The pitch was a proxy for a test we couldn't afford

I listen to David Senra’s Founders podcast most weeks, and episode #426 is about Christopher Nolan. I didn’t expect to come away from an episode about a film director with an idea about AI, but I’ll take it.

Nolan has said Inception was “something I’ve thought about off and on since I was about 16”. He said that to the Los Angeles Times during a break in shooting in the summer of 2009, which puts about twenty-three years between the idea and the first day of filming. He pitched it to Warner Bros. around 2001, right after he shot Insomnia. They were into it. He pulled it back anyway. His account, given to FilmInk: “They were interested and excited about it, but I ultimately decided that I wasn’t ready to finish writing it. I wanted to write it on my own, and then bring it back to them. I didn’t think that it would take nine more years though.” The movie came out in 2010.

How much agreement it takes

Nolan holding onto Inception for that long reminds me how much agreement it takes to build anything inside a company. First you convince someone to fund the idea, and you edit the idea a little to make it fundable. That edit is rarely dramatic and the person with the idea rarely resists, because doing so means the project goes nowhere. By the time anyone builds anything, the idea has already been shaped by the people who had to approve it.

AI changes this in some domains. It won’t let the next Nolan make their movie alone, but it can compress building a technical product far enough that internal projects and startup ideas get conceived, built and tested by one person. So the need to persuade drops away, and with it the need to bend the idea to suit a decision maker. But we lose something without the approval step, namely the refinement that comes from the approvers’ experience.

What twenty-three years actually gave him

Nolan made Inception on a $160 million budget with a studio crew behind him. If working alone were the answer, he’d have shot it at seventeen on whatever camera he could borrow.

The wait gave him standing. In 2001 he had Memento behind him and Insomnia just shot, and he pulled the pitch back himself because the script wasn’t finished. By 2009 he had The Dark Knight, and a director in that position doesn’t have to make an idea fundable, because he is the reason it’s fundable. That’s the version of this everyone wants: the idea reaches the world without being edited to suit anyone.

Twenty-three years is a long time to wait for it.

The persuasion was doing a job

The persuasion you skip was a filter. Most ideas that get watered down deserved to be. Nolan gets a podcast episode because he turned out to be right, but nobody records an episode about the person who protected a vision for twenty-three years and was wrong about it. There are a lot more of them.

Take the filter out and put nothing in its place, and you get the same bad ideas as before, built faster and defended harder, because a working demo persuades whether or not the idea under it is any good.

What the pitch was actually for

Persuasion was a proxy.

Building was expensive, so somebody had to make the call before any evidence existed, and a pitch was the cheapest feedback available. Studios green-light from a treatment for exactly that reason. Film costs too much to find out by shooting it, so an executive reads a description of the movie and decides. Software had its own version of that committee, and it ran on the same logic. Six engineers for two quarters is a lot of money, so we asked for a document and a room full of opinions about the document.

What changed is the cost of finding out, and that cost was the whole reason for the pitch. When building gets cheap enough, you stop soliciting feedback on a pitch and focus on measurement instead, which is a straight trade of the pitch for the test. I like that story better than the one where AI sets the builder free, because now the working software and the people trying it are what tell you no, instead of one person in a room reacting to a summary. You still get told no. It just happens after the software exists.

That also puts the filter back, in a better form. The old filter ran on somebody’s judgment about a description. The new one runs on whether the software works for the people it was built for, which is the question the description was standing in for the whole time.

Cheap tests kill slow ideas

Here’s what I can’t resolve. A cheap test is very good at killing an idea quickly, and Inception spent about fifteen years as an idea nobody could have tested, then nine more after Nolan pulled the pitch back. Give that sixteen-year-old the tools to build a rough version the same year and it would have landed badly, and I doubt he holds onto the idea for another two decades after that.

Measurement beats a committee at telling you what’s working now. What I don’t know is what it does to the ideas that need two decades of somebody thinking about them before they’re any good. Those ideas never survived the committee either. They just used to survive in a drawer, where nothing could test them.

Key takeaways

• Nolan had the Inception idea at about 16, pitched it to Warner Bros. around 2001 and pulled it back to write the script himself, and started shooting twenty-three years after the idea first turned up.
• The wait gave him standing, and standing meant nobody could ask him to change the idea to make it fundable.
• The persuasion AI removes was filtering, and most ideas that got watered down deserved it. Remove the filter and put nothing in its place, and bad ideas ship faster with more conviction behind them.
• Persuasion was always a proxy for a test we couldn’t afford to run. Cheap building trades the pitch for the test, so the no comes from working software and the people trying it.
• A cheap test kills a slow idea fast, and some ideas need twenty years before they’re any good.