an MIT Sloan lecturer who handed an AI a stack of his old teaching notes and got back a draft that seemed, in his words, to read his mind - down to an ice-breaker question and a tax aside he had never actually typed out. As a professor of business, I’ve had this experience, too, and I want to build on one thread in the article rather than add another round of applause.

The detail I keep returning to is where the “mind-reading” came from. Not from a long-accumulated relationship with the tool - that two-year account was not even in the room for this task. It came from the explicit artifacts Chris handed over in a single sitting: an older case paired with its note, and a few others. The system inferred his moves from documents, not from history. And that, to me, is the hopeful part of the story, because it means the thing that felt like magic is portable. It could have happened on any platform he fed those same notes.

I want to be careful about what actually transferred, though. A lot of what the system caught was the shape of Chris’s notes - their structure, the opening move, the order in which he builds toward the hard question. That is a real thing to reproduce, and it is not the same as reproducing his judgment about what is worth teaching or whether a claim holds up. For a piece about learning, that gap is the whole game. A student can absorb the form of expert reasoning, the confident structure and the well-placed question, without the substance underneath it. The Sovos note worked because an expert with the substance was standing right there to supply it. Take him out of the room and you can be left with something that has the silhouette of judgment and none of the weight.

Tawnya raises the flip side that concerns many of us - that systems accumulating context about you get hard to leave, and that the difficulty is the point. I think that is true as friction, and worth taking seriously. But her own story suggests a gentler reading. What the system “remembers” about us is a set of compact summaries it has written, not a bond it has formed. The workaround in the piece points right at it: you can ask the tool to summarize what it knows about you into a profile and carry that across. A memory you can write down and move was never a moat. It was a portable document the vendor happened to be holding.

It is summaries all the way down, honestly, for the AI and for us. A colleague who has known me ten years does not replay every conversation - they carry a compressed model of me, and when they are swamped, they forget things too. The AI is the same, except its compression is explicit and editable. That strikes me as an advantage. You can read it, correct it, version it, move it.

So the habit I would draw out of Chris’s experiment - and push just one step past what he did - is this. Externalizing your judgment is not only an input move, handing the system your old notes so it can learn your moves. It is also something worth asking for on the way out. In the middle of the work, have the tool state plainly what the two of you have actually co-created: the decisions you reached, the reasoning behind them, the pattern it thinks it is following. That artifact does triple duty. It is a checkpoint, so you can catch drift before you are ten pages deep. It is a mirror, because seeing your own thinking written back to you often surfaces something you had not quite articulated - a smaller version of Chris’s “reading my mind” surprise, pointed inward. And it is portable, because now the record of how you think lives in a file you own.

One caution about that mirror. It can flatter. A fluent summary of what the two of you “decided together” is easy to nod along to, and some of what it reflects back will be the model’s confident guess wearing your voice. So read it as a draft to check against what you actually meant. The value is in catching where it is wrong, not in the reassurance that you were right.

I should be honest about who this works for. Chris could tell in a single morning what was right, what needed his hand, and which of the system’s own flags to trust, because he had been writing these notes for years. The habit I am describing assumes that kind of judgment is already in the room. A student, or an instructor teaching the material for the first time, cannot externalize judgment they have not built yet, and cannot catch the errors a confident draft will hand them. Tawnya raises this open question in her piece, and I do not have a clean answer to it either. Portability helps the expert most and the novice least, which is backwards from where the need is greatest. So take the hopeful part of this as a note for people who already have the goods, and treat how we get everyone else there as unfinished work.

None of this diminishes the wonder Tawnya is pointing at - if anything, it is what makes the wonder repeatable. The same discipline that made Chris’s teaching note so good works in both directions: give the system an explicit record of how you think, and get in the habit of asking it to give one back. Do that, and the question of which tool “knows you best” starts to matter a lot less. What knows you best is the record you have kept - and that comes with you.