Software Development is a Search Algorithm

How we think about something can have a profound effect on how we do that thing. Thinking of the process of software development as a search algorithm rather than as the process of building software can lead to making better decisions about the products we create and the way we go about creating them.

The Search For Value
The end goal of designing software is to produce something which provides a benefit that is large enough for one or more customers to at least cover the cost of its production. Of course, the higher the value of that benefit, the more that folks will pay for it. On the other side of the equation you want to provide that benefit at the lowest cost. There is a point of diminishing returns both in looking for the highest possible value and in looking for the lowest possible cost.

As much as we might want this process to be deterministic, it isn’t. Just because you produce something doesn’t mean people will use it, and just because they use it doesn’t mean they will pay you what you want for it. The only way to truly find out is to produce something and see how much value can be realized from it, for instance by charging money for it.

The Max For the Minimum
In other words, producing software is a classic min/max search. You are searching for business opportunities with maximum value and implementations which minimize cost. You can think of it as a huge graph of opportunities and below each opportunity node is another graph of possible implementations. Complicating this is the fact that you would like the features to be close to each other from a market perspective in order to maximize the leverage of your sales and marketing departments and you would also like to have as much common implementation as possible in order to maximize the leverage of your existing implementation and engineering resources. For both you would also like to leverage your core competencies.

The faster you can traverse this graph, the faster you will unlock business value. Unfortunately, you only have a finite amount of time and resources, so you can’t run a search by producing all possible product variations and seeing how they sell. Plus, the search space is infinite. So, you have to make predictions. You look for ways to eliminate certain paths and to get more information about other paths without actually implementing them. Whatever your methods are for figuring out what to produce, they are all based around searching the solution space as fast and efficiently as possible.

Searching For Value in All The Right Places
You may think that only parts of the development process are parts of this search, but doing so will limit the effectiveness of your searching. For instance, if you arbitrarily decide that a particular feature is what your customer wants and you then spend a huge amount of time talking to your customer base about the requirements for that feature, you are limiting your search to the various ways to do that feature, but what if the perfect implementation of that feature is only worth $10? You are then searching within a low value set of possibilities. Before you dive in too deep, spend a fair amount of time at a high level so as to maximize the value of searching at the next level down.

The entire development lifecycle is part of your search. The first level of searching is to find potential high value opportunities from which to create new functionality. But don’t fall in love with a particular opportunity too quickly because implementation may cost more than the value to the customer. There are usually a variety of ways to implement something, so don’t fall in love with a particular design too quickly because there may be a cheaper implementation. And of course in order for the customer to determine whether they want to pay you or not they need to get their hands on it or at least see it in action. So, you are searching for opportunities and then within the opportunities you are searching for implementations and then you need to circle back with the customer to check the results.


Next: Specification is Planning is Design is Coding is Approximation