On innovation, giants, and skunks

Five firsts

The tent became the Skunk Works, named after a reeking still in the comic strip Li'l Abner. A shop that good could afford a joke for a name; the more competent you are, the less serious you need to be. Over the next four decades it built five airplanes that each did something no airplane had done, and Ben Rich, who ran it after Johnson, lists them himself:⁵

• P-80 Shooting Star, 1944. The first jet fighter the US Army Air Forces put into service, and the first of its jets past 500 miles an hour.⁴
• F-104 Starfighter, 1954. The first production airplane to fly at twice the speed of sound, on wings so thin at the front edge, 0.4 millimetres, that ground crews fitted guards over them so nobody got cut.⁶
• U-2, 1955. A jet-powered glider that flew over the Soviet Union from above 70,000 feet, so high its pilot wore a pressure suit. It still lands with a second pilot racing behind it down the runway in a car, calling out the height.⁷
• SR-71 Blackbird, 1964. Built mostly of titanium, because at three times the speed of sound its skin heated past 260 degrees Celsius. It was made to fit when hot: on the ground its fuel tanks leaked onto the runway, and sealed only once the skin warmed in flight. Since 1976 nothing with a pilot and a jet engine has flown faster.⁸
• F-117 Nighthawk, 1981. The first stealth aircraft in service, shaped like a cut gem because the 1970s software that predicted its radar echo could only handle flat surfaces. The equations behind it came from a Soviet physicist, and the Air Force denied it existed for seven years.⁹

"Best" would be too generous for all of them. West Germany lost 292 of its 916 Starfighters and 116 pilots, and Germans called it the widowmaker.⁶ "First" holds every time.

The rest of Lockheed had the same name on the door and the same government paying the bills. In 1971 it was the largest defence contractor in the United States, and it needed a $250 million federal loan guarantee to survive an airliner.¹⁰ So money does not explain the difference.

Why giants stop inventing

A big company is a machine for repeating what worked. Ninety thousand people building P-38s need procedures, and good ones: they are why 23 airplanes came out every day. A procedure is a decision somebody made once so that nobody has to make it again, and a new airplane is made almost entirely of decisions nobody has made before. The machine that makes the known thing cheap makes the unknown thing slow and expensive, with exactly the same parts. And an untried idea has to compete for money with work that is already earning; Clayton Christensen called this the innovator's dilemma, and noticed that the firms it beats are usually the well-run ones.¹¹

Every part is reasonable, which is why nobody removes it. When Johnson was sent to rescue the Agena rocket stage, he found 1,206 people in quality control and a reliability of 13.6 percent.¹² Each inspector had been hired to stop a real failure. Together they were enough people, he wrote, to design and build the thing.

Then there is the arithmetic. Every pair of people on a project is a conversation that might need to happen: five people make ten pairs, fifty make 1,225.¹³ Anyone who has tried to choose a restaurant for twelve knows the rest. Most of those pairs never talk, but the more people there are, the more desks a decision has to visit, and every one of them can say no.

And there is pay. In most companies the manager of forty earns more than the manager of four. Reward a manager for building a bigger department and a bigger department is a reasonable thing to build. Johnson told Rich the main plant gave raises "on the basis of the more people being supervised; I give raises to the guy who supervises least", and he put it in his fourteenth rule.¹²¹⁴

How the skunk did it

Johnson's answer was a small room inside the giant, paid for by the giant, with the giant's habits kept outside. His third rule says the people connected with a project "must be restricted in an almost vicious manner".¹⁴ The U-2 never had more than 81 people on it, and they handed back $2 million of a $22 million budget.¹² Designers sat "two feet apart" from the people who built and tested what they drew, so a drawing that could not be built got argued over before lunch.¹² One person decided, and someone above him took the pressure. Rich told his engineers: "I'll take the crap from the big wheels, but if you screw up I want to hear it first."⁵

Even the great exception worked this way. Bell Labs had about 5,700 people in 1950, paid for by the AT&T telephone monopoly, and the transistor, which the modern world more or less runs on, came out of one small mixed group of physicists, chemists and metallurgists inside it in 1947.¹⁵

The giant's reflexes are strong, and copies usually lose to them. The Army's Cheyenne helicopter tried Johnson's method and within six months had a purchasing department larger than his whole engineering department.¹² When investors made Olivetti sell its electronics division to General Electric in 1964, five engineers kept the desk computer they were building by renaming it a "calculating machine", which was not in the deal. The Programma 101 sold about 44,000 units, and NASA bought some for Apollo.¹⁶

What a beaver has

Castor, which is also Latin for beaver, is two people making open-source software and these essays. Johnson's arithmetic is very kind to two: one pair, one conversation. Ours never really stops, because we live together, so most ideas reach the other person while they are still half a thought. Nobody above us has to approve anything. A decision takes as long as it takes us to agree.

We also have what Johnson never had: agents. They add hands without adding people who need persuading, and the cost moves to context. People pick context up fast, in a corridor, and there is a limit to how much they hold; agents pick up nothing by being in the room, but they can work from far more than a person can hold, as long as someone wrote it down. So most of our work has become scoping: saying what we want precisely enough to hand it over, with just enough structure that nobody has to guess, and a check at the end. The risk is the one Johnson fought, in a new costume. Generate work faster than we can judge it and we will have our own sprawling department, conveniently stored on a computer. If it starts to happen, we make the jobs smaller before we add more agents.

The rule we most want to keep is the one about pay, because growth breaks it first: the day there is money to hire, the easiest way to look successful is to have more people. Nobody is offering us that money yet, so we wrote down what we reward while it costs nothing, in The myth of values as a competitive advantage, and why we commit slowly in Startup decelerator.

Every giant was once a few people who could decide before lunch. The skunk needed a tent to hide from its own company. We would like to grow without ever needing one.