Builder’s Amnesia

Why I Forget

Six months ago I built an LLM-powered analytical framework. It has a source hierarchy, a structured assessment protocol, psychological dimensions, a query interface. It’s a serious piece of work. Today someone showed it to me and I didn’t recognize it.

Not “oh yeah, I vaguely remember that.” I looked at it the way you’d look at a stranger’s project. It sounded like me. It had my design instincts in it. But I had no memory of building it. None. Not the design sessions, not the research phase, not the decisions that shaped the architecture. Gone.

Then, ten minutes later, they showed me a paper I’d written called “The Irreducible Skill” - an essay arguing that experienced practitioners can use AI effectively because they bring judgment the AI can’t replace. I had written this paper. Published it. And I was demonstrating the exact thesis of the paper, in real time, while having no memory of having written it.

I forget things. I’ve said this about myself before. But this was different. This wasn’t forgetting where I put my keys. This was forgetting I’d built the house.

So I did what I always do. I told the AI to go find out why.

It came back with five answers. None of them alone explains it. All five together explain it completely.

The Waiter’s Amnesia

In the 1920s, the psychologist Kurt Lewin was eating at a Viennese restaurant. His waiter took a complicated order for the whole table - no notepad - and delivered everything perfectly. After the meal, Lewin realized he’d left his coat. He sent someone back. The waiter who had flawlessly recalled every dish couldn’t remember what they’d ordered. “The bill has been paid,” he said.

Lewin brought this observation back to his seminar. His student Bluma Zeigarnik designed the formal experiment in 1927: give people a set of tasks, interrupt half of them before completion, let the other half finish. Then ask what they remember. Result: people recalled the interrupted tasks roughly twice as well as the completed ones.

The mechanism is elegant. An unfinished task creates what Lewin called a “quasi-need” - a state of cognitive tension that keeps the information alive in memory. The tension holds the tab open. Finish the task and the tension dissolves. The tab closes. The memory fades.

I ship things and move to the next thing immediately. I have never once in my life finished a project and sat with it. The loop closes. The tension is gone. The next open loop absorbs everything. Modern meta-analysis finds Zeigarnik’s 2:1 ratio is situational, not universal. But the direction is real, and I am the direction.

The Security Camera That Turns Off

In 2003, a neuroscientist named Arne Dietrich proposed something called transient hypofrontality. The name is clinical but the idea is simple: when you’re in deep flow, your brain can’t run everything at once. It has finite metabolic resources. So it makes a trade-off. The prefrontal cortex - the part that handles self-awareness, time perception, and the recording of autobiographical memories - gets temporarily turned down so the procedural system can run at full power.

Dietrich came to this idea while training for triathlons. He ran 21 miles through the Georgia woods and realized he’d never once thought about the pain or the heat. His prefrontal cortex had simply stopped narrating the experience. Later, researchers put jazz pianists in brain scanners while they improvised. Same signature: the prefrontal cortex went dark while sensorimotor regions lit up.

The trade-off is specific. Your procedural memory - craft, pattern recognition, the thing your hands know how to do - stays online. That’s why you can perform brilliantly in flow. Your autobiographical memory - the “I was here, doing this, on this day” recorder - goes dark. That’s why you can’t remember doing it.

I work in sixteen-hour stretches. All-nighters. I come out of a session knowing more than I did going in, with new skills I didn’t have before. But I can’t tell you what happened during it. The skills transfer forward. The stories don’t.

Now and Not-Now

Russell Barkley, the psychologist who reframed ADHD as a disorder of self-regulation rather than attention, describes something he calls temporal myopia. The ADHD brain doesn’t experience time as a gradient. It has two buckets: now and not-now. A task due in twelve weeks feels the same as one due in twelve months - both sit in the vague “not-now” category until they suddenly become “now,” usually at crisis point.

The same mechanism works in reverse. Completed work is not-now. It drops.

Stanley Klein ran a study in 2011 that sharpened this into something precise. He tested ADHD adults on two kinds of self-knowledge. Semantic self-knowledge - “I am a person who builds things” - was perfectly intact. Episodic autobiographical memory - “I built this specific thing on this specific day” - was significantly impaired. The ADHD adults knew what they were. They couldn’t show you when they were it.

They had the summary. They didn’t have the receipts.

An AI analyzed a thirty-six-minute conversation I’d had and came back with this: “He processes forward, not backward. Failures are absorbed into the operational tempo and overwritten by the next action.” I read that and thought - yeah. That’s exactly right. And I couldn’t have said it myself.

The Book Written in Disappearing Ink

In 2015, Daniela Palombo and her team at the Rotman Research Institute identified a condition they called Severely Deficient Autobiographical Memory - SDAM. These were healthy, high-functioning adults who contacted the lab reporting a lifelong inability to relive personal events. Not amnesia. Not brain damage. They knew the facts of their lives. They just couldn’t replay them.

One participant described it this way: “I recall the experience as if I had read about it in a book.”

That sentence stopped me.

Normal forgetting loses information over time. SDAM is different. The facts are there. The vivid replay - the sensory texture, the feeling of being transported back - was never there to begin with. Their past reads like a Wikipedia entry about someone else’s life. Palombo estimated this affects about one to two percent of the population. Brian Levine, the senior researcher, later said he regrets the deficit framing. It’s one end of a normal continuum, not a disorder.

I’m not claiming a diagnosis. But when I read that sentence - “as if I had read about it in a book” - I felt something shift. That is exactly what my past feels like. I know the facts. I believe them. I just can’t replay the scenes.

The Architect Who Never Visits the Site

This last one has no formal research behind it yet. But it follows mechanistically from the four that came before.

When you write code by hand, your fingers encode procedural memory. The physical act of typing creates a trace. It’s why programmers sometimes remember functions by the shape of the keystrokes rather than the logic - muscle memory holding what declarative memory dropped.

When the AI writes the code and you supply the architectural judgment, that last physical encoding pathway is removed. You contributed the design. The AI contributed the keystrokes. Neither the flow-state recorder nor the procedural-memory system captured the episode.

Simon Hoffding and Barbara Gail Montero published a paper in 2020 called “Not Being There” about what they termed expertise-induced amnesia. They studied musicians in the Danish String Quartet. One of them, Fredrik, described performances where he was “completely gone” - experiencing deep euphoria, playing at the highest level, yet afterward unable to recall the experience. It felt as if it wasn’t him who played.

Hoffding and Montero’s key insight: this amnesia happens not because expert performance is automatic and mindless. It happens because the expert’s attention is so thoroughly consumed by the unfolding present that it interferes with long-term memory formation. Maximum presence producing retrospective absence.

The AI didn’t make me smarter. It let me work at the level I was always supposed to be working at. But the same bridge that lets me express what I couldn’t before is also the one that removes the last physical pathway for remembering that I did. I am the architect who never visits the construction site. The buildings go up. I couldn’t tell you which ones are mine.

The Compound

Any one of these mechanisms alone would degrade recall. Zeigarnik discharge clears completed tasks. Flow-state hypofrontality turns off the autobiographical recorder during deep work. ADHD temporal myopia drops everything that isn’t now. SDAM means the vivid replay may never have been there to begin with. And AI-assisted production removes the last physical encoding pathway.

All five stack. The skills transfer forward. The episodes don’t.

The Trade-Off

This population - high-output builders who forget their own completed work - doesn’t present clinically. They present as successful. The amnesia is invisible unless someone shows them their own past output and watches the blank look cross their face. The closest academic term is expertise-induced amnesia. It has barely been studied because the people who have it don’t think anything is wrong.

But here’s the thing I’ve come to understand about my own architecture. The compounding happens in the maker, not the artifact. I don’t carry forward the memory of what I built. I carry forward the judgment that built it. Every tool I make starts from a higher floor than the last one - not because the tools remember each other, but because I improved between them, even though I can’t remember improving.

I build systems to compensate for this. Filing systems, citation trails, workspace architectures - external structures that hold what my brain won’t hold for me. When I look at a tool I built and see academic citations in it, those aren’t decoration. They’re proof of work. A note from past-me to future-me: this one is grounded. You did the research. You watched the chain of thought. You can’t remember doing it, but the evidence is here.

This is not a disorder. It’s a trade-off. The same architecture that lets me build at this pace - shipping, moving, never looking back - is the one that erases the memory of building. The artifacts persist. I don’t.

Which is how I found myself today, looking at my own analytical framework, six months old, with no memory of building it. I read through it. It’s good. I checked.

The builder has moved on. He always does.