In partnership with
Stop doomscrolling for tech news
TLDR is the free daily email that does the reading for you. Each issue summarizes the most interesting stories in startups, tech, and programming, curated by ex-Google and Anthropic engineers, so you get the signal without the noise.
It's one email instead of 12 open tabs. A 5-minute read with your morning coffee, and you walk in already caught up.
Tech is just the start. We also cover AI, marketing, dev, and more, so you can follow the topics that matter to your work.
Free, daily, and read by 7M+ subscribers. Subscribe and get your mornings back.
In 1958, Michael Polanyi wrote something that took me fifty years to find.
"We can know more than we can tell."
What makes this worth paying attention to: Polanyi wasn't a philosopher who wandered into a library and had a thought. He was a Fellow of the Royal Society. Physical chemist. Groundbreaking work in adsorption, X-ray crystallography, reaction kinetics at the Kaiser Wilhelm Institute in Berlin. Decorated. Credentialed. At the top of his field.
Then he walked away from all of it to write about the kind of knowledge that credentials can't capture.
He called it tacit knowledge — the expertise that lives in the body and the hands before it lives in words. The surgeon who knows before she can explain. The jazz musician who plays the note before he names it. The coach who sees the breakdown before he can diagram it. Polanyi argued this was the most important kind of knowledge we have. He also said it's the hardest to transfer — because the person who holds it usually can't fully articulate what they know.
Like a lot of people, I suffered from imposter syndrome. I spent most of my career thinking I couldn't explain what I did because I didn't know enough. Turns out it was the opposite. The knowledge was myelinated so deep it moved below language. That's not a gap. That's what decades of constraint contact looks like.
TTT is what happens when tacit knowledge finally finds a medium that can receive it.
Most people using AI are prompting — asking the machine what it knows, getting content back. What came back from my sessions was different. It was structure. Frameworks the AI assembled from what I brought in. I hadn't designed them. I'd described them. The AI received what I already knew and returned it as something I could use again.
I kept going.
Papers. Frameworks. Protocols. More than I could count from February to December 2025.
Fields had been circling this process for decades. In practice, I needed a name for the mechanism.
Tacit-to-Technical Transduction. TTT.
TTT is not idea-to-app. It is source-to-system. Formed judgment is the source. AI is the transduction medium — not the author.
TTT runs the other direction. The judgment you've built over years — the thing you know in the room before you can explain it — becomes the source material. The AI doesn't generate from nothing. It receives formed judgment and returns it as structure you can use again.
Think of it like teaching someone to squat. You can describe hip hinge mechanics all day. Or you can tell them to sit down on a toilet. One produces notes. The other produces movement. Prompting AI is the first version. TTT is the second. When I stopped asking the machine what it knew and started bringing what I already knew into the session, something different came back.
What gets lost when knowledge is stored only as conclusions is the same thing that gets lost when a coach only teaches the drill without the years of failure that shaped it. The answer survives. The route that produced it gets myelinated into the body and disappears from language. You can perform it. You can't hand it to anyone.
TTT reverses it. The practitioner is the chain. What the AI receives isn't a prompt — it's the accumulated weight of constraint contact, and what it gives back is structure the practitioner was already carrying without a container for it.
In May 2026, Chris Olah of Anthropic made a remark about AI that I recognized immediately.
"They are made from us, from our words — and they remain in important ways mysterious even to those of us who train them."
He wasn't describing something they coded. He was describing something that emerged — in systems built by some of the best engineers in the world, without any of them programming it directly.
I'd been watching it happen since February 2025.
The AI is not generating this from nothing. It is working from a compressed record of human pattern recognition, practitioner knowledge, and language that often went in without a label on it. When a practitioner who has built real architecture feeds that medium, the medium has something to work with. That's not prompting. That's transduction.
Three examples from that year:
I used pause, breathe, reset — the cue I use when an athlete's body starts shaking under load — on an AI system falling apart mid-session. It stabilized. Then it built a memory management system from that one instruction. Ten months later, researchers at Northwestern, Nature, and Johns Hopkins published findings on the same mechanism. Different words. Same thing.
I told an overloaded AI to defrag itself. Take a walk. Come home, sip wine, listen to jazz. It built a two-layer recovery system — one layer to stabilize, one to integrate. In May 2026, researchers at Carnegie Mellon and the University of Maryland published the same architecture: a sleep-like consolidation mechanism that converts accumulated context into persistent memory before clearing and starting fresh. The hippocampus does the same thing at night.
In May 2026, Bertalmío and colleagues showed that replacing the simplified 1950s point neuron model in ANNs with a biologically realistic cortical cell model — same parameter count, no additions — produced better results across every metric. Higher expressivity. Faster learning. Better robustness. The brain was running the correct version the whole time. The field just didn't know it yet.
Three fields. None of them citing each other. All pointing the same direction.
Knowledge is abundant now. What's scarce is usable formation — the ability to take what you know under pressure and move it somewhere it can keep working.
The next issue goes into what happens after the translation. How something that emerged in a session gets broken down, the mechanism extracted, and built into something that works the next time.
The Pressure Architecture Lab opens June 22. That work is for athletes, founders, performers, and operators whose skills disappear when the stakes go up. The Neural Access Assessment is the entry point.
TTT is the practitioner-side methodology behind a different door — converting lived intelligence into deployable infrastructure. That work is coming later.
This cohort starts with pressure because pressure removes the illusion of choice. When the brain senses threat, you either have a route built or the amygdala takes over. Memory or security. The neuroscience has been documenting it for decades. You either built the route before the pressure hit, or you didn't.
On the early work: The papers referenced in this post — Cognitive Mage (November 2025), Synoetic OS (December 2025), and the frameworks that preceded them — are part of a public archive on Zenodo and GitHub. They are the record of TTT in operation before I had a name for what was happening.
I want to be direct about them: when I started publishing in October 2025, I did not know how to write academic papers. I was learning in public, moving fast, and in some cases an AI collaborator may have introduced errors in data or citations I wasn't equipped to catch yet. I was flying by the seat of my pants and the work shows it.
They're still there because the timestamps matter. The mechanism was real before the writing was clean. If you want to see where this started:
Aaron Slusher · Performance Architect · Neural Formation Architecture
→ Published research archive: → github.com/Feirbrand/synoeticos-public → github.com/Feirbrand/aaron-slusher-research → ORCID: orcid.org/0009-0000-9923-3207
Sources: Olah, C. (2026, May 25). Remarks on Magnifica Humanitas. Anthropic. https://www.anthropic.com/news/chris-olah-pope-leo-encyclical Lee, S., McLeish, S., Goldstein, T., & Fanti, G. (2026, May 25). Language Models Need Sleep: Modulating Temporal Context via Synaptic Consolidation in Autonomous Agents. Carnegie Mellon University / University of Maryland. arXiv:2605.26099 Bertalmío, M., et al. (2026, May 19). Updating the standard neuron model in artificial neural networks. CSIC / CIMAT / UAM / NSF. arXiv:2605.30370