As the dust only just begins to settle from the collapse of Accell Group—yet another shock to the industry—this month we’re going to embark on a thought experiment of sorts.
We combine an interesting new perspective on the possible shape and function of the future economy on the one hand with the bike industry on the other.
In a rapidly moving technoscape, no clear conclusions can be drawn. Staying vigilant, monitoring the changes and reflecting on where they might lead has always been a cornerstone of business prosperity
In this uncertain future it’s an absolute necessity.
Contents
The Agentic Economy
Let’s begin with a definition of agentic that will inform the following disscusion. Agents are AI systems that autonomously carry out tasks just as a human might, except at a speed and scale which leaves the human way behind.
There’s more to it than that, a LOT more in the form of loop, harness, and graph engineering. That’s beyond our scope for now; we’re still figuring everything out. Let’s go with this minimal definition in the interim.
Two Trends are Actually One
Jeremy Allaire, the CEO of Circle, a finance company focusing on crypto stablecoins, published a long treatise earlier this year, The Agentic Economy.
The central idea is that AI and blockchain are more than two tech trends. Not only are they are absolutely key to the future, they are intricately interconnected. One implies the other. Each needs the other.
Before you click away, this bears directly on the structure of the economy which is being restructured around AI. And central to that restructuring is the transfer of value on blockchains, particularly Etherium.
True, these are very early days yet but as the saying goes, “The future is already here. It’s just not evenly distributed.”
The disruption of existing systems and the assertion of these new forms remains extremely uneven. But there’s no denying their relentless push from the shadows into the light.
Whereas AI pushes the marginal cost of thinking towards zero, blockchains radically reduce the cost of paying bills, invoices and the like in which transaction and settlement are almost the same thing.
At the least, settlement takes place light years faster than is currently the case. Having transaction and settlement occur almost on top of each other is significant.
The End of the Firm
We had a close look at this idea ourselves a few months ago.
Allaire makes a similar case although with different emphases, and implies the reorganization of existing structures rather than radical, total restructuring.
The argument is that companies exist today because it’s cheaper to coordinate labor inside four walls than to hire it out job by job. Once AI agents can be dispatched and paid instantly, however, the calculations change; everything changes!
Work gets broken into tasks and routed to whoever—or whatever 🤖—does it best. A company stops being a fixed structure and starts being an orchestration layer dispatching agents.
Trust comes to be written down and if you’re going to hand real tasks to software agents, you need to know who they are and hold them accountable (hence the rise of Graph and Harness Engineering and as essential areas of expertise).
Allaire calls this the identity and trust stack or in other words cryptographic identity combined with real-world accountability and a reputation record. Autonomy without anonymity, in his terms.
And when money moves at machine speed—and that’s pretty close to the speed of light—agents can’t wait around for a transaction to “probably” clear.
They need settlement that’s actually final, instantly. Allaire points out that such an innovation replaces the old trick of bank leverage (lending the same dollar many times) with velocity (turning the same dollar over so fast it does the same economic work, without the same risk).
Credit is restructured in other words.
Machine underwriting makes it cheap to evaluate small borrowers banks didn’t consider their targets or simply worth the risk.
Agents, and the small businesses behind them, can now borrow working capital against contracted income instead of physical collateral.
What happens to software in this? Well, it is repriced so that the “seat” of the SaaS model based on paying per human user no longer makes sense since the software itself is doing the work.
Pricing shifts to metering actual output which reduces costs down to fractions of a cent per task, a circumstance made possible because crypto tokens are far more finely divisible in comparison to the financial instruments currently in use.
In this reimagining of the traditional firm in which control is everything—in fact, the modern firm is defined by the key function of channelling power through layers of authority and hierarchy—are we looking at total decentralization or the opposite, the tyranny of agentic AI?
Neither. Power still concentrates, only the chokepoints are different.
Most of the stack, the apps, and the AI models themselves end up commoditized, and pretty quickly. (See the current struggle between closed and open source models—no prizes to guess which will triumph there).
However, two layers escape what on the surface appears to be absolute autonomy on the part of agents.
The first is identity, because a verified identity gets reused everywhere and there’s no forking it. The second is override keys which confer on the holder the power to halt or reverse the system.
Whoever controls those two things controls the economy running on top, no matter how decentralized everything else looks.
That’s the frame. Here’s where it gets interesting for people who make bikes.
How This Relates to the Bike Industry
Well, it may or may not. But on balance, it probably will since the structure of business organization generally is going to be transformed—the agentic economy is coming for all of us in one way or the other.
So, a brand that makes frames, for example, doesn’t weld its own tubes, right? Usually anyway. It buys tubing from a specialist, perhaps Ming Huei or Nan Long. Then a technician on site may briefly spot weld the tubes into place using a jig to more or less position down tube, head tube, seat tube and so forth.
Frames then go out in a batch to a welding specialist. Take a drive around the Daan/Dajia/Yuanli district on any given day and you are sure to spot one of the ubiquitous little blue trucks filled to the brim with racks of frames on the way, or on the way back from welding.
Next welded frames go out for T4 baking and T6 heat treatment: T4 hardens the welds; T4 hardens the frame overall.
Then the frames are sent off to a paint er specializing in decals and paint application including baking and quality control. Getting those little specks of dust and dirt out of the final coat before the clear coat, and later the clear coat itself, is a time-consuming and tedious job). All back and forth in those little blue trucks.
Here’s the Point
Nowhere in this chain is one firm doing everything across many departments. Even big operations like Giant and Pacific Cycles and so forth outsource just about everything (depending on the product level). All players are each a separate business, coordinated by phone calls, LINE groups, and relationships built over decades.
Allaire’s decomposition of the firm isn’t a prediction for this industry. It’s actually description of what’s happening every day, and you could make the argument that Taiwan’s bike supply chain has been running as loosely coupled, specialized nodes since before anyone called anything agentic anywhere.
In a nutshell, the important question (existential in the medium to long term) isn’t whether decomposition happens. It’s what changes when the coordination layer currently running on relationship and memory gets replaced by software that formalizes it.
Formalized Orchestration
Right now, matching a brand’s spec to available capacity across dozens of subcontractors runs on institutional memory.
Someone at the brand knows which welding shop is free this month, which anodizer is backed up, which paint line just lost a client and needs volume. That knowledge lives in people’s heads and phone contacts.
In an age of agents, an orchestration layer does that matching automatically.
Feed it a spec, it finds capacity, checks reliability, routes the job. Something like Alfred Tsai’s Bicycle Cluster data platform is already a step toward this: a shared, structured view of who makes what, at what capacity, at what standard.
That sounds efficient, but it also means somebody owns the layer that decides who gets found.
The Chokepoint? (It’s important!…)
Here’s the part of Allaire’s argument most important for a Taichung workshop with three to thirty CNC machines and no marketing budget: power in a decentralized system moves to whatever can’t be copied. And that is 1. Identity; 2. the override switch; and 3. the layer that decides who’s visible and who isn’t.
If an orchestration and trust layer gets built for Taiwan’s bike industry, somebody builds it.
Who might that be? A big OEM? There are plenty of those around who if they take the initiative (to exit what we’ve called elsewhere the Pit of Uncertainty), could position themselves in an unassailable position as far as the future goes.
Other candidates include a platform company, or possibly a government-backed consortium (T22 Revitalization project has shades of this). Whoever it is becomes the new gatekeeper, except now the gate is code instead of a buyer’s Rolodex.
The Gate becomes Code and no longer a buyer’s Rolodex.
That’s not obviously worse than the current system. It might be better. But it’s not neutral, and it’s not obviously good for the small shop either.
Whatever the case such a structure is likely to be enforced should Allaire’s vision of a world-wide agentic economy come to pass. (And if it does, the arrival of this alien way of structuring business will be sooner than later).
Being Seen
The current system runs on tacit knowledge. A workshop’s reliability is known because a buyer has toured the floor, possible multiple times, watched the welds, had dinner with the owner, and done this for years.
That knowledge built up slowly over time and hard to replicate. It’s also invisible to anyone outside the relationships constituting the structure.
An identity and trust stack writes that down. On-time delivery rate. Defect rate. Capacity utilization. Payment history. All of it, explicit, queryable, comparable across every workshop in the system at once.
Explicit is exploitable in ways that tacit is not.
A workshop that used to compete on relationship now competes on a score.
A buyer who used to need years of face time can compare fifty shops in an afternoon. That’s good for buyers, but a mixed bag—At. The. Very. Least— for a shop whose entire value used to be the relationship itself, not just the metrics behind it.
As we write, data governance (standardization of data forms and structure) which is at the heart of such a system is currently being piloted in Taiwan.
So, Then, Who Becomes the Orchestrator?
To hazard a guess, running this layer won’t fall to the small shops which constitute Taiwan’s bike component sector.
The controller will be whoever already has the scale and the buyer relationships to make an orchestration platform worth building. Giant? Merida? (Depends on how enlightened management is about the shape of the future . . . which is possibly the key variable to making this a crucial competitive edge.)
Could it be a platform a nobody’s built yet but with the same instinct as Alfred Tsai’s data work, just with settlement and payments bolted on? You’d have to think that something he’s got his eye on.
If that’s right, the small subcontractor doesn’t get freed from its dependency on a handful of big OEMs and buyers.
It gets a more efficient, better-documented version of the same dependency. Its pricing and reliability become visible to everyone at once, which sounds like opportunity until you remember visibility cuts both ways. (And here a key differentiator will be authentic marketing that steers well clear of AI).
It’s also easier to route around a workshop once its capacity and quality are legible to a system than when they lived in one buyer’s head.
Counterpoint
Let’s step back a bit and consider the following.
Allaire’s argument on machine underwriting and working capital is specific.
Agents borrow against contracted cash-flow streams instead of needing collateral or a bank relationship.
For a small electroplating shop or decal house currently eating a 60-day float on every job—it’s potentially HUGE.
Empowering.
Faster settlement and capital access without a physical asset to pledge against?! “Game changer” is a much abused term these days. But if anything qualifies, it’s this.
Still, faster cash flow doesn’t fix a worse bargaining position.
A shop that gets paid faster while becoming easier to replace hasn’t come out ahead. It’s just been made more comfortable on the way down.
What, Then, Are We Left With?
The open question is whether any of this substitutes for what currently makes the system work: the banquet, the factory tour, twenty years of a buyer trusting a workshop because he’s watched it operate through good years and bad ones.
An orchestration layer can approximate that trust with data. It can’t replicate the part where a buyer sticks with a shop through a bad quarter because of the relationship, not the score.
Maybe it’s not an issue.
Maybe the industry moves toward the version where trust is a number and everyone’s fine with that.
No idea. But we continue to speculate.
The early days of the world wide web promised democratized point-to-point access to information and communications.
This vision was true in the early days, but by the mid 2000s had faded away. Now big tech—so called FAANG or now MAGNA MOBSTA…whatever—and Internet Provider nodes dominate the interwebs.
Companies savvy in internet marketing and SEO could become influential and powerful back then. Yet ultimately only at the behest of Google and the other platforms.
Now, agents running on a company’s local server (the open source models already provide this) promise significant decreases in cost and significant increases in productivity (failures are due to lack of expertise in loop/harness/graph engineering…you need staff that understand this stuff).
The promise of an escape from delayed payments in the form of instant settlement and instant finance based on accumulated digital capital (the argument for a Strategic Bitcoin Reserve is relevant here) is potentially revolutionary.
Once again however, the platform effect, shall we say, threatens capture of a system before it gets going.
Times are moving fast; we’ll keep watching.
EN 
DE