Technologies that Bridge the Gaps Between Domains

As we grapple with AI, and with the vast variety of industries it will affect, the hard question is not whether this new tool will have vast society-altering impact: It will. Nor is the question whether AI will create a permanent underclass or dystopia. It will not. As I have argued previously, any such underclass or dystopia will last only so long as our society refuses to admit that our political structures must be updated to match the incredible and, in the main, incredibly positive technological shift that AI represents. Which might be longer than we hope, but it will not be as long as we fear, and certainly not forever.

The hard question is how do we think about the size of the field it creates? How many trades does it bridge, how far does mastery travel across those bridges, how large does a unified ranking system become once formerly siloed professions can compare specializations. We might call this field, the AI-space.

A generation ago, a different tool opened a smaller version of the same problem. I found reviewing the impact of that tool on the industries it touched instructive, and I hope you will as well.

I. The Puppet-Maker’s Hand

Our story begins in 2011, when Ivan Owen, a steampunk prop-maker in Bellingham, Washington, built a giant articulated metal hand for a costume, worked by cables, and posted a video of it flexing. The video went viral, and it caught the attention of a carpenter in Johannesburg named Richard Van As, who saw its potential to solve a different problem. He had just lost four fingers of his right hand to a table saw, and the prosthetics on offer cost more than he could pay. Inspired by the potential of the new technology, he began writing to the puppeteer.

For months the two men passed designs across nine time zones, first as sketches and photographs, then as printable files. Partway through the work, the mother of a five-year-old South African boy named Liam, born without fingers on his right hand, found their work and reached out to them. They built Liam a working mechanical hand on a desktop 3D printer, for roughly the cost of its plastic.

Owen released the designs for free instead of patenting them. Building on the virality of the earlier videos, the files spread, and the network that grew from that decision, e-NABLE now counts roughly forty thousand volunteers in over a hundred countries and has delivered prosthetic hands to more than ten thousand children.

A puppet-maker and a carpenter, neither with a day of medical training, did the work of a medical-device company and then out-distributed it. A generation earlier this story would have been impossible. The barriers to entry around prosthetics, like the barriers around every licensed trade, were enormous: the training, the equipment, the credentialing, the supply chains, were effectively insurmountable to non-credentialed makers without a massive research institute behind them. The new CAD toolkit leveled those barriers without even trying.

All Owen and Van As, and other players like them, needed was a level playing field to interact with, and they could compete with the largest names in the world.

A level playing field makes skill legible. Legible skills develop into ranks. As more players join these ranks, hierarchies naturally form. For Owen and Van As, the tool that leveled this field was CAD, and as we will see, natural justified hierarchies, hierarchies of competence, form when that happens.

II. The Rehearsal

The tool was Computer-Aided Design, riding on the desktop 3D printer and the cheap 3D scanner, and for about a decade it had a clubhouse. Dale Dougherty launched Make magazine in January 2005 and held the first Maker Faire in San Francisco in 2006. TechShop, the first major commercial American makerspace, opened in Menlo Park that October. MIT’s FabLab specification grew from forty-five sites in 2010 to roughly three thousand across some hundred and sixty-five countries, and by 2015 the faires had drawn over a million attendees. For a membership fee, anyone could walk into a shared building, use machines no individual could afford, and learn a language that allows machines to model physical shapes.

As I argued in my Domain Mastery essay, domain mastery is transferable between “near” domains, which run on the same or similar operative principles and can be accessed by similar mental frameworks or metaphors. But when two domains run on different operating principles, and the mental frameworks, the metaphors that a person needs to use to be successful are foreign and alien to each other, we call those domains “distant. It is these distant domains that create the Tower of Babel problem from the previous essay in this series: men sealed inside siloed hierarchies, who cannot compare and compete with each other in virtuous, non-lethal ways. Without a shared field, honest ranking fails. So domains are distant because they lack a shared set of abstractions. But what is a piece of software, what is a user interface, if not a mental framework that allows you to influence or affect the underlying domain using a set of agreed upon abstractions.

We might call this a bridge tool, a tool that does comparable work across separate professional domains through a shared interface, a shared output format, and a shared body of practice. The interface is itself a mental framework: an abstraction, a metaphor, that lets the user manipulate every domain the tool touches under one set of controls. Most pieces of software respected the way that humans already divided up the world. CAD did not. Through the software family that we are referring to as CAD, users can interact with family jewelry design, dental aligners, aerospace components, and architectural models. The interface is shared, the output is a common file format, and the practice travels with the user. A surveyor with ten years of CAD mastery can move into manufacturing work in weeks or months rather than years, because the primary skillset of both jobs is making CAD files on a computer and the surveyor is already REALLY good at that.

When you look at the maker-movement part of what made it so exciting was that people transferred, in numbers and in style. Jim McKelvey owned a glassblowing studio in St. Louis and lost the sale of a glass sculpture because he could not accept the buyer’s credit card. He walked into TechShop, prototyped a little white card reader, and co-founded Square; the Museum of Modern Art acquired the reader in 2011. Bre Pettis was teaching art in Seattle public schools before he co-founded MakerBot; he sold it four years later for $403 million. Edgar Sarmiento, a Colombian-born design student, won a crowdsourced competition and became the lead designer of a production autonomous shuttle without ever working for an automaker. A glassblower in payments, an art teacher in manufacturing, a design student in automotive, a puppeteer in medical devices. CAD Mastery turned out to be the only credential, the new industries needed.

The barriers to entry fell even between industries are still quite distinct. Nearly every custom hearing aid made on earth is now 3D printed: scan the ear, shape the scan in CAD, print the shell, with over ten million units in field use. Dentistry runs the identical pipeline on teeth. The two trades have adopted mechanically identical workflows, because the shared tool dictated the optimal workflow’s shape. In a modern lab, one Formlabs resin printer can run jewelry casting resin on Monday, dental surgical-guide resin on Tuesday, and engineering prototype resin on Wednesday. Only the cartridge changes.

Again what allowed the movement in these stories was mastery of the shared Bridge Tool, the interface named above. I describe the effect as similar to how a gaming controller (xbox or playstation as you prefer) can allow a skilled player to quickly achieve success in shooting games, fighting games, driving games, and flying games within hours, though the genres share almost nothing cognitively. A player’s mastery of the singular interface of the game controller, means that his skills transfer to other domains with incredible speed, far faster than if he had to master custom tools for each process no matter how much better the theoretical control that customized tool SHOULD be.

III. Temporary Egalitarianism, Durable Hierarchy

By most contemporary accounts, the American makerspace movement peaked around 2014. TechShop filed Chapter 7 in 2017. Maker Media went insolvent in 2019 after Autodesk, Intel, and Microsoft withdrew their sponsorships. The natural reading is that the CAD movement failed: the clubhouses closed, the magazines died, and the bridge tool must have been a fad.

There is a joke, or at least a story, that people tell in many emerging industries about the gold rush: the miners go broke, and the men selling picks and shovels get rich. The moral of the story is, don’t be the miner, be the pick and shovel salesman. There is a time and place in which this is true, namely in the height of a gold rush, when a bunch of otherwise unprepared people come and using money that they earned doing SOMETHING OTHER THAN GOLD MINING, buy a pick and shovel and decide to try their fortune. But by its very nature the fate of the unprepared masses is that 90%+ have no skills, are unable to capitalize on whatever gains they made, and lose their shirts in short order. When enough miners quit, the pickaxe sellers have nothing left, and have only been a parasitic free-rider on a short-term fad.

By contrast the stable model is organized mining: a hierarchy of labor, leadership, and deployed capital that can predictably extract gold over a longer period of time. Working with and supplying those meticulously organized companies is less exciting and creates less profit per transaction. In the pick and shovel context this is literally true, the mining company is going to send an experienced negotiator to buy their shovels and pickaxes in bulk, or set up a manufacturing center to build them at cost. But over the long-haul, it is relationships with these efficient high-quality producers (who are always intensely focused on identifying and respecting competence hierarchies) that produce real wealth.

One of the features of the short-term fad is a massive amount of egalitarian feeling. This is because the hierarchies are not known. Egalitarianism during the initial rush is solution to an information disparity, not a claim that all of the people you are treating the same will actually be wealth-creators on a longer time-scale. Any situation where the hierarchies are not known creates an information gap: nobody yet knows which walk-in will matter, which shop will produce, which master will wind up being ranked high enough to remember old slights. While that gap lasts, the rational short-term move is to treat everybody equally, or at least to treat everybody well, because it costs very little to be polite and a great deal to write off the wrong person too early. “Do not bet against any player” is a let’s-not-offend-anybody posture. It is prudent. It is rational. It is prudent and rational BECAUSE the players are ignorant, not just of each other’s rankings but of their own.

However, once the sequence of iterative games begins and people begin to accrue wins and losses, hierarchy forms and the egalitarian bubble pops.

CAD is not an exception to this process. The bridge between domains creates a new playing field, a vast unsorted space full of massive opportunity for wealth creation. This was followed by a decade of egalitarian expectation before a durable hierarchy emerged. MakerBot sold for four hundred and three million dollars on a thesis of monetizing mass access; that thesis failed as a commercial model, and Stratasys pivoted toward industrial customers. These were the pickaxe businesses. Their thesis was access was the main bottleneck and therefore cheap access for everyone would create broadly distributed wealth. The overwhelming majority of their customers never produced anything profitable. When the thesis of abundant access was tested it failed.

People read the bankruptcies listed abovce as the death of the CAD movement. That reading confuses the failure of the egalitarian viewpoint (which is inevitable) with the death of the CAD hierarchy. They are different events. What actually happened was a CAD aristocracy emerged: ranked skill, portable credentials, and a labor market that prices mastery across industries. The result of that change was that left-leaning outlets that had cheered the open clubhouses stopped covering the story once the outcomes stopped matching their narrative. In reality the hierarchy kept compounding.

The hierarchy shows up today on resumes rather than in crowds. CAD mastery is a certified, ranked, portable credential. Autodesk runs a tiered ladder from Certified User to Certified Professional, with separate tracks for construction and for manufacturing; SolidWorks runs one from Associate through Professional to Expert, each rank earned by examination. Employers in architecture, engineering, construction, manufacturing, aerospace, and medical-device development read the same certificates the same way. Engineers Rule reported in 2015 that, according to those who do the hiring at Disney, a SolidWorks certification will not necessarily guarantee a job, but it will land the interview. Rank earned inside the tool’s hierarchy is honored inside every industry hierarchy the tool touches. A friend of mine took a job with a surveyor precisely for the CAD experience that came with it; once he had years on the tool and a certification, he could move anywhere he called CAD-space. A man who holds the Professional rank can move anywhere in CAD-space in both senses of the word: anywhere in the modeled geometry, and anywhere in the labor market the tool touches.

Four vendors — Autodesk, Dassault, PTC, and Siemens — hold roughly two thirds of the CAD market. These are the dukes of CAD. Their moat is less the software alone than the certification apparatus that issues patents of nobility for peak users who hold the bottleneck in their brains and fingers. In August 2024 Stratasys sued Bambu Lab over patents. In January 2025 Bambu locked its firmware against the open-source tools that had built the field. Those who bet that forever-open mass access was the permanent order lost their shirts. Those who bet on rank and on protecting hierarchies of skill did not.

So to restated this as a prediction, the pattern is as follows: A Bridge Tool creates an information gap; egalitarianism rides that gap for a season; the pickaxe sellers prosper while the field is illegible; then rank consolidates, the access businesses die, and the hierarchy remains. CAD already finished the cycle across a score or so of industries. AI is about to do the same thing at a civilizational scale.

IV. The Second Tool Is Bigger, Much Bigger

In December 2025, a Sydney data analyst named Paul Conyngham drove ten hours to inject his dog with a cancer vaccine he had designed himself. Rosie, a Staffy cross he had rescued in 2019, had mast cell cancer; surgery, chemotherapy, and immunotherapy had each slowed it and none had stopped it, and the vets had given her months. Conyngham had seventeen years in machine learning and not one day of biology. He opened a chatbot and asked how to treat the disease, and the model walked him through it: sequence the healthy and the tumor cells, find the neoantigens, design an mRNA protocol to train the immune system against them. He paid about three thousand dollars to sequence Rosie’s tissue, ran the mutations through AlphaFold to map the mutated proteins, and used three different models to design the protocol, refine it, and check it. Then he carried it to Páll Thordarson at the University of New South Wales, whose RNA institute evaluated the protocol, cleared the ethics review, and manufactured the vaccine in under two months. By late January, Rosie’s largest tumor had shrunk by roughly three quarters, and she was clearing fences after rabbits again.

Now this is not to say that “AI cures cancer.” Caveats abound: This is one dog, with no control group, under a treatment no one has replicated; partial remission is not a cure, and mast cell tumors wax and wane on their own. Nor did an AI allow a data analyst to recreate a gene-lab in his basement: a state of the art lab at the University of South Wales actually made the treatment and took responsibility for putting it into a living animal. But even despite all those caveats, its a crazy impressive story. Compare it to the CAD puppet-maker and its a similar feat... albeit on a medical problem at least one order of magnitude more complex. When asked Conyngham put it in these terms: the chatbots, he said, empowered him to act with the power of a research institute. That is the collapsing siloes effect but this time the domains are sales consulting and oncology.

CAD used a file type (or perhaps a file family) to bridge the gaps between trades that model, cut, or deposit amounts in a defined 3-D space. The new tool’s medium is language and code, and language and code are what every serious profession already runs on. The lawyer’s brief, the physician’s chart, the farmer’s spray log, the engineer’s test suite: all of it is language and code. And the same workflow disciplines are being implemented by everyone who is trying to access the massive wealth the efficiencies of AI seem to create.

We are all listening to Andrej Karpathy. Orchestrate the model. Verify its output. Keep auditable state. Constrain it with tests. Open the trade press of any field this year and you find the same article wearing different nouns: the bar journal teaching lawyers to verify model output against the record, the medical informatics journal teaching clinicians to verify it against the chart, the agronomy newsletter teaching growers to verify it against automated sensors and farming machines. The shared tool, the shared interface, the common metaphor and domain masteries built or adapted to that mental model is dictating the workflows of every industry.

So ask the CAD question at the new scale. The certified drafter can move anywhere in CAD-space. What is the AI-space? Try to name an industry that is not being impacted, segmented, and reconnected by a tool that translates natural language into code-precise implementation. The humble terminal is the best proof. For fifty years the command line was the most powerful and least usable interface in computing, a priesthood’s instrument, accessible only to the most precise, most autistic members of our body politic. Now, with AI wrapper interpreting the natural language, it now takes instructions in plain English and outputs working commands with the machine grade precision the terminal requires.

The way that the release of ChatGPT affected Super Mario developers is instructive. All the way up to GPT-2, Researchers who wanted Super Mario levels out of the pre-wrapper generation of models (MarioGPT) had to be hyper precise in their prompting, grafting on new attention weights and fine-tuning through fifty thousand training steps, in order to have a chance of getting playable output. After the GPT-3, that same request is a sentence. Natural language in, code-precise implementation out, in any domain that can describe its own work: that is the AI-space, and we are all struggling to name an industry out side of this.

V. What will AI hierarchy look like?

So the process of a hierarchy emerging in this new massive playing field, as the disparate silos are deconstructed comes in three phases. First, the floor will drop: a newcomer in any bridged domain can collaborate with practitioners of any other through the shared interface. Second, the ceiling rises: masters become comparable across domains, because the shared field makes shared ranking possible. Third, a hierarchy emerges: where comparison is possible, ranking follows, and the ranked masters consolidate above the local hierarchies.

The floor has already dropped; that is essentially what vibe-coding is. The ceiling is rising now. Marc Andreessen has been watching early leading-edge companies in the Valley circle a job title loosely called “builder.” Programmer, product manager, and designer used to be three separate jobs. Each specialty now expects AI to cover the other two: a “Mexican standoff” in which Andreessen predicts all three sides are correct. Each can generate the work of all three, so the job becomes builder. One person becomes responsible for shipping complete products, with AI filling the gaps outside his original training. Masters in that shared field are becoming comparable across the old specialty walls.

A similar thing is happening in law, tax, and accounting, where a practitioner more familiar with the new tools and the underlying principles can achieve parallel performance, or even outperform a seasoned professional in their own specialty.

My prediction is that soon the best harness-builder in a law firm and the best harness-builder on a farm will be legible to each other, because both are visibly operating the same instrument with different cartridges loaded. You can watch the local version in any serious business environment this year. Some person has quietly become the one others bring their stuck problems to, the one whose harness everyone else’s work flows through, and that person’s is performing a disproportionate amount of work regardless of their formal domain expertise. Multiply that room by every trade that runs on language, and you have the early rungs assembling.

Once these types of domain cross-overs start happening it is only a matter of time until a hierarchy emerges. That stable hierarchy of tool-masters will consolidate across every bridged profession the way it consolidated over the design trades, because the same mechanism is running here with more force and more reach. As with everything involving AI it appears we are speed-running this process. The frontier models are already, barely 3 years in, straining under the pressures the makerspace businesses only faced after a decade.

The pickaxe sellers are already feeling it. In the spring of 2026 Microsoft cancelled most of its Claude Code licenses inside the Experiences and Devices division, the group that builds Windows, Microsoft 365, Outlook, Teams, and Surface, and told engineers to migrate to GitHub Copilot CLI by the end of June after the token bill ate the annual AI budget. Uber’s CTO told The Information the company had burned its entire planned 2026 AI-tools budget in four months on Claude Code and Cursor, with heavy users spending hundreds to thousands of dollars a month in tokens; the COO publicly admitted the firm still could not draw a clean line from that spend to shipped consumer features. They were not outliers. Bain’s 2026 survey of nearly a thousand large companies found that AI cost savings routinely landed below the bands executives had budgeted for, and Gartner placed generative AI in the trough of disillusionment, forecasting that a quarter of planned 2026 AI budgets would slip into 2027 as proofs of concept died in procurement. The tools worked well enough that engineers used them constantly. What failed to materialize was the durable work product the budgets had been sold on. The pickaxe sellers appear to have neglected to notice that you have to be digging where the gold ACTUALLY IS in order for your tool to be useful.

And yet there are what are being called 10x tokens. Meta’s CTO told a room that his best engineer was spending a salary’s worth of tokens and running five to ten times more productive for it: easy money, no limit. Jensen Huang said he would be alarmed if a half-million-dollar engineer burned only a few thousand dollars in tokens in a year. The industry has begun to treat heavy token spend as the mark of the new peak performer.

That framing is convenient, because it lets people talk about tokens and budgets while ignoring the politically incorrect conclusions underneath. Most major corporate org charts are the product of deceptive and non-meritorious sorting processes. The most immediately available piece of evidence for is that Elon was able to cut 80% of the bloat at Twitter and performance improved. The hierarchy of credentials that staffed the last forty years of the profession is deeply corrupt. A tiny fraction of engineers whom the powers that be have been suppressing for those decades are about to gain actual political power and relevance, because a bridge tool makes their competence visible in a way the credential ladder never would.

Many people are framing this as a “10x” tokens question. That framing strikes me as hilariously wrong. Behind the “10x tokens” as far as I can tell ,are just the peak engineers. The same tool that burned Uber’s budget without a clear line to shipped features returns an order of magnitude for the operator who already knows what he wants the work to become. Eric Raymond, responding to the public complaints that language models spit out worthless code, asked whether the naysayers were on weak models or had some kind of skill issue he cannot see because for him, his mental habits and communication skills naturally allow him to direct AI to massive effect. He describes getting 40 years of work, a lifetime of output, done in mere months. What he is pointing at is the scarce capacity he elsewhere assigns to the human side of the loop: architectural thinking, the higher-level pattern judgment the model is blind to when it generates to specification. Billions are being wasted on misdirected token spend, he noted. Be clear in your thinking, tell the model what you want with precision, and good things happen.

This also explains Jensen Huang’s situation at Nvidia. He has hired for the best and has not allowed partisan and ethnic bias to corrupt the sorting process, so AI supercharges an organization that was already sorted for competence. Nvidia is also incredibly small by comparison with the corporate giants that spent the same decades building the corrupt credential hierarchy into their hiring. Those giants are, in all probability, unable to change, and they will die.

In that sense AI is an alkahest. In Renaissance alchemy the alkahest was the theorized universal solvent: the liquid that could dissolve any composite substance down to its primary components without destroying what was essential in them. It was believed that upon dissolving all that could be dissolved that what would be left would be the philosopher’s stone, the key to changing lead into gold and creating elixirs of immortality. AI is an alkahest of our corporate credentialing system. It sifts out the philosophers stone from base elements. It dissolves the sealed silos and the credential smoke-screens that hid people from one another, and in dissolving them it searches out the top-level thinkers who can do the architectural thinking Raymond describes. Perhaps we might say instead of finding the philosopher’s stone it finds the philosophers.

But this is just the same kind of uneven distribution that we have seen before, and that if you will permit me that we ALWAYS see, in analogous situations. The promises of egalitarianism always prove false, and there was always a tiny minority that capture the vast majority of the wealth BECAUSE THEY WERE THE PRODUCTIVE AND TALENTED ONES FROM THE BEGINNING.

VI. Genuine Equality of Opportunity Always Creates a Hierarchy of Competence

Remember that the assumption that genuine opportunity creates equality of outcomes is a fundamentally left-wing premise and is not supported by any evidence whatsoever. A shared interface opens the field to anyone who walks in. Once the field is open, skill becomes visible across domains that used to hide inside credentialed siloes. When skill is Legible it can be compared. When skill is compared humans recognize ranks. Recognized Ranks consolidate the field into a hierarchy of competence. The less that the opportunity is gamed by leftist assumptions about equality of outcome, the cleaner the ranking, because nobody can hide behind crutches and handicaps meant to ensure that no one loses too badly.

Again the thing I want to emphasize is that CAD ran the pattern on a limited set of industries. Anyone with a membership could use the laser cutter; the winners of the CAD race are intimately tied to a certification system that is brutal in its ranking. AI is running the same pattern at civilizational scale. Vibe-coding drops the floor. The builder title raises the ceiling. The 10x-token debate is not a story about magic tokens. It is a story that keeps trying not to notice the competence hierarchy the tool is already making legible.

The equality of opportunity that AI is giving us is going to produce what genuine opportunity always produces... which is to say hierarchy, deep, inescapable, politically incorrect hierarchy. Without a shared field there is no honest ranking, only sealed silos and ultimately a tower of Babel story. But with the alkahest of AI, the siloes are dissolving, the players meet, and competence sorts itself the way it always does once the games start to iterate.

One final point. When there is a hierarchy, there will be an apex. Every ladder has a top rung. The man who sits there is the one who masters the master tool entirely: who can make it write, make it speak, and make it move things in the world. Armies and hospitals alike will bid for him, and no team of ordinary size will stand against him. The next essay is about that man, and about what sits at the top of the AI hierarchy. I call him the AI Paladin.