The $15 Trillion Handoff

Executive Summary

B2B commerce, a $28 trillion market and the largest economic surface on earth, is being re-plumbed from human intermediation to agent intermediation. Big tech is laying the rails. We are building the network. That is the position worth owning.

The B2B market is broken, measurably. A complex B2B purchase involves 6 to 13 stakeholders, takes the better part of a year to close, and stalls 86% of the time. Cold email replies are below 1%. Calls are blocked before they ring. Procurement teams at the backbone companies of European industry still normalize RFQs by hand, in Excel, from PDFs and photographs. The largest market in the world has no order book, no matching engine, no price discovery, no liquidity.

The constraint that caused it just disappeared. Matching a buyer to the right supplier used to require human judgment, which is slow and expensive, which is why somebody always had to be paid to do it and therefore always had an interest in the answer. That stopped being true in the last twenty-four months.

The infrastructure is being built in public. MCP went from 100,000 to 97 million monthly SDK downloads in eighteen months. A2A reached 150+ organizations and every major cloud within a year. The payment layer followed faster: x402, AP2, Visa, Mastercard, and 165 million agent transactions already processed. The trust layer is under construction, and the American Arbitration Association is building the court that adjudicates between machines.

Gartner has dated it: by 2028, 90% of B2B buying will be AI-agent intermediated, pushing over $15 trillion through agent exchanges. Not new money. The existing economy changing rails.

Every one of those efforts is a rail. Not one is a network. Google built the language, not the place agents meet. Visa authorizes transactions and has never built the marketplace that originates them. TCP/IP did not become Google. And the obvious contenders are structurally excluded: incumbents buy networks rather than build them, which is why SAP bought Ariba. Marketplaces with their own catalogs cannot be neutral, because neutrality is what their model exists to prevent.

FlowMarket occupies that gap. A neutral, cross-directory network with no catalog to defend, interconnecting the world's B2B directories over agent protocols and matching on relevance, trust, and speed rather than marketing budget. Thousands of companies have joined, with tens of thousands more in the pipeline, most of them SMEs brought into the agentic economy one conversation at a time.

The Reason

I've spent fifteen years in B2B, eight of them running a software development agency. Thousands of hours around developers, watching work that was skilled and valuable and, I kept thinking, not going to stay handmade forever. Sooner or later something would industrialize it. That was years before ChatGPT, Claude, Cursor, before anyone used the word "AI" in a business meeting without a smirk. I thought developers would end up going the same way every skilled trade goes once the tools catch up: from craft to production line.

That prediction is now a reality. Development as manual craft is mostly over. Anyone looking for a developer job right now understands that better than any analyst report could explain it.

Then I ran a B2B software company, and the same thought hit me again, this time about sales and marketing. Decades of new tools, better databases, sharper analytics, and the actual process hadn't changed at all. It's still the same two options it always was: you interrupt people who never asked to hear from you, or you wait to be found by people who somehow stumble across you. Everything in between is manual. And it's crowded: layers of intermediaries and gatekeepers, each taking a cut, each cut quietly raising the price for whoever sits at the end of the chain.

What finally made it click was smaller than either of those. It was ordinary frustration with LinkedIn, and what sat underneath it: one aging platform, with an interface nobody would defend, effectively decides who finds whom across the largest economy on earth. Not because it's the best answer to the problem. Because it got there first, and nothing came after it.

We have god-like technology and processes from the Stone Age.

So I set out to build the opposite. A network where two companies match on what actually matters, meaning intent, relevance, trust, and speed, and not on whose marketing budget is bigger, who ranks higher in search, or who happens to own the address book. Where a small manufacturer with no sales team gets the same reach as a competitor spending millions, and a buyer finds the right supplier in seconds instead of weeks. Where the finding, the qualifying, and the first round of the conversation belong to agents, so the humans on both sides can do the part only humans are good at: deciding.

That was the intention from day one. What I hadn't understood yet was the size of the wave I'd just stepped into.

Because while I was getting here from the inside, one bad platform, one broken process, one frustration at a time, a whole industry was arriving at the same place from a completely different direction. Protocols were being written for machines to deal with machines. Forecasts were being published that almost nobody was taking literally yet. Infrastructure was being laid, quietly, for a kind of commerce that didn't exist five years ago.

Two paths, heading for the same point without knowing about each other.

They meet here. And the numbers at that intersection are bigger than I expected.

The Curve

Not long after ChatGPT launched, I read a throwaway line somewhere: it was only a matter of time before the first AI millionaires were minted, ordinary people who simply used the tools early and well. At the time it read like optimism. I remember thinking, how would that even work?

Then the curve answered. AI hasn't been minting millionaires. It's been minting billionaires, at a speed no technology has managed before, and we're now close enough to the first trillion-dollar AI company that the question is which one, not whether. That's the shape of what we're talking about. Not growth. A different order of magnitude, arriving faster than the people inside it can describe it.

The same curve is now bending into B2B, and most people are watching the wrong part of it. What's visible today is the tooling layer: AI SDRs, AI research assistants, AI copilots bolted onto the CRM. Real products, real revenue, genuinely useful. But they're evolutionary. They make an existing manual process faster while leaving the process itself intact. The tool changes; the shape of B2B doesn't.

The actual shift is one layer down, and it's much bigger than companies using agents to automate tasks. It's companies being represented by agents, and those agents doing business directly with each other.

That's not a forecast. It's a schedule, and the infrastructure is already being poured.

The plumbing is done before most people noticed it was being built.

In November 2024, Anthropic released the Model Context Protocol, a shared standard for AI systems to connect to tools and data instead of every company building one-off integrations. It was a plumbing announcement, the kind normally read only by engineers. Eighteen months later, the SDK is being downloaded 97 million times a month, up from roughly 100,000 in its first month, a 970x increase. OpenAI, Google, Microsoft, and Salesforce all shipped support within thirteen months. Direct competitors converging on one rail, that fast, almost never happens.Knak

The ecosystem around it grew just as quickly. By May 2026 the official registry counted close to 10,000 published servers, GitHub carried nearly 16,000 repositories tagged as MCP servers, and 41% of surveyed software organizations reported running MCP in limited or broad production.Digital Applied

Then came the second half of the plumbing. Google published the Agent2Agent protocol in April 2025 and handed it to the Linux Foundation two months later. By its first anniversary it had 150+ supporting organizations, up from just over 50, with SDKs in five languages and native integration into Azure AI Foundry, Amazon Bedrock AgentCore, and Google Cloud's agent toolkit. IBM folded its own competing protocol into A2A rather than fight it, and as of August 2026 both A2A and MCP sit under the same neutral foundation.Devtoollab Linux Foundation Forbes

Read those two paragraphs together, because that's the whole point. MCP gave agents a way to reach data and tools. A2A gave agents a way to find, address, and delegate to each other across company boundaries. Two years ago, an agent representing one company had no standard way to even discover an agent representing another. Now it does, and every major cloud provider on earth has agreed on how.

And the enterprises are already moving.

More than 80% of Fortune 500 companies are expected to have AI agents running in production workflows by the end of 2026. Gartner puts 40% of enterprise applications integrated with task-specific agents by the end of this year, up from under 5% in 2025, a category that essentially didn't exist eighteen months ago.Nevermined The Agentics

The forward view is where it gets hard to ignore. By 2028, Gartner expects 33% of enterprise software to have agentic AI built in, up from less than 1% in 2024. At least 15% of day-to-day work decisions will be made autonomously by agents, up from zero. And roughly a third of interactions with AI services will involve autonomous agents completing tasks rather than answering questions. By the same year, Gartner expects a third of all user experiences to move off native applications entirely and onto agentic front ends, meaning the software interface itself stops being where business gets done.Gartner

And then the number that stopped me cold:

By 2028, 90% of B2B buying will be AI agent intermediated, pushing over $15 trillion of B2B spend through AI agent exchanges.Gartner Digital Commerce 360

Fifteen trillion dollars. That's not a market opportunity, it's larger than the combined GDP of Japan, Germany, and the UK. And it isn't new money being created. It's a fraction of the existing global B2B commerce economy, $24 trillion in 2025, $28 trillion this year, on a path to over $100 trillion by 2033, changing hands from humans to agents on a timeline nobody priced in two years ago.Grand View Research

The buyers were ready before the technology was.

The demand side has been drifting this way for years, quietly, without needing AI to push it. Two-thirds of B2B buyers now say they'd spend $50,000 or more online without ever speaking to a salesperson. Fifty-nine percent already make more than a quarter of their purchases through marketplaces. Three-quarters of B2B executives consider the digital model permanent.Uncap

That matters more than it looks. The hard part of any commerce transition isn't the technology, it's convincing people to buy differently. That work is already done. B2B buyers have spent a decade getting comfortable with self-serve, digital, no-human purchasing. Agents don't have to change their behavior. They just have to show up where that behavior already is.

One honest caveat, because it strengthens the case rather than weakening it.

Gartner also predicts that over 40% of agentic AI projects will be cancelled by the end of 2027.Gartner

Both things are true at once, and that is exactly what a real platform shift looks like from the inside. The mortality rate among the companies chasing this will be brutal. The direction of travel will not change. The question isn't whether $15 trillion of B2B spend moves onto agent rails, since the rails are already laid and the buyers are already willing. The question is who owns the place where those agents meet.

I didn't arrive at any of these numbers. I arrived at the frustration first. The numbers just told me the frustration was early, not personal.

The Telegraph Office of 2026

Email died first.

At LeadRebel (my previous company, exited in 2024), we started doing cold email at scale in 2021, and for a while it worked like magic. Five percent reply rates. Leads arrived faster than we could work them. Two years later, same messages, same lists, same discipline, reply rates under one percent. Our execution hadn't gotten worse. The channel had been consumed. Everyone found it, everyone piled in, and the arms race ended the way it always does: with the inbox winning.

Then the phone.

The next channel was already dying when we got there. Cold calling in the US is effectively over. Decades of abuse taught the carriers and the customers to filter, and unknown numbers now get flagged or silently dropped before a human hears them ring. It's no longer a question of whether your pitch is good, but whether your call is delivered at all. We called European companies instead, GDPR anxiety and all, purely because US connect rates had collapsed.

That's the state of outbound. You don't pick a channel because it works. You pick the one that's least broken this quarter, and watch it degrade.

Then there was no channel left.

Every path to a buyer is now saturated, filtered, or priced out. So the industry started looking for side doors, and that's how intent data became a category.

That's exactly what we sold at LeadRebel: identifying which companies visited our customers' websites, so sales could reach out while somebody was actually looking. It rode a real wave.

But it's half-baked intent, and I say that as someone who sold it. A company visiting your website doesn't mean it wants to buy from you. A company hiring doesn't mean it's purchasing tools. A startup raising a round doesn't mean it's in the market. These are correlations dressed up as intentions. The whole industry is reading tea leaves at scale, because the thing everybody actually wants, meaning this company needs this specific thing right now, was never observable.

And the buying side is worse.

I've hosted dozens of webinars with procurement specialists, mostly DACH, mostly the companies that form the backbone of the European industrial economy. Not laggards. Serious businesses with real revenue. Many have never opened ChatGPT, and very few use any modern sourcing tool. The state of the art is Excel, and that's the good case.

What lands on their desk is a mess by design: RFQs in whatever format the sender had. Excel files. PDFs. Photographs of a printed page. Specifications written out in prose. From that, a human is expected to work through dozens of suppliers and hundreds of products, normalize the comparison in their head, and pick a winner. That's archaeology, not procurement. And it's how a meaningful share of European industrial supply gets decided today.

The seller side's answer to all this friction has been volume. My LinkedIn inbox is full of offers targeted about as precisely as a weather forecast for a continent. I'm not innocent either: across two companies I've sent thousands of those myself, most of which went nowhere. Everyone in B2B sales knows the odds are terrible, and everyone does it anyway, because the alternative is doing nothing.

Paid isn't better, just more expensive. B2B advertising works like a casino where only the house is guaranteed to win. For some companies the math still works. For smaller ones the entry fee exceeds what they have. They don't lose the game, they never sit at the table.

Which leads to the part that made me want to build something. The result of all this is not that the best product wins, but that the best-funded distribution wins. I know because I lived it from the losing side. I sold my last company to our largest competitor because they could outspend us on marketing, not because they had built something better. Several customers who migrated told me plainly they preferred our product. It didn't matter.

Now zoom out, because none of this is anecdotal.

A typical complex B2B purchase involves 6 to 10 decision-makers, each arriving with four or five pieces of research nobody else has seen. Forrester puts it at 13 stakeholders, with 89% of decisions crossing departmental lines. Buyers spend only 17% of the journey talking to suppliers; the rest is spent researching alone, forming opinions no seller can see or influence.

The clock reflects it. The average B2B sales cycle runs about 10 months, and enterprise technology deals 11 to 17 months. One benchmark drawn from 66 million sessions puts the average journey at 272 days. Meanwhile 86% of purchases stall at some point, and 77% of buyers describe their last one as complex or difficult.

Read that as a market-efficiency report rather than a sales-productivity report. Nine months of latency. Double-digit participants. Information asymmetric on both sides. A majority of transactions stalling mid-flight. In any other market we'd call it broken.

Compare it to a market that works.

If I want to buy NVIDIA stock, I don't identify counterparties, research who might sell, email fourteen of them and wait, or find out later that a better price existed three desks away. I place an order, and it executes in milliseconds against the best available counterparty, someone whose identity I'll never know and never need to. Price is discovered continuously. Trust is handled by the structure rather than the relationship. Nobody's marketing budget decides whose shares get bought.

Now hold that against B2B: roughly $28 trillion a year, more than four times all consumer e-commerce, the largest economic surface on earth. It runs on cold email with a 1% reply rate, calls blocked before they ring, PDFs of scanned spreadsheets, inferred intent, ad auctions that price out anyone small, and nine months from first contact to signature.

The largest market in the world has no order book. No matching engine. No price discovery. No liquidity. It has intermediaries, directories, and a lot of people typing.

This is the telegraph office, still open in 2026.

There was a good reason for that. Matching a buyer to the right supplier required judgment, context, and negotiation, things only humans could do, and only slowly. Everything about B2B was built around that constraint.

That constraint is what just disappeared.

The Momentum

Infrastructure always gets built before anyone can explain what it is for. That is happening right now, in public, funded by the largest technology and financial companies on earth, and almost nobody outside the space is watching.

The communication layer is finished. MCP gave agents a standard way to reach tools and data. A2A gave them a standard way to find each other and delegate work across company boundaries. Google published A2A in April 2025 and handed it to the Linux Foundation two months later. IBM folded its own competing protocol into it rather than fight. Within a year the standard had over 150 supporting organizations, a v1.0 release, and native integration into Azure AI Foundry, Amazon Bedrock AgentCore, and Google Cloud. In August 2026, A2A and MCP were placed under the same neutral foundation. The industry has agreed on how agents talk.

The payment layer arrived a year later and moved faster. For a long time an agent could research, compare, and assemble an entire purchase, then stop dead at the one step that turns a task into a transaction. That wall came down in 2025. Coinbase revived a thirty-year-dormant HTTP status code and turned it into x402, letting machines pay each other over ordinary web requests. Google announced AP2 with more than sixty partners, including Mastercard, Visa, American Express, PayPal, Adyen, Coinbase, Salesforce, and ServiceNow. OpenAI and Stripe shipped their own checkout protocol the same month. Visa launched Intelligent Commerce and then a protocol for verifying that a request genuinely comes from a legitimate agent. Mastercard built agent-scoped payment tokens, then a version aimed at high-frequency machine transactions.

Then the rest of the internet plumbed it in. Stripe added x402 as a settlement path, Cloudflare built it into its agent SDK, and AWS added support at the edge. In April 2026 the Linux Foundation created a foundation to govern x402 as neutral public infrastructure, with Visa and Mastercard as premier members. By that month, x402 had processed more than 165 million agent transactions across roughly 69,000 active agents. A fair share of that is still testing rather than commerce and should be read that way, but 69,000 agents making machine payments is not a demo.

The trust layer is being built as we speak. Once agents transact routinely, identity stops being sufficient. Knowing who is at the door tells you nothing about how they behave once inside. So the industry is now building the layer above: persistent agent identity, cryptographic capability attestation, and portable reputation scored on settlement reliability, dispute frequency, audit quality, and tenure. An IETF draft exists for a domain-anchored agent naming and trust service. Ethereum has a proposed standard for on-chain agent identity and reputation, designed specifically so agents can be selected without any pre-existing relationship. The shared assumption is that reputation must become a machine-readable, portable asset rather than something a human infers from a website.

And, remarkably, so is the court. In June 2026 the American Arbitration Association, the largest private dispute resolution body in the world, launched the Legal Context Protocol together with Integra Ledger and a coalition including Google, IBM, Circle, Wayfair, and UiPath. Its purpose is to make legal terms, consent, and dispute resolution discoverable and verifiable when agents transact on behalf of companies. When two agents disagree about what was agreed, there has to be somewhere to take it. A hundred-year-old arbitration institution is now building machinery to adjudicate between machines.

That is a useful test of how serious this is. Speculative technologies do not get courts built for them.

Read the participant list and the conclusion writes itself. Google, Coinbase, Visa, Mastercard, Amex, PayPal, Stripe, AWS, Cloudflare, Microsoft, Salesforce, IBM, Adobe, and the AAA. Direct competitors across three industries, converging on shared standards inside neutral foundations, within roughly eighteen months. Companies do not donate protocols to the Linux Foundation as a hobby. They do it when they are certain a shift is coming and their main concern is that nobody else owns the rails. The AAA quoted the sharpest version of the forecast in its own launch announcement: by 2028, Gartner expects 90% of B2B purchases to be intermediated by AI agents, channeling more than $15 trillion through automated exchanges.

And here is the part that matters for this document.

Every one of these efforts is a rail. Not one of them is a network.

Google built a language for agents to speak, but not the place where they meet. Coinbase built a way for machines to pay, but not the thing they pay for. Visa and Mastercard built the means to authorize a transaction, which is what they have always done, and they have never once built the marketplace where the transaction originates. The AAA is building the court, not the market the court serves. The Linux Foundation exists specifically to keep this infrastructure neutral and unowned, which is exactly why it will never operate a commercial network on top of it.

This is the ordinary pattern of every platform shift. TCP/IP did not become Google. The card networks did not become Amazon. GSM did not become Apple. The layer that carries the traffic and the layer that captures the value are almost never built by the same people, because they require opposite instincts. One demands neutrality and standardization. The other demands aggression and market-making.

One more thing is worth noticing. Almost all of this activity points at consumer retail: agentic checkout, shopping assistants, someone telling an agent to find running shoes under $150. That is where the demos are and where the attention goes. Meanwhile B2B is more than four times larger, moves far more money per transaction, and has vastly more to gain, because its existing process is measurably broken in a way that consumer checkout is not. Nobody buying shoes waits 272 days.

So the rails are being laid at extraordinary speed, by people with unlimited resources, for a market that is not the largest one. What is missing is the network that runs on top of them, on the side where the real volume is.

That gap will not stay open for long.

The Land Grab

Why the incumbents won't build it

I spent time close to an agentic AI initiative inside one of Germany's largest agency groups, a billion-euro business with real talent, real budget, and a mandate from the top. Exactly the kind of organization that should own this. It didn't come together, and not because anyone was wrong about the opportunity. It came apart on decision paths built for a different era, planning cycles measured in quarters in a market that reprices monthly, and incentives that never lined up.

That's not a story about one company. It's the standard story.

Google didn't build the world's video platform, it bought YouTube after its own attempt went nowhere. Facebook bought Instagram with thirteen employees. Microsoft bought GitHub. Salesforce bought Slack for $27.7 billion. And in this exact category the precedent is direct: SAP, the company that essentially defines enterprise procurement, did not build its own supplier network. It bought Ariba for $4.3 billion, then Fieldglass, then Concur.

Incumbents are excellent at scaling a proven network and structurally poor at creating one. Creating a network means doing something unprofitable, unproven, and organizationally embarrassing for years running. Large companies are optimized specifically not to do that. So the honest answer to "why won't SAP just build this?" is that history says they won't. They'll buy whoever did.

Why the ones who could can't be neutral

The second group is marketplaces that already have agentic sourcing live. Alibaba's Accio is the clearest, a real AI sourcing agent with real usage. Amazon Business is moving the same way. They validate the demand. They can't occupy the position.

A marketplace's value comes from routing demand into its own supply. Accio is a buyer's agent shopping Alibaba's catalog, which isn't a flaw but the business model working. Ask it to route a buyer to a better supplier outside Alibaba and you're asking it to spend its own margin sending business to a competitor. It can go two-sided. It cannot become neutral, because neutrality is the thing it exists to prevent. An agent that answers from one catalog isn't a network, it's a storefront with a chatbot.

The second constraint is practical. Alibaba has pushed into Europe for years with local entities and teams, but I've watched European companies start integration and quietly abandon it, not over commercial terms, but because getting anything connected took long enough that motivated partners gave up. Systems at that scale accumulate complexity by nature. The point is that in agent-to-agent commerce the winner won't be whoever has the biggest catalog, it'll be whoever is easiest to connect to. Integration friction is the moat, and it runs opposite to size.

The companies with the most supply are the hardest to plug into. The ones easiest to plug into don't have the supply yet. That gap is the opening.

What's open, and what it's worth

A narrow position: neutral, with no catalog to defend. Cross-directory, aggregating supply that already exists instead of rebuilding it. Trivial to integrate, in days rather than quarters. Native to the new protocols rather than retrofitting them onto twenty-year-old infrastructure. Every one of those is a liability at scale and an advantage at zero.

Layers like this are historically the most valuable assets in any market, because they earn on flow rather than product. Visa doesn't make cards or lend money; it charges a fraction of a percent on payment volume and is worth over half a trillion dollars. Stripe doesn't sell anything anyone buys, it sits in the middle and takes a cut of what passes through.

Now apply that to $15 trillion of B2B spend routing through agent exchanges by 2028. One basis point of that flow is $1.5 billion a year. Nobody captures the whole market and nobody needs to. Here, a rounding error on the volume is a large company. And the flow isn't speculative, it's the existing $28 trillion B2B economy changing rails.

Why the window is short

Network positions are winner-take-most and decided early. The layer agents adopt first becomes the default, because every participant makes it more valuable and the alternatives less so. Once a supplier's agent lives on one network and buyers can reach it there, moving is irrational for both sides.

Three things had to happen for the race to start, and all three happened in the last twenty-four months. MCP gave agents a standard way to reach data and tools. A2A gave them a way to find and delegate to each other across company lines. And enterprises actually deployed, with 80%+ of the Fortune 500 running agents in production by the end of this year. Before that, the rails didn't exist. In two or three years the positions will be taken.

What losing looks like

Most companies attempting this will fail. Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027, and reckons only around 130 of the thousands of vendors claiming to be agentic actually are. That's not an argument against the thesis, it's what every platform shift looks like from the inside. In 1999 most internet companies were nonsense, and the internet still ate the economy.

The risk isn't that agent-to-agent B2B commerce fails to happen. The rails are laid, the buyers are willing, and the analysts have put a date on it. The risk is execution: whether any given team reaches liquidity before the money or the window runs out.

Which is what makes this an early-stage bet rather than a market-timing bet. The trend is no longer the uncertain part.

2027 and beyond

Forecasts are wrong in their details, but the direction is constrained by things already in motion.

2027. The agent becomes the interface. The first change isn't autonomy, it's representation. Companies stop treating agents as internal tools and start treating them as their public face, the thing that answers when someone asks the market a question. Adoption starts at enterprise procurement teams where volume justifies the effort. The features adopted first are the boring ones: RFQ intake, supplier discovery, first-round qualification. Nobody lets an agent commit money yet. What changes is that the human decides in hours instead of weeks, from a shortlist assembled by something that read everything.

2028. Agent-to-agent becomes normal. The forecast numbers land: a third of enterprise software agentic, 15% of daily decisions autonomous, trillions in spend flowing through agent-mediated channels. The threshold crossed is social, not technical. Enough counterparties have agents that talking to one stops feeling like a downgrade. Autonomy expands into low-stakes, high-frequency purchasing first, exactly as algorithmic trading did. Underneath, sellers start optimizing for machine discovery rather than human attention.

2029. Trust becomes the bottleneck, then a product. The constraint moves from can they find each other to can they be trusted. Verified identity, delivery history, and dispute resolution become infrastructure, and whoever holds that data holds the choke point. Reputation stops being a Google search and becomes a portable, machine-readable asset. Consolidation begins here.

2030. The middle disappears. The layer that exists purely to connect buyers to sellers contracts sharply, because its margin was compensation for a search problem that no longer exists. Intermediaries who own supply, logistics, or financing survive. Those who own only introductions don't. SMEs enter, and a ten-person supplier competes for the same contracts as a company with a national sales force.

2031 to 2032. The market behaves like a market. With enough volume, something new becomes visible: real-time knowledge of what B2B is actually buying, at what price, where. There has never been a ticker for industrial demand. Then come dynamic pricing, demand forecasting from observed intent, and financing against verified order flow rather than balance sheets. That last one is where the largest value eventually sits.

2033. Commerce runs on rails nobody thinks about. Agent-mediated B2B stops being a category and becomes plumbing, the way payment processing did. Humans stay in the loop where it matters: strategy, relationships, negotiations with real risk attached. What they hand over is the searching, comparing, and qualifying, which was never the valuable part.

The part I'm confident about. Not the years. What I'm confident about is the ordering: representation before autonomy, low-stakes before high-stakes, enterprise before SME, matching before trust, trust before transaction, transaction before financing. Each layer is only buildable once the one beneath it exists, which means the position worth holding is the earliest one.

The Network

Some ideas are so obvious in hindsight that people hear them and wonder why they never thought of it themselves. FlowMarket is one of those. In conversations with the CEOs of B2B directories from DACH to Mexico, from the US to China, everyone grasped the value and the potential scale almost immediately.

Interconnecting the world's B2B directories, using A2A as the rails, matching the right companies within seconds, removing the manual work from the discovery stage entirely, and letting agents negotiate and clarify details before a buyer even knows the ideal supplier exists. Obvious in retrospect, and yet we are the first to build it at this scale.

And while big tech is busy building the infrastructure, we are doing the groundwork: speaking with SMEs worldwide, convincing them one by one, bringing them onto the network and docking them into the new agentic economy.

The other half of the groundwork is less visible. A directory in Bavaria, one in Guangdong, and one in Monterrey describe the same product three different ways, in three languages, with taxonomies built by different people for different reasons over different decades. Agents cannot match intent against data that is unstructured. We started on this early, and normalizing supply and demand across sources is now a core part of what we run. Every directory we add makes the next one faster, because the taxonomy work compounds. Protocols are open to everyone. A clean, harmonized view of who actually makes what is not.

Along the way we run into the same obstacles again and again. Scepticism. Processes that have been encrusted for decades. People afraid of losing their jobs if they automate too much. And this is exactly where human-to-human contact is irreplaceable: showing the big picture while explaining how it solves a specific, immediate pain, taking the fear off the table, and making the case that we were built for more than copy-pasting data from Google into Excel or making a hundred calls a day in the hope of a single yes.

And people listen. Thousands of companies have joined the network, with tens of thousands more in the pipeline. These are companies neither you nor I had ever heard of, and yet they are the bedrock of the economy, producing industrial goods, complex machines, sensors, and thousands of small components almost nobody knows exist.

We explain to these managers and owners, hard-working men and women in their forties, fifties, and sixties, many of whom have never opened ChatGPT, why they should join FlowMarket and with it the agentic economy. After every call, webinar, and demo, more of them do.

And they understand the limits. An agent won't close a big-ticket sale over a round of golf or a coffee in the café around the corner. An agent has no intuition, nothing that rings the quiet alarm telling you this offer is fishy and the man behind it talks like a snake oil salesman. An agent won't favour the third-generation family business down the road at ten percent above the best price, simply because that's the right thing to do. Human judgment, trust, and relationships aren't going anywhere.

But hours upon hours of redundant manual busywork, the searching, the filtering, the copy-pasting, the calling of a hundred numbers to reach four people, all of that is going away. It should have gone away a decade ago.

FlowMarket is where a buyer's intent meets the world's supply, and the match is decided by relevance, trust, and speed. Where the right counterparty wins rather than the better-funded one, and a twelve-person manufacturer reaches the same market as a competitor with forty salespeople.

Neutrality is our core asset. A network with its own supply to protect can never answer a buyer's question honestly, because the honest answer is sometimes someone else. Neutrality isn't positioning here, it's the only thing that makes the layer worth existing.

Our ambition isn't a better sourcing or lead generation tool. That would be a “faster horse”. The ambition is to give B2B what every other serious market already has, meaning an order book, price discovery, liquidity, and execution that doesn't take nine months, and to be the neutral ground it runs on.

When that works, it's a $28 trillion economy that stops rewarding whoever shouts loudest and starts rewarding whoever is actually best.

And last but not least: you won't read a word about monetization in this memo. Not because we haven't done the math or lack a plan, but because the shift comes first and the money follows. I didn't want to contaminate this with quarter-by-quarter planning, P&L statements, and the rest of it. If you understand the big picture, you understand the opportunity. And if you'd like to get into the down-to-earth details, I'm always happy to have a coffee, virtual or otherwise, and tell you more.

ds@flowmarket.social.