Tesla’s New Car Smell

... Thus, when [Freeman] Dyson arrived at the Institute for Advanced Study in September 1948, Feynman diagrams in hand, he immediately joined ten fellow ‘postdocs’ in theoretical physics, the largest cohort yet.

[J. Robert Oppenheimer had reorganized the Institute to emphasize post-doc training for physicists; he was quoted as saying that "the best way to send information is to wrap it up in a person."]

In fact, the new building that had been planned to hold offices for the newly expanded ranks had not been completed on time, so all of the postdocs wound up sharing desks in one large room for the first 6 weeks of their stay – an architectural arrangement perfectly suited for the sharing of tacit knowledge.

They huddled around large tables swapping ideas, pressing Dyson to explain the details of how to use Feynman diagrams. He delivered several sets of long lectures to the group, supplemented by constant, informal conversations.

By the winter of 1948–49, Dyson had organized them into several groups, each working on distinct diagrammatic calculations. One pair, Kenneth Watson and Joseph Lepore, worked with Dyson to calculate high-order terms in a certain model of nuclear forces, while another duo, Norman Kroll and Robert Karplus, took up Dyson’s lead to undertake similarly complex diagrammatic calculations in QED.

So effective were Dyson’s tutorials that Wolfgang Pauli wrote to another of the young Institute theorists that May, asking what Dyson and the rest of the ‘Feynman-school’ were working on.

(Extra paragraphing added.)

Many often wonder why large software system projects are difficult. Many often wonder why running large organizations such as governments is difficult. In this topic some wonder why large scale vehicle manufacturing processes are difficult. These fundamentally share properties.

One key challenge is that system, organization, and process complexity do not grow in a linear manner. If you have ever tried to estimate software development time for larger projects you understand how wildly inaccurate it can be. The number of interaction points multiply with growth. Progress feels extremely fast at the beginning. As the system surface area grows it becomes increasing difficult to push incremental improvements forward. A single person or small team comprehensively understanding the entire system and keeping enough of it in working memory becomes a challenge. There are design and management strategies to combat the challenge but in my opinion only time and iteration can truly overcome.

The single most important long-term factor here is whether quality and continuing improvement is part of the culture or not. I believe they are, as suggested by the production target miss. A process of prioritizing and then fixing sub-optimal system and process interaction points, both internally and externally, will drive the company to high output while maintaining quality over time. TPS [0] concepts are well known but one doesn't simply download and install it in an afternoon. It can take decades to optimize.

The single worst decision the company could make is to pump out vehicles to meet a production target metric while sacrificing quality. That has been tried before in this country (and at the same factory several decades ago no less [1]) and did not work out well.

For perspective: Competitive threats are just not a very big deal. Globally, 72 million cars were produced last year [2]. If Tesla captures just 1% of that market it is still a huge achievement - 720k cars per year. Even if we assume they sell only the lowest price Model 3 at $35k and ignore all other vehicle and energy products that would equate to over $25 billion in revenue - 10 times the 2016 annual figure. The market will also grow as the global economy does. There is plenty of room.

[0] https://en.wikipedia.org/wiki/Toyota_Production_System [1] https://en.wikipedia.org/wiki/NUMMI#Background [2] https://www.statista.com/statistics/262747/worldwide-automob...

My Odyssey can't do that.

Now I'm confused why Tesla would sound so limited, unless they are trying to cover their ass legally on how you SHOULD use autopilot versus how you CAN use it?

Personally, we don't share much - she drives hers, I drive mine and only share duties in the weekend.

With our separate commutes, and a PHEV and and EV we could literally not visit gas stations except on road trips.