Endpoint Specification, or Good Help is Hard to Find

If you want to convey the coordinates [-1.0219,18.7172,.3181,182132.2290], what is the minimum amount of information you must convey, in order to be certain that the recipient of the information understands it?

Well, supposing there is a monument there, you could give somebody the name of the monument, and the time. However, maybe the monument shares a name with a local cafe which is more familiar to them, and they show up there instead.

You specify that it is the monument by that name, aware of that potential risk. They show up an hour late—when you gave them the time, you gave them their local time, and they helpfully converted it to the timezone where they were going.

You specify the timezone, and they’re still an hour late; you didn’t realize they were very specific about Eastern Standard Time versus Eastern Daylight Time, and interpreted your statement incorrectly.

You’re very particular about when they show up, but when they get there, they haven’t brought the zoom lens you wanted for the photo shoot.

What, did I forget to mention the photo shoot?


In a sense, the problem here is miscommunication. However, that is merely labeling the problem; I have a personal label for the problem here as well, which makes sense internally, and will make no sense to you, because the problem here isn’t actually communication: The problem here is that the minimum amount of information to fully specify exactly what you want is, effectively, you.

Alternatively, the minimum amount of information to specify exactly what you want, is the thing you want, in which case you already have it; see programming.

If you’ve ever worked with a client, you will be familiar with the experience of doing exactly what the client asked for, completing the work, showing it to them—and discovering that “what they asked for” does not actually match what they want. From your perspective, they asked for the wrong thing; from their perspective, they asked you to show up at Eastern Time, and you showed up an hour late; they asked for the correct thing—you just didn’t hear the correct thing. And this bit is blue, and clearly it should be red, why did you make it blue? They didn’t specify? Well, why would they need to specify that it was the version of the time zone everybody in the state is already using?

The people who are really successful at dealing with clients—and by really successful, I also mean really expensive—do not do what they are asked to do, but instead do what the client wants. Which is the same thing, but it is not.

Consider a bureaucrat. You may be familiar with the term “malicious compliance”—it is the term we use when somebody rebels against a system by following the rules of the system to the letter. Observe, for a second, that you recognize a divergence between what they “should” be doing, and what they are “actually” doing—and what they “should” be doing is not, in fact, what the rules say they should be doing. Otherwise we wouldn’t have a term like “malicious compliance” to describe the act of doing what the rules say instead of what they obviously should be doing instead.

How much information would it require to specify the endpoint for the average bureaucrat? Obviously far more information than is contained in the rules, which, if we think about it, are more like guideposts, or lighthouses, than either paths or destinations.

I would suggest that in practice, the landscape that the rules help guide us through is, in a fundamental sense, fractal; you can increase the specificity of the rules infinitely without correctly specifying what it is you actually want. Or, to put that differently—every exception necessarily has at least one exception. Or so close to that as to make no practical difference.

Suppose, for a moment, that you manage a bureaucrat; you’ve laid out some rules that dictate the basics of the process you want them to follow. Now, what do you actually want them to do? Well, obviously, that depends on what you want to accomplish. And here we return to the idea that the minimum complete specification for what you want is you.

Now, complete is actually better than we need; in practice, good enough is good enough. However, even that requires that the bureaucrat is able to, within “good enough” margins, guess at what it is you actually want, in order to do it.

It’s interesting how valuable it is, as we consider this, to be predictable. The lowest common denominator of consumption is the most predictable form of consumption; it’s also the cheapest. Weirdness is expensive.


You don’t employ the bureaucrat, however, not like you employed the photographer. You’re both employed by somebody else; you are merely managing the bureaucrat. Your manager isn’t your employer either. They take some set of explicit instructions, guess at what is actually wanted, and write up some instructions for you. You take the explicit instructions, guess at what your manager actually wants, and write up some instructions for the bureaucrat. The bureaucrat takes those instructions, guesses what you really want, and gets to work.

It seems like each stage introduces some level of error, doesn’t it? That’s the intuitive reaction as you consider what is going on here? That each participant in the game of telephone amplifies the error?

Consider, however, an actual game of telephone. Let’s start with two people. Which two people?

Well, suppose we have the business owner and a programmer. The owner asks the programmer to create a website to sell widgets, and the programmer creates a website that sells widgets, and guessing that the owner wants to be able to sell other things later, makes it able to do this. The owner gets upset and fires the programmer; this document cannot possibly be long enough to describe everything that is wrong with the website; notably, the programmer did guess correctly that other things would be sold.

Let’s add a manager to our game of telephone. The owner requests a website to sell widgets. The manager guesses that the owner actually wants a company website, of which selling widgets is an important feature. This information is passed along, and the programmer correctly guesses that other things would be sold. We end up with something passable but the owner is still unhappy.

I can theoretically add additional people to the chain; I don’t think it’s necessary. Now, this example has an element which I think requires attention; in particular, the manager is making what is indistinguishable from a guess at what -any- owner would want in a website. It is worth paying attention to this, because while there is some value in that kind of guess, a good manager isn’t going to just guess at what any business owner wants in a website, but what this specific business owner wants in a website.

Given finite amount of work available from the programmer in a given time frame, the value of the manager to the business owner is, when you get down to it, going to be in that manager’s ability to prioritize the use of the programmer’s time to get the features most valuable to the business owner. Which means it isn’t just the ability to guess what you want, but also to guess what you don’t want; if it was merely adding features you might want, we could just keep adding people to the game of telephone, and compile a massive list of features.

Wait, you ask, couldn’t a programmer do all this as well, and predict what the owner wants? The answer is, of course, that yes, a programmer could do this.

However, an observation: The ability to correctly guess what somebody else wants is, in fact, valuable; the better somebody is at it, the more they will be in demand. (Also implying that the more predictable the desires, the less expensive it will be to hire somebody to fulfill them.)

Programmers, of course, are also in demand, and valuable. Value in this case compounds. If you only need one employee, the programmer who can guess what you want might be the employee for you. However, as you add employees, it becomes more and more cost effective to hire one person who can guess what you want, to pass that information down to other people.

Alternatively, of course, you could just specify exactly what you want in the first place, so there’s no ambiguity. However, if you’re doing that, you’ve already done the job you would have hired the programmer to do in the first place.


Let’s consider the value of being able to guess what somebody else wants. First, we may observe that one can be better or worse at this. Second, we may observe that one can be better or worse at this with respect to specific individuals; there exists somebody who is better at me at guessing what people want in general, who is worse specifically at guessing what my wife wants. Third, we may observe that one can be better or worse at this with respect to groups of individuals; a devout Catholic may have some advantage at guessing what other devout Catholics want.

So a cluster of things can come together to create a price point for hiring somebody at a given level of proficiency at guessing what is wanted. Insofar as cultural effects matter, a dominant culture may get what effectively amounts to a discount on this skill; certainly it seems like the level of this skill necessary for a highly homogeneous organization to effectively coordinate will be lower.

These things matter, and will have effects on society. Also, as society gets more complex, it’s probable this skill is increasing in demand. I would highlight that, but I’m not going to, so I’ll just repeat it in different words: As society gets more complex, jobs increasingly require greater degrees of the ability to accurately guess what other people want. Even low-level jobs. This is important, but I’m going to move on, because it isn’t what this article is about.

Personally, I don’t just think of this as “guessing what somebody else wants”, because, while accurate in a certain sense, it also isn’t a complete description. There’s a secondary motivational component; being able to be motivated to achieve what somebody else wants.

Imagine the low-level manager trying to get the employees to care about merchandising displays.

I call this second set of things “caring”, because that is the word we use to describe the behaviors associated with it; it’s also how I think about this kind of thing as a whole, in that it is expensive to hire somebody who cares.

I will forward an idea: There are businesses you can start which, if you care and have a basic level of competency and actually show up and run your business, will not fail, and will be quite profitable to you. They operate in weird economic vacuums in which the ubiquitous corporate presence is mysteriously absent, or, if present, mostly serving as a facade of respectability, and predictable customer experience, for an owner-operator.

You can probably think of some examples. I’ll pick one.

Why is it that you can go and start, say, a laundromat, and start making money? Seriously—why haven’t corporations taken all these opportunities and made that money for themselves? Why are there any such economic niches to fill at all in the world? And why aren’t corporations going around buying up all the independent profitable businesses once established? Opportunity cost is of course part of the answer, but it isn’t the whole answer.

Hiring somebody who cares is expensive, and you need somebody who cares to run these businesses.

It’s easy to guess one thing the corporation wants out of the laundromat. They want to make profit. That, however, is not the end of what they want: They also want to avoid PR or legal problems (a proper discussion of which would make this article unbearably long, but treating PR and law as specialized forms of “guessing what other people want” is only slightly wrong). Whoever they hire is a potential source of PR or legal problems; they don’t just need somebody who cares about what they care about, but somebody who legibly does so. And somebody needs to care about making sure those people actually care, and so on and so forth.


Returning to the question of “If you want to convey the coordinates [-1.0219,18.7172,.3181,182132.2290], what is the minimum amount of information you must convey, in order to be certain that the recipient of the information understands it?”, some observations. Somebody might notice the absence of a reference coordinate. Somebody else might notice that the coordinate system is unspecified. Somebody else might notice that the universe itself is unspecified. I’m not sure it is even possibly to specify the minimum amount of information you would need to convey; personally I’m not certain the number of dimensions is finite, so we might need to specify an infinite number of additional coordinates in additional dimensions. (Heck, how much of the language used to specify that needs to be conveyed, as well?) The question is, basically, nonsense.

How the heck can anybody even guess what anybody else wants, given that the space of wrong answers is infinite? Well, fortunately, we’re pretty convergent. Being predictable is valuable, and human beings actually practice being predictable (when you get down to it, that’s what language is, the practice of predictability; I could guess what you’re thinking, but it’s better if what you are thinking is more predictable). That’s its own topic. The point is, we can. Some people better than others.

Caring is also a thing; on its own, useful, like the low-level manager who gets the merchandising done right. And combined with accurately predicting what other people want, it is quite powerful, at least for other people’s purposes; and quite valuable, for our own.

There’s no major conclusion here; this is just an attempt at conveying a particular perspective. I could say something about how these two things tend to be denigrated and despised, I guess, so let’s go with that. Hey, it’s valuable to be able to care about stupid crap. It’s valuable to be able to figure out what somebody else wants you to do without them having to say. It’s valuable to be predictable.

Here’s to middle management, and everything you represent; you got something important right.