AI continues to bulldoze the economy at a dizzying pace

It is August 2026 and Earth continues to spin through space at roughly the same speed as it did just 10 years ago. Yet it was only just a blink of an eye ago when “Clawd Bot” (Open Claw), “Claude Code” and “Claude Cowork” silently changed human technology forever.

What everyone knows is that AI continues to rapidly shape the world. What few people stop to ask is: which technology is changing the world faster? the fastest? or nuclear-world-ending-ly-fast!

We all know that Anthropic, Open AI, Nvidia and the so-called “Megascalers” will no doubt make money, lots of money. Even considering the case that AI fails to meaningfully replace white-collar work at a profitable margin, AI revenues will not be at risk, but rather the status quo of the world economy will adapt accordingly. The cat is out of the bag; Pandora’s box has been opened. AI is simply too big to fail at this point.

AI harnesses fixed something in LLMs that make them real engineering tools

Leading up to the first flight of the Wright brothers in 1903, many people died attempting to achieve flight by jumping off cliffs on wildly imaginative prototypes. It was commonly assumed at the time that the obvious predecessor to powered flight was unpowered flight, so much research was conducted on improving gliders.

The little-known secret at the time that the Wright brothers deeply believed in was that the predecessor to the airplane was in fact a kite, not a paper airplane. They flew several prototypes as flight: pitch, yaw, roll, and throttle. People laughed at them and compared them to children playing with kites. Little did they know, the brothers had invented a critical piece of engineering that we can’t imagine working without in today’s time: wind tunnel testing.

So it was in knowing and believing in this little secret that allowed them to achieve controlled flight without any casualties and with just a few cheap bike parts. They were exceptional engineers, knew what they were doing, and firmly believed in the lunacy of the masses.

The impact of AI harnesses on the future economy is at risk of becoming an endangered species that might not survive tomorrow’s harsh environment without active conservation. It is not how many trillions of parameters, how many tokens per second, how much money a token costs, how many workers (“FTEs”) the agent system can replace, which reinforcement learning framework, which dataset, which model architecture, etc. None of these things matter in real-world engineering. The single concept that turns LLMs from children’s toys into real tools is that of Verify-ability.

Claude Code is the only mass-market verify-able harness

When an LLM outputs tokens it has no idea whether it will be useful to the user that prompted it. That is, unless it can be verified to some standard of truth. Thus, to bridge the world of bits to the world of atoms, the LLM’s output needs to be 100% and reliably true.

Previous attempts to achieve this have involved adding the reasoning “think” tags, adding reinforcement learning to train LLMs to call tools like web search to include up-to-date content. However it was the verification loop that checks the LLM’s expected output against a real result obtained from the physical world that finally achieved the 100% all the time breakthrough.

Let’s say a user wants an LLM to write code to print “Hello World”. Before the invention of the harness, we were only 99% sure the program output would actually be “Hello World”. But imagine if after the LLM generates the script, the program knows to run the code and then check whether the resulting output is exactly “Hello World”. That is exactly what a harness does, and it even sounds kind of stupid when you consider this basic application case.

Claude Code does in fact work on this basic, almost stupid principle, and it has transformed the entire field of software engineering with this one simple trick. But just imagine what other verification steps can be performed by physical world-facing sensors like cameras, motors, lidar, etc. Imagination is truly the only limitation!

Harnesses will change the physical world in ways nobody is ready to talk about

Perhaps the people who know the importance of harnesses are not necessarily going around telling everyone they know and don’t know about it. This is of course not a conspiracy theory or anything, but why go out of your way to educate everybody in the world about a once-in-a-lifetime wealth creation opportunity?

The bakery of the future will require no humans and be 100% operated by a harness

What fundamentally is a bakery beyond turning flour, water, yeast, and salt into some number of delicious ready-to-eat bread? Of course the salesperson in the bakery will always exist, but is there any reason for the bakery to need any humans?

What is bread dough mixing except a process by which mixing speed is controlled such that the raw ingredients can be measured out using a digital scale, the dough temperature is measured by a digital thermometer and the stickiness of the dough can be determined by a camera? And which of these tasks is not verify-able by an agentic harness?

What is bread rising except a process by which the dough volume is measured by some camera or lidar sensor, the fermentation temperature and time measured by digital sensors? And which of these tasks is not verify-able by an agentic harness?

The pharmaceutical manufacturing factory of the future can be 100% operated by a harness

What about making high value-added goods like pharmaceuticals, supplements, or cosmetics? Are these processes not simply more complex versions of a bread factory? Which analytical tests of product efficacy or safety cannot be performed by a harness to generate a verification for another harness?

Let’s not limit out imagination here, I think it is clear that nearly all industrial processes are suitable to be replaced by harnesses. And if any humans are needed at all, that’s where tele-operated humanoid robots can come in to deliver the final touch of automation.

The economy of the future will commoditize everything – that is everything except for business models that take agentic harness outputs as their inputs

This essay is not meant to argue that agentic harnesses are a great business opportunity right now, they most certainly are not. Harnesses may be limited to coding only in August 2026, but soon enough, and probably sooner than you and I might think, the diffusion of this technology will lead to mass commoditization. At first only the manufactured goods will become commoditized, but soon enough the commoditization will begin to cannibalize itself and commoditize agentic harnesses itself. Thus, I predict there will only be one or two mega-companies that will control almost all of the harness business.

I argue that the best business opportunities to pursue right now are in the business models that do not yet exist today, but will be enabled by mass commoditization. This requires one to believe that rapid change by AI is definitely coming and that a lot of things in this world are about to get much cheaper, faster, better, stronger. I know this may be a difficult thing to believe in, but I hope that there is sufficient evidence around you to convince you to be this rationally optimistic.

Commodities as inputs paired with a compelling need is the winning formula of the post-harness economy

What can be certainly known about the winning business model of the future is that it will still follow the fundamental golden formula: buy low, sell high. What is being bought at low prices will certainly be the goods and services that will become commoditized by the agentic harness economy. What people will need in the future and are willing to pay premium prices for remain open to speculation, and of course to disciplined market research.

Emotional satisfaction and the offline third space as a fundamental unmet market need?

Following the trend of online-to-offline (Disney, Pop Mart, experience economy, etc.), the gamification of offline third spaces is projected to grow to meet the rising unmet emotional needs of the perpetually online.

Health and wellness treatments that are currently cost-prohibitive will become commercially viable?

Expensive elective tests like longevity biomarker testing, RNA, DNA sequencing will become more accessible in the post-harness economy and will meet the growing unmet need of longevity management.

Did you find this essay helpful?

If you did, feel free to reach out to Jerry to schedule a casual coffee chat. If you have read this far, Jerry is very eager to meet with you and listen to your feedback. Jerry is currently based in the Zhejiang-Jiangsu-Shanghai area and may also regularly travel to Beijing, Tianjin and Shenzhen.