About The Durability Curve
Where value migrates when intelligence becomes cheap.
A research project about where value migrates as artificial intelligence commoditises the layers beneath it.
Most AI commentary asks what the technology can do. This project asks the question underneath it: when the cost of intelligence collapses, what stays expensive, and who keeps the margin?
I run a research system that turns AI, product, and market noise into instruments for seeing what survives when the surface changes.
The Durability Curve runs on that system. It reads far more than I can, tags what survives contact with real deployment, and surfaces the contradictions I would otherwise miss. The essays are the visible part. Underneath sits a vault of thousands of notes that argue with each other, and the five laws below are the patterns that kept holding after the arguments settled.
The lens is loudest in AI, because that is where the ground moves fastest, but the same pattern runs through product strategy, markets, capital allocation, and any system that has to outlast its own context.
Harry Floyd
The five laws
Five claims do most of the work here. Each one is stated so it can be argued with, and each links to the section of the flagship essay that tests it.
-
Bottleneck Migration
Value moves to the most stubborn adjacent layer. Usually upward as the layer below commoditises, and downward to the physical substrate, meaning compute, power, fabs and access, when the scarcity is physical.
Near · 1–5y -
Difficulty Is Load-Bearing
The hard part is the mechanism that produces the value. Strip the difficulty out and the value leaves with it.
Structural -
Architecture Outlives Content
The scaffold persists while the content turns over. Structure is the moat; the output that fills it is replaceable.
Long · 5–20y -
Instruments Over Theory
Hidden structure stays hidden until you build the instrument that measures it. Theory points at the field. The instrument reads it.
Structural -
The Targeting Problem
Capability aimed at the wrong layer makes things worse. More power on the wrong target only enlarges the miss.
Structural
The method
An analysis with no law, no falsifier, and no named comparison is a vibes answer. That is the bar, and a piece that clears it shows the work on the page: the law in the byline, the falsifier in the text.
-
Declares
which law it stress-tests, stated up front.
-
Falsifies
the specific evidence that would break the claim.
-
Compares
two real entities with real numbers, so the claim is checkable.
Where to start
You do not need to read everything. Follow the problem you brought.
-
You are new here
Start Here is the shortest map of the lens, and it routes you onward by the problem you actually have.
-
You build AI systems
Most Verification Is Just Bigger Classification names the layer that decides whether an output counted, then Same Model, Different Product shows why the harness, not the model, is the product.
-
You allocate capital
The Forest Floor Is The Product argues the substrate case, and The Investor's Substrate Test runs it on a company you are looking at.
-
You are choosing what to protect
The Model Is Not The Moat is the argument, and The Substrate Map is the layout of where durable positions actually sit.
-
You want the vocabulary
The Lens Lexicon defines the ten load-bearing terms the essays keep reaching for.
-
You would rather run something
The rack holds 7 instruments you can point at your own stack, metric or decision. Free, no login, nothing sent anywhere.
What you get
- The rhythm
- Several essays a week, with the week’s main piece every Tuesday at 10am UK time, plus Notes through the week.
- Free
- Every essay on this site and every instrument on the rack, complete. No paywall, no opt-in, no login.
- Paid
- A tier on Substack for the deeper instruments, the ones you run on your own numbers, and the full archive.
Questions
Who writes it?
Harry Floyd. I run a research system that turns AI, product, and market noise into instruments for seeing what survives when the surface changes. The essays are the visible part of it.
How often does it publish?
Several essays a week, with the week's main piece every Tuesday at 10am UK time, plus Notes through the week.
What does it cost?
Every essay on this site and every instrument on the rack is free and complete, with no paywall and no opt-in. A paid tier on Substack carries the deeper instruments, the ones you run on your own numbers, and the full archive.
Do the instruments need an account?
No. Each one runs in your browser, takes no login, and reads only what you give it. A few save your work to that browser so you can come back to it. Nothing is sent anywhere.
How do I know a claim here is any good?
You check it. The bar is that an analysis names the law it is testing, states the evidence that would prove it wrong, and compares two real entities with real numbers rather than arguing in the abstract. Where a piece clears that bar it shows the work on the page: the law in the byline, the falsifier in the text. Where it does not, you are entitled to say so.
Is any of this investment advice?
No. Everything here is research and education, not personalised investment advice. I disclose positions for accountability, not as recommendations.
The allocation principle · Law I × Law III
Own the durable architecture at the moment the bottleneck is migrating toward it.
Subscribe to get the next structural lens in your inbox.
Everything here is research and education, not personalised investment advice. I disclose positions for accountability, not as recommendations.