How do you design a car? – It is one of the most famous diagrams in the agile world drawn by Henrik Kniberg.
I’m guessing many readers know this already: one approach (top of the diagram) is to iteratively design and build all the pieces, put them together and you have a car. This is one interpretation of iterative but until you put the pieces together there is no feedback and no value.
After all, who wants half a car? We know what we want from a car, right? Who needs feedback? Who wants a car with three wheels? – Why waste time experimenting?
The alternative, advocated by minimally viable product people everywhere is to redefine the problem; we don’t want a car so much as transportation, we could start with a very simple – and quick to deliver – transportation system (the skateboard). Because it is delivered sooner we get feedback sooner. We see how people use it, and evolve it over a series of iterations into a motorised car.
Since I know this, you know this and it is on the back of every agile cereal packet one can rest assured that Larry Page, Tim Cook and Jeff Bezos know this, right?
Well no, not according to the Financial Times recently. The FT carried a piece about the development of autonomous cars – “Robotaxis: have Google and Amazon backed the wrong technology?” – paywalled. Since we already have the sort taxi someone drives the development effort goes into advancement.
For the last few years Google/Waymo, Apple and others have sunk billions of dollar – yes billions – into developing self-driving cars. And of course, we all know what we want from a car, even a self-driving car so this is an engineering problem.
These cars now work but there are a number of challenges before the achieve world dominance. Most of the challenges now are less to do with the technology and more to do with the market: customer acceptance, insurance, pricing, etc. Still, billions more are needed before any return can be achieved. In other words: Google etc. took the first approach, build the pieces.
What is less well know, and what the FT writes about, is that another group of companies has take the other approach. Component suppliers like Bosch and Magna, and tech companies like Mobileye (Intel) have been developing desecrate technologies that can be incorporated into existing cars which evolve towards self-driving dream. Not only is this cheaper but it is easier to market and clear regulatory hurdles because humans are assisted in driving not replaced. (Tesla is also in this camp as they add more and more capabilities to their auto-pilot features and have been getting feedback from real customers for years.)
Now it seems evolutionary approach may win-out against the big clean sheet of paper. The race is not over yet but the evolutionary suppliers are making money while the new designer are still burning cash. The evolutionary suppliers are integrating their tech into cars and getting regulatory approach while the new designers have to navigate many regulators.
Engineers often object to the evolutionary model because: “we need to see the big picture”, “you need an architecture”, “you can’t evolve a skateboard into a car”. And indeed, one of the Google engineers, quoted in the FT saying:
“Conventional wisdom would say that we’ll just take driver assistance systems and we’ll kind of push them and … over time, they’ll turn into self-driving cars … well that’s like saying that if I work really hard at jumping, one day I’ll be able to fly.”Chris Urmson, 2015
When you consider the problem purely in engineering terms this rings true, but while one needs to respect engineering it is not the only frame of reference. One need to consider the commercial and marketing aspects, as well as customer acceptance and other factors. To give any single line of thought a privileged position is to expose yourself to risks from the others.
At the end of the day, as I have argued repeatedly: engineering is about creating solutions within a context, within constraints. To any given problem there are many potential solutions – many ways to slice the onion. The “best” solution is the solution which best fits those constraints.
The evolutionary approach allows you get feedback sooner which allows you to uncover those constrains sooner, that allows for course corrections and it also means less time and money has been spent on solutions which don’t meet the constraints.