Designing for a use case


When you develop an application or some software artifact you consider what task the program is to complete. You make a model of who your customers are, and how they interact with your service, and then you proceed to implement that. Unfortunately you cannot predict every possible use case for your program, you end up choosing a few prime use cases and focusing on those. This makes them easy to discover and use, which is great. But what if those aren’t the operations your users really want?

I would argue that these compromises make it very difficult for your customers to complete complex tasks, and also alienates the user if their task is one of the marginalized ones. They could imagine that their task is so insignificant that nobody would bother to produce a system that could do it. But this really does not have to be this way, we could instead give the users power over what they do with their data. We need to expose the functions people want in a logical consistent manner, and then let them interact with those items however they wish.

This requires logical consistency throughout the system. An operation on one item should have the same effect as on all other items that are logically equivalent. Users must not worry about their data-structures, and those data-structures ought to be dynamically optimized to account for the users usage pattern.

Common operations across collections

Anyone ought to be able to apply bulk operations over any collection, but right now the only collections that really support this are music, emails, and perhaps transactions. I ran across this problem trying to remove old iPhone apps from my iCloud list. Apple lets you remove them, but only one at a time. There is no way to select more than one item, and remove all of those selected items from the list. There's no doubt that it is technically possible for this to be a feature, but Apple doesn't care if you can do it easily. Easy should be the default, and adopting a universal way to deal with data is part of the solution to this problem.

The issue is that developers don’t think its important for their application to support those operations. Users as a result think they don’t need them, and they’re not relevant to what they want to accomplish. The result is a severely suboptimal environment for those using our products.

I think the solution is to expose the programming to the user, not thru a terminal, but thru a GUI. The operations have to make intuitive sense for any reasonable person. Data structures will have to be persistent and invisible to the user. This is why I’m currently working on UDB, a collection framework that dynamically optimizes data structures given the most common access operations. This flies directly in the face of optimization and performance, however this is not a huge issue as collections are generally small, and programmers can optimize for certain use cases, eg photo applications, or email, to provide better performance. They key thing though, is that all operations have to be supported by all underlying data structures. This gives people a consistent frame of mind, without it the system becomes to complex for normal people to grasp.

Common operations across logical datatypes

Another issue is that the same types of data are represented in different ways, without letting a universal set of operations that apply to data of the same logical type. This is another issue where the user does not understand why it needs to be this way, and it really doesn't. The key to fixing this issue is providing a large library of operations that work for logical primitives. Developers can specify a single method to translate their current data format into the format that the user prefers. Some examples of logical primitives would be times, locations or various forms of media. There are quite a few other primitives, you can find a full list here.

Because these types are consistent, and supported throughout the system, you can apply common filters, sorts, or apply functions to every item in the collection.

For example, I would like to find a contact of mine, but I've forgotten their name. I know I added them recently, and that they didn't give me their email. I ought to filter by contacts who do not have an email, then sort by how recently I added them. I have about 600 contacts in my address book, and neither of these operations would take much time, even if they are not heavily optimized.

Imagine the power if you could do that for all these data types. You could find the closest x just by getting a list x then because they have a location, you can find out their distance from your current location. The coffee shop use case has been polished, and for coffee shops its super easy. What if you're looking on someones blog for good gluten free restaurants in Sacramento. You think this person knows what they're talking about, and you'd like to know how far away all of them are. Well, right now you have to type them all in to Google maps, or have knowledge of the area to figure out how far away they are. This is obviously sub optimal. If you have a list of restaurants, even just their names and a general geographical location, you ought to be able to find the distance from you to them without having to code something to do it, or type in 10 address to Google, and compare across multiple tabs.

Programmers provide the data, and the user manipulates it

One of the huge benefits of such a system is that developers can focus on the applications that are critical to what they do. That is their business logic, or leveraging their unique dataset. They don't have to develop an app, they don't have to really bother with data conversion, and they don't have to focus on the client side. Furthermore any added function to a primitive will immediately propagate throughout the entire system. This means that code really only has to be written once, and each developer doesn't have to spend time solving problems that have already been solved.