on platforms

I’ve worked for Google for about 4 months now, perhaps a bit less. In that time, my thinking on the future of computing has changed (because I’m sad I think about this a lot).

This thinking is the result of a vague feeling that the existing client platforms have all run out of steam. Windows Vista, OS X and the latest Linux distros are all, give or take a few details, pretty much the same. Most of the new features are not inspiring. Apple are doing some good work in UI design, but it’s mostly just polishing the edges and in some cases, they are just coming up with fancy workarounds for deeper problems (expose, time machine, etc).

Progress seems slow. The Big Problems that we’ve known about for years don’t get solved. There are 3 big problems that interest me:

• Usability
• Security
• Environmental impact

I’ll only talk about usability today.

Usability

Luis Villa articulated one side to this long before I really understood it, which just goes to show that he’s smarter than me and in about 5 years I’ll probably wish I’d gone to law school. Some things are easy to do with web apps, and hard or impossible to do with client side apps. Of course the opposite is also true, but in general, when an app can be implemented as a web app, it usually is. This says to me that most people prefer it if they can get it.

Luis says:

Three big things:

• Web development/deployment is easy, and desktop is not.
• Web development makes certain things easy; primarily location independence and collaboration.
• Desktop development has advantages web devel does not (rich inter-app integration, localized search, etc.) but taking advantage of them is a PITA.

Luis is thinking from a developers perspective so it’s fair enough that he doesn’t mention the user experience in all this. But what I’ve realised in the last 4 months is that a well written web app gives a significantly better user experience than the equivalent client app. This is the opposite of conventional wisdom, which is that web apps are “less rich” and therefore somehow worse.

The way web apps beat client side apps works like this:

1. Firstly, people love speed. I knew this in a vague sort of way before, but now I can quantify it. Every millisecond you can shave off an operation will increase your userbase. Most good clientside developers know this, but web apps let you actually watch traffic come and go as latency changes. The correlation is remarkable and it really focusses the mind on making things happen NOW.

2. Secondly, less is more. Web apps make it hard to pull off stuff we take for granted in the desktop world – customisable toolbars, tree views, file dialogs, tab widgets etc. This is a good thing.

   The “richness” that we (programmers) love, is despised by many people. They’ll never tell you that if you ask, because they won’t know how to express it, or even what’s wrong. Instead the people who do understand it and have learned all its quirks and odd little features will tell you how important it is. This is really misleading. Don’t be fooled.

   A classic example of this is tree view controls. Programmers love trees. We think about them all the time, we see them everywhere we go. To us a tree is the most natural thing in the world. Because we love trees so much, it’s an obvious step to represent them in the user interface in some way. Like, say, by using a tree view widget. Some people love tree widgets so much they spent weeks recreating them in DHTML and JavaScript. Big mistake! Usability studies and real world experience indicates time and time again that lots of people have difficulty with trees. Some people will just avoid using them completely. Most will maybe use the top level of the tree, but won’t go in for nesting (ie they’ll have Documents/Foo and Documents/Bar but not Documents/Foo/Bar).

   Because trees are such a PITA to recreate using HTML, most people don’t bother with them in their web apps. They find some other way of doing it. GMail uses labels, Flickr uses tag clouds, the web itself uses search. Most web apps just somehow avoid it entirely. All those users who never really cared much for tree views but couldn’t really identify why are suddenly happier. They can’t explain it but somehow they know this new thing is better than the old thing.

   This follows for quite a few other kinds of client side UI construct that are now avoided because they are known to be bad. MDI, multi-level tab widgets, and incredibly complicated pref panes for instance. Once upon a time, MDI and multi-level tabs was the height of cool. Everybody wanted them. Nowadays they are avoided like the plague. How many other widgets that we take for granted will follow this fate? The humble toolbar, perhaps, which is cursed with ugly and small buttons identified only by an obscure picture? Is the “less is more” web solution of hyperlinks and image buttons sized in proportion to their importance going to replace them? I hope so.

3. Luis touches on deployment but doesn’t discuss it further. Havoc Pennington also talks about it in the context of well behaved desktop apps. One of the good yet annoying things about the Mac is that many apps now ship with Sparkle and auto-update themselves. It’s a slick little library and works well, but unfortunately it shows release notes before doing the update and generally requires user interaction/thought. Once all your apps use this, you quickly realise that (a) every app has bazillions of small releases and (b) every developer lists every change in excruciating detail.

   Web apps don’t install. They don’t update. They just appear when you need them and magically get better over time. In a world where people don’t even want to wait 2 seconds for something to appear, this is a HUGE deal! Software installation and update is, even in the best cases, slow fragile and annoying. Better to dispense with it entirely (as ZeroInstall tries to do).

4. Often, web apps have more information, so can do a better job. For instance GMails spam filtering can take advantage of knowing who is in your address book, whether a message is part of an existing thread you took part in, and so on (disclaimer: i am not saying it actually does this, just that it could do theoretically). The traditional server/client split prevents this setup, reducing the quality of the spam filtering for everyone.
5. Final point. There is no cult of consistency in web app development. You are free to create something as ugly or beautiful as you are able. Users really respond to looks, all companies know the importance of appearance and brand but traditional client side development makes it hard to achieve this. Apple, Microsoft and many other companies therefore waste millions of dollars whilst each app creates their own widget toolkit to get a distinctive look. The web recognises the importance of style from the get-go.

There are many things web apps cannot do, but perhaps to bring about the fundamental improvements I talked about in a previous entry the right solution is to extend the web (as opposed to Luis’ approach of making a client side platform that has the same advantages as the web).