The Company Is the Product
The fastest deployment pipeline at Microsoft in the early nineties was a window seat. A customer's network was down, the fix mattered, so you flew out with a floppy disk, sat down at their machine, and installed the code by hand. I know because I was the one on the plane, patching assembly in the network redirector. That was the hotfix, the emergency path. Every other line of code I wrote took a year to reach a customer.
I have spent thirty-five years watching that year collapse into seconds. This is about what the collapse did to the company itself, and to the business of funding companies.
The Latency Collapse
Code-to-customer latency is the most important number in software, and almost nobody tracks it. When I started, it was a year: code shipped in boxes. The web cut it to months. SaaS cut it to weeks. Continuous deployment cut it to days. Agents cut it to seconds. I shipped at every point on that curve.
Here is what the number actually controls: the unit of shipping. At a year, you ship patches. At months, you ship features. At weeks, you ship products. At seconds, the unit moves up a level again, to the thing that contains products. What we used to call a company is now a product.
Four orders of magnitude does not make the game faster. It changes what the game is played with.
Early Is Indistinguishable From Wrong
After Microsoft I founded BeComm and wrote an operating system for devices in C. It connected any media source to any media sink and kept audio and video synchronized across a home network. We built pinch, swipe, and zoom before there were smartphones to put them on. One of the largest chipmakers in the world chose it for their new consumer tablet, a bet on the tablet era a decade before the tablet era. I put ten million dollars of my own money in. Our growth peaked into September 2001, and the company was dead on arrival. Not wrong. Early. The market could not tell the difference, and neither could my bank account.
The lesson was not "be less early." You need to be smart, you need to work hard, and you need to be lucky, and the third one does not keep a schedule. The only strategy against luck is to still be standing when it arrives. Under the old physics you could not afford to stand and wait, because knowing whether you were right required building the product first: a year of latency and a funded team, spent before the answer existed. That is why early and wrong were indistinguishable. They ran out of money in the same place.
The Time Machine
The wreckage had a vault in it. I was a named inventor on nearly a hundred patents from BeComm, and the licensing campaign that followed generated tens of millions. The money was the lesser lesson. Silicon Valley has always disliked patents, for reasons that were once good: when building took years, execution was the moat, and the builder won. The latency collapse ended that luxury. Copying is now as fast as creating, and when implementation is a detail, what remains defensible is what cannot be prompted into existence: proprietary data, customer pull, and the right to exclude. That is what a patent is. Not a trophy, and not a right to build: it turns a parked idea into an option that only you can exercise. The collapse made that option more valuable, not less. A parked idea used to be a promise you could not afford to keep: exercising it took a year and a team. Now it takes weeks. Most parked ideas will expire worthless. That is what options do. You hold them for the one whose market arrives, and you strike the moment it does. I had filed on everything at BeComm. If I had exercised those ideas a decade later instead of spending ten million dollars being early, the story reads differently.
Hold that thought: ideas can be time-shifted. It is half the answer to the timing problem.
The Expensive Part
For twenty years I made the same expensive mistake, in good company. Eight people at Strings. Ten at Aimi to start, upwards of forty by the time we learned we were wrong. A full team at a juice company I founded that was secretly a software company (the ERP kept small juiceries within 48 hours of both the produce and the customer). In every one of them, more money went to the team than to anything else, and it went out before the market had validated anything, because it had to. The team was the build capacity. You bought the team to build the product, and you built the product to find out whether you were right. Every business book canonizes this: culture first, team first, never waver. For that era's physics, the books were correct. They also could not fix the failure mode: when the market finally speaks and the vision moves, the team you have is no longer the team you need.
In January 2024 I wrote in Nasdaq that businesses would move from prioritizing engineers to creative workers who leverage AI to do the knowledge work. That is what happened: AI absorbed the build capacity that teams used to be. Last year I shut down an eight-person offshore team at Strings and rebuilt the entire product myself: a scalable backend, an iOS app, an Android app, at feature parity and finally at the speed of the vision instead of the speed of headcount. Engineering is a tool, not a team. Which inverts the oldest rule in the startup book: the expensive part of company building now happens after the risk is gone.
The Company Builders
Put the two collapses together, build cost and team cost, and a new kind of builder comes out the other side. I can describe them precisely, because I have spent a decade becoming one.
They will build the tools, the infrastructure, the systems that used to require teams. Not just the product: the machinery around the product. Systems that patent what they invent, sell what they build, market what they sell, run what they own, and keep idle capital working. Departments used to do this work. Machinery will do it, machinery that replaces the headcount builders used to buy before they could afford to be wrong.
They will fail ideas constantly, and the failures will cost almost nothing. Failing fast will not mean discarding. It will mean iterating cheaply until the world is ready or the idea is disproven, and the difference matters: a parked idea is not a rejected idea. It is a held option, kept alive by patents and by a build cost so low that waiting is finally affordable.
They will let the vision move. The never-waver founder was a creature of the old physics: when a bet costs millions and years, you need an operator too invested to question it. When code follows the vision in real time, stress-testing the idea becomes the discipline, and stubbornness stops being a virtue. The obsession does not disappear. It gets assigned to the stage where it pays: at spinout, the product that proved itself gets a founder who will never waver about it.
And they will not staff anything until it proves itself. The reveal will not be a demo, a waitlist, or the builder's own conviction. It will be a real customer using the product in production and pulling for more. When a product passes, it will not get a project team. It will get a founder, a company, and a cap table. It will spin out. The company is the product.
None of this is a prediction. In January 2012, Forbes published my description of the model: multiple products developed at once, "a disciplined and repeatable process" for prototyping and market testing, engineered for "the early and low-cost failure" of some products and the rapid growth of others, in a company organized by function instead of by product. Five months earlier, someone had written that software was eating the world. This is what it looked like from inside the kitchen. The only thing missing in 2012 was the technology. It arrived.
We have run the full loop. A decade ago we filed patents on AI-generated music before that phrase meant anything; they are in the public record. When the product proved viable, the first hire was not an engineer. It was a founder, found by reading research papers until the right PhD surfaced, and the company was built around him. It spun out with the patents. One of the largest companies in the world made a run at acquiring it; we declined, and it closed an institutional round within a year. Then the market moved: foundation models arrived, a purely generative competitor leapfrogged us and got stuck exactly where we knew the bodies were buried, rights and royalties. So we did the thing only a builder with a machine can do. We recalled the product, pulled it back to two people from the original prototype team, and rebuilt it on the new stack. We learned. We adapted. We built more with two people than we did with forty. And the loop keeps accelerating: this year Implicit Media went from first customer meeting to shipped product in weeks, with a team of one, past incumbents built to sell finished products, not to build solutions.
I wrote that AI would equalize knowledge. It did. What it doesn't equalize is the machine. Everyone has the models. Nobody has the machine. The models are the commodity. The machine each builder assembles around them is the moat.
The Renaissance Engineer
Who are these builders? The tempting answer is anyone: if building is free, why not a twenty-one-year-old in his parents' basement?
The answer is a kind of engineer, not a resume. Anyone can prompt a simple web service into existence now. That is exactly why the classic SaaS trade, monetizing the difficulty of building services, is dead. I have used the anyone-can-code tools myself. They are opinionated for a reason: the guardrails keep you from building crap by keeping you from building anything complex. The guardrails are the product, and the guardrails are the ceiling. Systems that scale vertically and horizontally, iterate weekly, and survive contact with sales, marketing, support, and a half dozen cloud topologies require architecture. AI made implementation a detail. It did not make architecture a detail.
Architecture is learned the long way. I took my first computer apart the day it arrived and spent two weeks putting it back together, soldering the memory I broke so my mother would not find out. I learned assembly by cracking game copy protection, learned a dozen languages before I ever shipped a line of code, and shipped my first line into a network stack where adding a print statement made the bug disappear. The kid has the same models I do. He does not have the failure modes.
For thirty years this industry priced experience as depreciation. The latency collapse repriced it. When implementation was the bottleneck, you hired specialists by the dozen and the generalist was overhead. Implementation is now nearly free, judgment is the bottleneck, and the renaissance engineer, who builds the product while understanding how to market it, sell it, and support it, is the founder of the future. The machine is that judgment with leverage, and it does not run alone: every product it ships leaves with its own founder and its own team.
Where Venture Goes
You know the old pipeline: years in a garage before a customer exists, friends and family bridging to a first deployment, a team hired around an unproven direction, an institutional round whose first purchase is more headcount, and the concept still unproven years in. Capital still ships the way software shipped in 1990.
Here is the industry's open secret: venture never really spread its bets. Look inside any fund's portfolio and you will find the same names funded again and again, the second company, the third, the pivot after the pivot. The spray was always concentrated on proven builders, because the founder was the only durable signal in the noise. The idea was the excuse. The person was the asset.
A company builder is that logic completed: what the repeat founder becomes when the team constraint is removed. Not the second company after the first, but all of the next companies at once.
The power law is real and I have no argument with it. The insane returns come from outliers that looked insane at seed, and nobody picks the one in a hundred in advance, because the deciding variables (timing, market readiness, the builder's own life) are outside anyone's control. The company builder's answer is not better picking. It is cheaper tickets. Hold the options at near-zero carrying cost, and staff the one that hits alignment. The machine does not pick the outlier in advance. It delays irreversible commitment until reality has begun to identify it. Capital compresses the way product did: deployed incrementally until validation, then poured on like fuel.
The old model built rockets without knowing the target. Fuel a hundred, and if you picked the right design, the right fuel, and the right crew, one hits and returns billions. That math worked for the fund and for no one else: the other ninety-nine crews spent years of their lives on rockets that went nowhere. The machine fails ideas, not people.
So follow venture's own logic to its conclusion. If the founder was always the asset, then a proven builder who no longer needs a team to build no longer needs seed capital to build either. What leaves a builder's machine carries mostly scale risk, which is exactly what growth capital prices well; the first company out of ours is something institutional venture bought. Early stage becomes manufacturing. Venture does not die. It moves to where the risk went.
At Endanik
We have been building this future under one roof for a decade. We call it Endanik. This is how we do it.
The first company out the door was Aimi. Patents filed years before the market existed, a founder found in the research literature, a team built around him, a spinout, an institutional round. And when the market moved, the recall: not to cut costs, but to build new models. We are not afraid of research. The machine's products begin as models and datasets, not features, and research did not end when the products shipped. It is a subsystem.
The follow-ons had the advantage Aimi's first life did not: agentic coding from the first commit. Implicit Media, a team of one, shipped past incumbents in weeks. Strings came back from an eight-person team as a product rebuilt at the speed of its own vision. Each company that proves itself leaves the machine with a founder, a team, and a clean story for the market that prices it.
Capital participates through an instrument we built because nothing existing fit: the SAFE Pool. Seed equity prices unretired risk. Fund interests charge a decade of fees for someone else's picking. Holdco equity traps the winners on the parent's balance sheet. The pool holds low caps across every spinout the machine ships, converting when the venture market prices each company later: the upside of a direct investment, with the optionality that is native to the model. The mechanics live in our term sheet. The concept fits on an index card: bet on the output, not the guesses.
Beyond that, we stay quiet. The products speak on their own schedule.
What Happens Next
Companies will be built this way from now on: quietly, cheaply, in machines that fail ideas by the dozen and ship the survivors with founders attached. The ones you can see today look like anomalies: a music company recalled and rebuilt by two people, a system deployed past incumbents in weeks. They are not anomalies. They are early. The builders running them will mostly be invisible until their companies are not. That is the nature of machines. They do not tweet. They ship.
It took a plane ticket to ship a bug fix. The next company ships before lunch.
I have no social media. This page is the only place I say things.
The company is the product.