Seam №001 · Essay · July 13, 2026 · 9 minInteractive experience added July 13, 2026
You Can Just Build Things
On thinking rocks, forbidden skies, and what we do with the knowledge we inherit.
scroll ↓
There is a dangerous way to get used to a miracle.
You put it in your pocket.
A black rectangle wakes under your thumb. Billions of microscopic switches become your camera, map, library, and a way to reach someone you miss.
We call this normal.
Look at what had to exist for it to feel that ordinary.
Almost everything I use rests on a stack beneath it and a chain behind it. The stack is what had to exist first: materials, tools, skills, standards, and people doing careful work. The chain is how that work survived long enough to reach me, through observations, experiments, corrections, copies, and teaching.
I kept noticing both in four things I watched and read: a Marques Brownlee video about chips, Dr. STONE, the DeepMind documentary The Thinking Game, and Orb: On the Movements of the Earth. They made the finished world feel less sealed. I could see more of the work, and more of the people, behind it.
That changes what “you can just build things” means to me. You can begin without mastering everything your work depends on, because other people have already made parts you need. Using their work also gives you something to look after: check it, credit it, and leave it usable for whoever follows.
01
The seam beneath the miracle
Marques Brownlee’s video about shrinking into a chip makes its scale feel physical. A chip the size of a fingernail can contain billions of transistors, tiny electrical switches. Some features measure only a few dozen nanometers. A red blood cell measures about 7,000 nanometers across. A nanometer is a billionth of a meter.1 My brain can say that. My brain cannot feel it.
Brownlee closes the gap with a comparison. If a transistor grew to the size of a car, the phone holding billions of them would grow to the size of the Earth.
The labels need care. Apple calls its M5 a third-generation 3-nanometer design. “3-nanometer” names a manufacturing generation, rather than the width of each transistor. The actual device contains many different dimensions.2
Now watch one get made. Chips are built on a wafer, a thin, mirror-smooth disc of silicon. Their patterns go on in layers through repeated steps that add, remove, and alter material. Light transfers patterns onto a light-sensitive coating in a process called lithography. At the end, the wafer is cut into individual chips.3
ASML makes machines for printing some of the smallest patterns. To produce extreme ultraviolet light, a machine shoots a falling droplet of molten tin with a laser twice. The first pulse flattens it. The second turns it into a plasma, a gas of charged particles that emits the light. The machine repeats this up to 50,000 times a second. Glass would absorb that light, so the optics use mirrors inside a vacuum chamber.4
Somebody built that. On purpose. And then they built more of them.
A chip is a treaty between strangers. Designing a transistor, polishing a mirror, and operating a production line are different jobs. Shared measurements and standards let the work of one team fit with another’s.
I did not build any of this. I can barely explain it. I use the result without seeing the tolerances, the night shifts, or the prototypes that failed. We inherit the finished verbs: call, render, search, translate, and forget the work underneath.
Looking at all that machinery can make building feel impossibly remote. But the division of work is what lets someone participate without understanding every layer. You can write something that runs on a chip without learning to manufacture one. The work beneath your project can be enormous while your first contribution remains small.
The world has seams everywhere.
02
Civilization, run backward
Dr. STONE, by Riichiro Inagaki and Boichi, opens with a flash of green light that turns humanity to stone. Thousands of years later, Senku and Taiju wake in a world where the cities have crumbled and the forests have come back. They set out to revive everyone and rebuild civilization through science.5
To work metal, Senku needs a furnace. Before the furnace can do its job, he needs fuel, suitable materials, and people to work the bellows. Every finished object opens into the work required to make it. Slowly, he builds the tech tree of the human race back up out of the mud, and I could not stop reading.
Senku can just build things. He is also drawn. He knows an absurd amount of science, and the plot wants him to succeed. The rest of us discover how much we are missing when we try to make something as familiar as a light bulb: a filament, glass, a power source, and tools and skills to put them together. Every prerequisite opens into more prerequisites.
Senku’s useful habit is seeing dependencies, the things that must be in place before the next thing can work. He looks at a finished object and reasons backward. What has to exist first? What can we make with what we already have? Which missing piece matters most, and who knows something I do not?
The series hands you a genius, then makes him depend on everyone around him. Knowledge does not hammer metal. A formula does not carry lumber. The Kingdom of Science works because it becomes a crowd of people who are each good at something else.
You do not need the whole stack before you can add to it.
Once they can bring people back, another problem appears. Senku wants to revive everyone. His rival wants to decide who deserves to return. Both plans use the same chemistry. Who benefits depends on the people controlling it.
03
A fair way to be wrong
The Thinking Game, Greg Kohs’s documentary about Demis Hassabis and DeepMind, follows researchers trying to predict protein structures with AlphaFold.6 A protein is a chain of amino acids that folds into a three-dimensional shape. That shape helps determine what it does. Working it out experimentally can take months or years for a single protein.
The researchers had to submit their work to a test they did not control. CASP is an independent assessment in which teams predict protein structures before the experimental answers are made public. Assessors compare the predictions with those answers. The test gives researchers a way to find out whether their confidence is deserved.7
At CASP14 in 2020, AlphaFold achieved a median score of 92.4 out of 100 on a measure of how closely predicted structures match experimental ones.8 The 2021 paper also reported limits. Predictions could be less reliable when few related protein sequences were available, or when a protein’s shape depended on interactions with other chains.9
The work then became available beyond the laboratory. DeepMind released AlphaFold 2 openly in 2021. By 2022, its public database with EMBL-EBI held more than 200 million predicted structures. Researchers could use those predictions, with confidence estimates, as starting points for further work.10
That decision matters to me as much as the score. It made a capability available to people who had not built it. They still needed to understand its limits and test what they did with it, but they could begin further along. Work that others can inspect, question, and build on becomes part of the next person’s inheritance.
04
The protagonist is transmission
Orb: On the Movements of the Earth, by Uoto, is historical fiction. In its imagined fifteenth-century Europe, investigating whether the Earth moves around the Sun can get you tortured and killed. People study in secret and hide their notes somewhere a stranger might find them.11
The story keeps changing main characters. The notes keep moving, the people carrying them keep dying, and I have never read anything with this much nerve.
The handoff is what had me gripping the book. An idea goes from a dying person to a stranger. The person leaving it behind cannot know whether it will be understood, believed, or kept safe. The next person has to decide whether to risk their own life for work they did not begin.
A truth can be correct and still vanish. Being right offers no protection from the people trying to suppress it. Sometimes courage is making a copy.
Knowledge has to be made survivable. It needs diagrams, books, demonstrations, archives, teachers, and people willing to protect those who carry it. The work includes keeping a question alive long enough for someone else to correct the answer. The person who preserves it may never be the one who solves it.
This gives the chain behind everything I know a different weight. I inherit the work of people whose names I may never learn, including people who preserved something they did not discover. We arrive in the middle of a conversation whose first speakers we cannot name.
Dr. STONE runs civilization backward, from an object to its prerequisites. Orb runs knowledge forward, from one person to the next. One asks what had to exist before this. The other asks who carries it after me.
05
Who pays for it
Everything I have just told you, I told you like a fan.
A fan’s mistake is to treat cleverness as innocence: to see a beautiful machine and assume the system around it is good. Making something extraordinary tells you little on its own about who bore the cost, who shared the reward, or who consented.
Sharing AlphaFold’s predictions was valuable. It does not settle every question about the company behind it or the uses of AI. I can admire that decision and still ask who controls the tools, who can afford to use them, and what happens when they fail.
Those questions should change what I do with my own work. Before asking someone to rely on a tool, I owe them an account of its limits. If it depends on their data or labour, their permission belongs in the plan. An experiment is less harmless when someone else has to absorb the mistake.
06
The tools that speak
In August 2025 I wrote an essay asking whether a piece of writing was made with AI. Its central argument still holds: use the tools and stay answerable for what they make. The person publishing the work has to explain their choices and deal with the consequences.
Slop is fluent output with no care behind it. What worries me is how easily a finished-looking answer can let us skip deciding what we think. A paragraph can sound convincing before anyone has checked what it says. A tool that helps me write should still leave me willing to question the result.
I also want someone who thought they had nothing to make to discover that they can. Help with writing, code, or a first attempt can make starting less intimidating. I refuse to be sour about that possibility.
The obligation travels with the inheritance. Credit the source. Respect consent. Pay people where their work creates value. Check the claims and decisions that others will rely on. When you cannot check an important part yourself, find someone who can before asking people to trust it.
Own what you make.
07
Leave a rung
The first version can be embarrassingly small: a rough script, a drawing, a repair, or a paragraph that finally says what you meant. Curiosity is enough reason to start. You do not have to turn everything you make into a product.
Rung 01 / Follow one thread
Pick something you want to make or understand. Spend ten minutes tracing one part of its stack or its chain. What does it depend on? Who worked it out, and where did they leave their explanation? Find one existing tool, source, or person that can help you begin.
Rung 02 / Build one thing
Make a first version small enough to try. Suppose people keep asking how to use a tool you know. Write a short guide with one worked example, then give it to someone new. Watch where they get stuck and revise that step.
Rung 03 / Leave a map
Write down what worked, what failed, and what the next person should know. Name the people whose work you used. Keep an example they can follow and leave it somewhere they can find it. Make it possible for someone else to continue without needing you beside them.
The just in “you can just build things” gives you permission to begin before you feel ready. It makes no promise about difficulty. Take hold of the part within reach, treat it honestly, and pass it farther than you found it.
Notes and sources
- ASML, “The basics of microchips.” Explains transistor switches and compares chip features with a red blood cell, approximately 7,000 nanometers in diameter. The video that prompted this section is Marques Brownlee’s “I shrunk down into an M5 chip.” Brownlee ends the scale journey with the car-to-Earth comparison used in the text. Both comparisons translate scale. Neither gives the dimensions of every transistor.↩
- Apple, “Apple unleashes M5, the next big leap in AI performance for Apple silicon,” October 15, 2025. Identifies M5 as a third-generation 3-nanometer design. For the distinction between a manufacturing node and a physical measurement, compare ASML’s “5 things you should know about High NA EUV lithography,” January 25, 2024. It describes 8-nanometer printing resolution for machines intended to manufacture 2-nanometer-node chips.↩
- ASML, “How microchips are made.” Describes wafer processing, the light-sensitive coating called photoresist, lithography, and the repeated steps used to form a chip’s layers. It also gives the figures in the chip scene: up to 100 layers aligned with nanometer precision, and hundreds of steps taking up to four months from design to mass production.↩
- ASML, “Light & lasers” and “Lenses & mirrors.” Explain the two-pulse tin-droplet light source, its rate of 50,000 cycles per second, and the use of mirrors inside a vacuum chamber.↩
- Riichiro Inagaki and Boichi, Dr. STONE. VIZ Media’s “The Official Website for Dr. STONE” credits the creators and introduces Senku and Taiju’s effort to restart civilization. The discussion of dependencies, teamwork, and the revival conflict is my reading of the series, rather than a set of instructions for reproducing its inventions.↩
- Tribeca Festival, “The Thinking Game,” 2024. Identifies director Greg Kohs and describes the documentary’s focus on DeepMind and AlphaFold. For the relationship between protein structure and function, and the time experimental work can require, see DeepMind’s 2020 explanation in note 8.↩
- Prediction Center, “CASP14: Critical Assessment of Techniques for Protein Structure Prediction,” 2020. Describes predictions made before experimental structures are public and assessment by independent researchers. Experimental results may still be forthcoming when a target is issued; the essential condition is that participants cannot see them. Approximately 100 research groups submitted more than 67,000 models across 90 targets.↩
- DeepMind, “AlphaFold: a solution to a 50-year-old grand challenge in biology,” November 30, 2020. Reports a median Global Distance Test (GDT) score of 92.4 across CASP14 targets. GDT measures structural agreement on a scale from 0 to 100. This is not the percentage of proteins fully solved, and it is not a CASP13 result.↩
- John Jumper et al., “Highly accurate protein structure prediction with AlphaFold,” Nature 596 (2021): 583–589. The section “MSA depth and cross-chain contacts” discusses the limitations described here. These observations concern the system reported in that paper, rather than every later AlphaFold version.↩
- DeepMind, “AlphaFold reveals the structure of the protein universe,” July 28, 2022, and the EMBL-EBI and DeepMind “AlphaFold Protein Structure Database.” The announcement recalls the open release in 2021 and reports the expansion to more than 200 million predicted structures in 2022. Database predictions have varying confidence and are distinct from experimentally determined structures. The discussion of open release here concerns AlphaFold 2 and this database.↩
- Uoto, Orb: On the Movements of the Earth. Seven Seas Entertainment, “Orb: On the Movements of the Earth (Omnibus).” The discussion concerns the fictional story and its transmission of astronomical work. Its persecution of heliocentrism should not be read as an account of specific fifteenth-century prosecutions.↩