The first thing you build for anything new inside a big company is the team, and you have to build it out of people who already have jobs. From outside, starting something inside one of the big platforms looks like the easy version of starting something. The money and the infrastructure are already there, and a billion users are, in principle, one code review away. The one thing nobody can hand you is conviction.
Startups collect conviction at the door. Anyone who joins a six-person company has already talked themselves into it, or they wouldn't be there. Inside a big company the filter runs the other way. The engineer you want has a team that likes her, a manager who rates her well, and a promotion that's eighteen months out on her current track and who-knows-when on yours. You can't offer her more money; compensation is decided somewhere far above both of you. All you can offer is the idea, plus your own certainty about it. So long before any executive hears about a new project, you're pitching it to the person two desks over, who has every reason to say no.
Recruiting has been the hardest part of every new product I've worked on, including the one I'm on now. It has also been the most useful part, though that took me a while to see. The engineers I wanted most were exactly the ones who were hardest to move, and their objections were almost never about risk to their careers. They were about the idea. Why would anyone want this? What breaks when it's a hundred times bigger? If a pitch can't survive twenty minutes with a skeptical engineer, it probably can't survive contact with users either. Recruiting turns out to be a cheap way to learn that early. Some ideas died in the hallway before anyone wrote a line of code, and dying in the hallway costs a week instead of two quarters.
The pitches that worked were never the grand ones. Grand visions are cheap inside a big company; everyone has seen a dozen, and seen eleven of them die. I contributed one of the eleven. My first pitch was a deck about everything the thing might someday become, and I'm fairly sure one slide said "ecosystem." The engineer I wanted most listened politely, then asked what the first user would actually do with it. I had no answer. That one died in the hallway, and deserved to. What moved people was specificity -- a particular broken thing, and a first version we could put in front of real users in eight weeks. That's roughly the opposite of what I'd assumed persuasion required.
Once a project becomes official you can staff it the normal way. The team I helped lead grew past twenty people, engineers and designers and data scientists and product managers and researchers. Plenty of them arrived by assignment rather than persuasion, and that's fine. But a team like that runs on the belief of about six people, and everyone knows who they are. The other fourteen watch those six the way passengers watch the flight attendants during turbulence. As long as the six are calm and certain, the twenty ship. When the six waver, no amount of headcount saves you.
Most work at a big company needs none of this, and shouldn't. If you're keeping a mature system healthy, the org chart is the team, and the system supplies its own conviction. I've done that work too (some of the infrastructure I've helped scale serves more than four billion interactions a day), and nobody has to be talked into believing it matters. It's visibly load-bearing. A new thing is, by definition, not load-bearing yet. Belief in it has to be manufactured, and belief is manufactured in small rooms.
This pattern is much older than software. The Royal Society, which more or less invented modern science, began as twelve men who stayed behind after a lecture at Gresham College in 1660 and decided to keep meeting weekly.[1] And in 1943 Kelly Johnson talked Lockheed into letting him take a couple dozen engineers into a rented circus tent next to a plastics factory, where they built America's first operational jet fighter in 143 days. The tent smelled so bad the engineers named the place Skunk Works.[2] People usually read these stories as lessons about autonomy, which is to say put the new thing far from headquarters and let it run. I think the real lesson is density. In a small room the fraction of people who believe is one hundred percent, and decisions happen at the speed of conversation.
Speed matters more than it seems, because conviction decays. Inside a big company it behaves like water pressure in a tall building. Strong on the ground floor where the idea started, a trickle by the fortieth floor where the approvals live. Somewhere in between, someone has to install a pump. On every project I've led, the pump was a demo. A demo does something no document can. It lets people feel the thing exist. I've watched a room of fence-sitters convert in ten minutes of clicking around a prototype, after weeks of design docs had moved nobody. So I've come to think the real audience for a demo is the team itself. Executives read it as progress. The team feels it as proof.
People say big companies can't do new things, and they usually blame process. I don't think that's quite right. Process is what a company builds so that conviction stops being necessary, so that ten thousand people can cooperate without each of them having to believe in what they're doing that day. For mature products this is a triumph; it's why the site stays up. But it means the machinery of a big company is designed, on purpose, to run without the one fuel a new thing needs. Nobody believes in a new thing yet. That's what makes it new. Conviction doesn't scale. It has to be grown one person at a time, by hand.
So if you're inside a big company and want to start something, don't begin by asking for headcount. Begin by finding the four or five people who can't stop arguing with you about the idea, the ones whose objections keep getting better. Then find a room small enough to hold them, and close the door.
Notes
[1] The lecture was Christopher Wren's; he was one of the twelve, along with Robert Boyle. And they didn't just resolve to meet. Each of them agreed to pay a shilling a week to fund the experiments, which is about as concrete as belief got in 1660.
[2] The smell, which drifted over from the plastics factory, only supplied the association. In Al Capp's comic strip Li'l Abner there was a foul-smelling factory called the Skonk Works, and one of Johnson's engineers, Irv Culver, took to answering the phone "Skonk Works, inside man Culver." The name stuck; Lockheed changed the spelling years later when Capp's publisher objected. The 143 days wasn't a stopwatch number either (the contract allowed 180, and they beat it by 37).