Paul Graham - Essays

Like it or not, programming languages are also for people, and I suspect the human brain is just as lumpy and idiosyncratic as the human body. Some ideas are easy for people to grasp and some aren't. For example, we seem to have a very limited capacity for dealing with detail. It's this fact that makes programing languages a good idea in the first place; if we could handle the detail, we could just program in machine language.

Remember, too, that languages are not primarily a form for finished programs, but something that programs have to be developed in. Anyone in the arts could tell you that you might want different mediums for the two situations. Marble, for example, is a nice, durable medium for finished ideas, but a hopelessly inflexible one for developing new ideas.

A program, like a proof, is a pruned version of a tree that in the past has had false starts branching off all over it. So the test of a language is not simply how clean the finished program looks in it, but how clean the path to the finished program was. A design choice that gives you elegant finished programs may not give you an elegant design process. For example, I've written a few macro-defining macros full of nested backquotes that look now like little gems, but writing them took hours of the ugliest trial and error, and frankly, I'm still not entirely sure they're correct.

We often act as if the test of a language were how good finished programs look in it. It seems so convincing when you see the same program written in two languages, and one version is much shorter. When you approach the problem from the direction of the arts, you're less likely to depend on this sort of test. You don't want to end up with a programming language like marble.

For example, it is a huge win in developing software to have an interactive toplevel, what in Lisp is called a read-eval-print loop. And when you have one this has real effects on the design of the language. It would not work well for a language where you have to declare variables before using them, for example. When you're just typing expressions into the toplevel, you want to be able to set x to some value and then start doing things to x. You don't want to have to declare the type of x first. You may dispute either of the premises, but if a language has to have a toplevel to be convenient, and mandatory type declarations are incompatible with a toplevel, then no language that makes type declarations mandatory could be convenient to program in.

In practice, to get good design you have to get close, and stay close, to your users. You have to calibrate your ideas on actual users constantly, especially in the beginning. One of the reasons Jane Austen's novels are so good is that she read them out loud to her family. That's why she never sinks into self-indulgently arty descriptions of landscapes, or pretentious philosophizing. (The philosophy's there, but it's woven into the story instead of being pasted onto it like a label.) If you open an average "literary" novel and imagine reading it out loud to your friends as something you'd written, you'll feel all too keenly what an imposition that kind of thing is upon the reader.

In the software world, this idea is known as Worse is Better. Actually, there are several ideas mixed together in the concept of Worse is Better, which is why people are still arguing about whether worse is actually better or not. But one of the main ideas in that mix is that if you're building something new, you should get a prototype in front of users as soon as possible.

The alternative approach might be called the Hail Mary strategy. Instead of getting a prototype out quickly and gradually refining it, you try to create the complete, finished, product in one long touchdown pass. As far as I know, this is a recipe for disaster. Countless startups destroyed themselves this way during the Internet bubble. I've never heard of a case where it worked.

What people outside the software world may not realize is that Worse is Better is found throughout the arts. In drawing, for example, the idea was discovered during the Renaissance. Now almost every drawing teacher will tell you that the right way to get an accurate drawing is not to work your way slowly around the contour of an object, because errors will accumulate and you'll find at the end that the lines don't meet. Instead you should draw a few quick lines in roughly the right place, and then gradually refine this initial sketch.

In most fields, prototypes have traditionally been made out of different materials. Typefaces to be cut in metal were initially designed with a brush on paper. Statues to be cast in bronze were modelled in wax. Patterns to be embroidered on tapestries were drawn on paper with ink wash. Buildings to be constructed from stone were tested on a smaller scale in wood.

What made oil paint so exciting, when it first became popular in the fifteenth century, was that you could actually make the finished work from the prototype. You could make a prelimary drawing if you wanted to, but you weren't held to it; you could work out all the details, and even make major changes, as you finished the painting.

You can do this in software too. A prototype doesn't have to be just a model; you can refine it into the finished product. I think you should always do this when you can. It lets you take advantage of new insights you have along the way. But perhaps even more important, it's good for morale.

Morale is key in design. I'm surprised people don't talk more about it. One of my first drawing teachers told me: if you're bored when you're drawing something, the drawing will look boring. For example, suppose you have to draw a building, and you decide to draw each brick individually. You can do this if you want, but if you get bored halfway through and start making the bricks mechanically instead of observing each one, the drawing will look worse than if you had merely suggested the bricks.

Building something by gradually refining a prototype is good for morale because it keeps you engaged. In software, my rule is: always have working code. If you're writing something that you'll be able to test in an hour, then you have the prospect of an immediate reward to motivate you. The same is true in the arts, and particularly in oil painting. Most painters start with a blurry sketch and gradually refine it. If you work this way, then in principle you never have to end the day with something that actually looks unfinished. Indeed, there is even a saying among painters: "A painting is never finished, you just stop working on it." This idea will be familiar to anyone who has worked on software.

Morale is another reason that it's hard to design something for an unsophisticated user. It's hard to stay interested in something you don't like yourself. To make something good, you have to be thinking, "wow, this is really great," not "what a piece of shit; those fools will love it."

Design means making things for humans. But it's not just the user who's human. The designer is human too.

Notice all this time I've been talking about "the designer." Design usually has to be under the control of a single person to be any good. And yet it seems to be possible for several people to collaborate on a research project. This seems to me one of the most interesting differences between research and design.

There have been famous instances of collaboration in the arts, but most of them seem to have been cases of molecular bonding rather than nuclear fusion. In an opera it's common for one person to write the libretto and another to write the music. And during the Renaissance, journeymen from northern Europe were often employed to do the landscapes in the backgrounds of Italian paintings. But these aren't true collaborations. They're more like examples of Robert Frost's "good fences make good neighbors." You can stick instances of good design together, but within each individual project, one person has to be in control.