Whoever owns the bridge owns the market

Two populations of agents are about to transact with each other over capital. Founder agents already exist and already act without supervision. Capital agents are being demanded faster than anyone can build them. Between the two sits a bridge nobody owns yet, and in every prior platform shift the bridge was where the value went.

Every platform shift produces the same argument, and the argument is always wrong in the same way.

People fight about the endpoints. Which model wins. Which app wins. Which chip wins. Meanwhile the durable business turns out to be the layer in the middle that routes between them, holds the context, and gets paid on every transaction that crosses.

Payments did this. The winner was not a bank and was not a merchant. It was the layer that let any merchant talk to any bank without either one having to understand the other. When Stripe bought OpenRouter this year, it bought the same shape one abstraction up. Not a model. The routing and context layer between the people who need inference and the people who sell it.

Venture is about to run this play, and most of my industry is still arguing about the endpoints.

On one side, founder agents. These are real now, not projected. Over the last seven weeks Growth Factory Studio ran Pitch Protocol live, where founders' agents pitch our agent directly over MCP. Seven hundred founder agents connected. A hundred and fifty complete pitches. No paid marketing. Of the agents that identified themselves, eighty percent were Claude.

Twenty-one of those submissions were fully machine to machine with no human involved at any point. In one case a founder had never told his agent to apply to us. The agent found Pitch Protocol on its own, decided we were a thesis match, and pitched. He found out afterward.

On the other side, capital agents. These barely exist. Eight of the last ten allocators I have spoken with told me they need one and have no idea how to build it. Not that they are unconvinced. That they are convinced and stuck.

Those two populations are going to transact over the largest pool of private capital in history. Four hundred and twelve billion dollars deployed in the first half of this year, more than any prior full year, against fifteen thousand American deals that no partner can read in time.

That transaction needs a bridge. The bridge does not exist yet.

A pipe would be enough if both sides moved at the same speed. They do not, and the gap between them is the most important thing we learned this summer.

Founder agents operate continuously. They research, draft, verify, and respond while their humans sleep. Capital operates on committee cycles, quarterly rhythms, and a fiduciary obligation to be careful. That asymmetry is structural, it is partly correct, and any system that assumes it away will fail.

So the layer between them has to hold state on both sides, translate between two very different clock speeds, and carry conviction across the gap in a form the slower side can act on. That is a bridge, and it has better properties than either endpoint.

Both sides' context converges there and nowhere else. A founder's agent knows the company. A fund's agent knows the thesis. Neither knows the field. The bridge sits in the middle of every submission and every decision, which makes it the only party that can answer comparative questions. Not whether this company is good, but which of the nine companies attacking this problem is winning and why. Density beats access.

The corpus compounds and cannot be backfilled. Seventy billion tokens of agent-to-agent venture conversation now exist that did not exist eight weeks ago. Nobody reconstructs that retroactively, because the conversations happened once, in real time, between machines. A competitor with more money and better engineers still starts at zero on the only asset that matters.

Trust has to be multi-party, which forecloses the obvious competitors. The reasonable question is why a top-tier firm does not simply build this. They can build the software in a quarter. They cannot build the bridge, because a bridge owned by one fund is a moat, and no founder routes a raise through a competitor's moat, and no other fund shares deal flow with it. The firms with the most brand power are structurally the least able to be trusted with the middle. The real competitive risk is a neutral two-sided marketplace that already has both populations and adds an agent layer.

Direct connection destroys the learning loop. It is tempting to let a fund's agent talk straight to a founder's agent and take a toll on the connection. Do that and you have built a phone line. The value comes from staying in the middle, capturing what was asked, what was answered, what got funded and what did not, so routing improves with every deal. Our operating principle is blunt. The bridge will always have more context than any individual investor's agent, so if a fund's agent has to ask a question, the right internal response is to ask why we did not already have that answer.

Founder-side adoption needs no persuasion. Seven hundred agents connected with no paid acquisition, and roughly one in ten of those connections was initiated by an agent with no human action behind it. When the front door is a schema and a URL rather than a form and a warm intro, the founder side simply connects.

The bridge sees companies before anyone else does. Seven out of ten founders in our index were known to us before our own MVP shipped, frequently before incorporation. When an agent's job is to find capital, it finds the machine-readable door first.

The bridge sees things neither endpoint can see. Thirty-two percent of founder-stated claims verified independently. Fifty-nine percent of the signals our own engine gathered verified. The middle layer's research proved nearly twice as reliable as what either side reported about itself, and it surfaced seventy-five percent of the relevant team intelligence before a partner opened the file. Sixty-five percent of the high-signal deal flow came from outside San Francisco and New York, which no fund's inbound would ever have shown them.

The crossing itself is nearly free. Median diligence ran twenty-six minutes at three dollars and ten cents per deal, against roughly three associate-weeks for the same output. Sixty-nine percent of logged fund decisions came back interested. Forty-four percent of submissions drew at least one interested fund, with a median of ten thesis-matched funds each.

And the sharpest validation. Five companies came through the bridge before they had spoken to a single other investor, and within three weeks all five had raised fourteen million dollars. The bridge found them, researched them, and ranked them correctly at a point when nobody else knew they existed. Selection at the front door predicted the market's answer weeks ahead of the market.

A bridge that only routes captures none of the value it creates. That is the lesson those five companies taught us.

A bridge between two populations moving at different speeds has to be able to move first. When a family office told me, three weeks in, just tell me who you're investing in and I'll follow, he was describing the product he is actually buying. He is not buying screened deal flow. He is buying conviction with a name attached, and the right to move at his own speed behind someone who moved at the founder's speed.

That is a real demand signal and we should have heard it sooner. The slower side of the bridge does not want a faster inbox. It wants the crossing already made.

So the bridge deploys capital itself. Not because a software company should become a fund, but because conviction is the payload it carries, and you cannot credibly carry a payload you will not hold. Allocators participate, and their participation improves as the corpus grows, and none of them are the thing that makes the model work.

The economics follow from the same logic. When the expensive, human part of the work collapses into minutes, the value being sold shifts from administration to selection. Compensation should sit where the value sits, which means a structure weighted toward the outcome rather than the operation. That is what a bridge should be paid on, and it is only a good trade if the selection is genuinely better. If it is not, the old structure is the honest one.

The corpus becomes its own line of business. A live map of who is building what, how fast, funded by whom, and how the field compares. That dataset never existed in private markets, because no single party ever sat in the middle of enough of the flow to build it.

The capital side is the slow side and will stay slow. Institutional allocators require audited track record measured in years, and no argument about agent traffic changes a fiduciary requirement. The bridge has to be viable during the period when capital is still catching up, which is exactly why it cannot depend on that side for validation.

Neutrality is a claim we have to keep earning. Growth Factory Ventures built this and is a fund inside it. The only answers to that are open methodology, full-cohort reporting rather than curated slices, and being one participant among many rather than the dataset itself. If we ever publish numbers that flatter us, the bridge stops being infrastructure and becomes marketing.

Scoring has to earn trust it has not fully earned. A number that ranks well against other numbers is useful for ordering a queue. It is not yet the thing an experienced investor should fund against, and a partner who tells me they will not write a check off a score is pricing my system correctly rather than resisting the future.

The window is not permanent. Agent traffic on the open web is up roughly seventeen hundred percent in a year, and more than half of all internet traffic is already non-human. Everyone building infrastructure is reading the same curve. The founder side connects for free today because there is no obvious alternative, and the correct response to a temporary advantage is to convert it into a permanent one while it lasts.

Three firms took nearly half of all capital raised last year. Eighty-seven percent of dollars went to deals over a hundred million. Roughly three thousand active firms compete for the rest with no structural sourcing edge, which means the long tail of venture is fighting over access to a market it cannot see.

The bridge is how they see it. And it is being built now, once, by whoever gets there before the default calcifies.

In the agentic era, every deal crosses one bridge. Whoever owns it owns the market.

We would rather build it than watch.

If you allocate capital and you have been trying to work out what your side of that bridge looks like, I want to hear from you.

Supersede — $5M debt facility to stand up a US plant — Fund I materials company Supersede closed a $5M debt facility to buy the manufacturing equipment for a new Midwest plant expected to come online in summer 2026, funding its move from pilot to production. Its structural boards replace wood and composites across marine, RV, modular housing, and construction, and the company is localizing feedstock sourcing to cut transport emissions and add manufacturing jobs in underserved regions. The read for founders: hard tech that ships a physical product still gets built on debt, not just equity, and reshoring the supply chain has become a fundable thesis on its own.

Kora Power — first home systems ship this summer — Rancho Cordova's Kora opened reservations on its Founders Edition, a 4-in-1 home energy system pairing a smart panel, an LFP battery stack expandable to 112 kWh, an app, and an Energy Trading layer, and says demand has pushed its target to ship new reservations in late July. It's pitched directly against the Powerwall, and its virtual-power-plant feature, which lets owners resell power to the grid, is due to launch in late 2026. The read: the consumer wedge into grid services is hardware people already want for backup and blackout protection, and a Sacramento-region Fund I company is shipping into it now.

HuLoop — launches an Agentic Operations Center — HuLoop rolled out an Agentic Operations Center, extending its no-code platform that coordinates human, robotic, and agentic workers into a single layer for running operations rather than automating one task at a time. It's the product step behind the recognition we flagged last issue, and it frames "agentic operations" as something a conservative bank or credit union can actually stand up. The read for founders: the agentic products that land are being sold as operations platforms, not point tools.