I’ve covered common thoughts about eliminating programmers by making their jobs easy enough for anyone to do in part 1. Now we need to tackle the question of whether computers will learn to program and eliminate programming jobs.

We Don’t Need Computers That Can Program

First you should know that we already have software that can write code. Today programmers write specifications in languages that they prefer and then computers write the instructions that they will execute later. That’s called compiling. We have compilers that can watch how code is being used as it runs and rewrite it on the fly to perform better. We also have programs that can write human-readable code (this is called code generation), but that doesn’t tend to be very useful – it just saves a little typing.

For the foreseeable future humans will need to tell computers what to do. Telling computers how to do something every time we want them to complete a task is tedious. When programming, we only need to explain how to do something once.

Let’s talk through an example. Your boss says, “I have a file with everyone’s birthday in it. I want you to make a program that sends everyone in the company an email on their birthday.”

By the end of the day you have the program ready for your boss and it looks like this:

def send_birthday_greetings(filename)
  parse_file(filename).each do |person|
    if person[:birthday] == Date.today()
      send_email(person[:email], "Happy Birthday!")
    end
  end
end

This code loads a file, looks at each person in the file, and, if their birthday is today, sends them a message that says “Happy Birthday!”. We don’t need to write that program every day, we can configure it to run daily.

The next day your boss comes to you and says she wants a program that will send a greeting on everyone’s hire anniversary. At this point, there is no need to write another program, you can modify the previous one.

def send_anniversery_greetings(filename, attribute, message)
  parse_file(filename).each do |person|
    if person[attribute] == Date.today()
      send_email(person[:email], message)
    end
  end
end

This time the code takes a file, the name of an attribute in that file that represents an anniversary, and the message to send to users.

We just made our program more abstract and flexible. We can even create a user interface for our boss so that if she wants to send out a message on everyone’s pet’s birthday she can do that without even talking to us.

If our job was to send birthday greetings, we’re unemployed. If our job was to write programs that send messages on certain days, we’ve programmed our job out of existence.

As programmers we avoid writing repetitive code or solving the same problem twice. We take repetitive tasks and make it possible to do them over and over at nearly zero cost.

Beyond eliminating repetition, programmers work to eliminate “incidental complexity” or the work that is not essential to the problem at hand. Committees of programmers create standards and enhance platforms so that we can stop dealing with trivial technical problems and focus on the fundamental complexity of what we’re working on.

For instance, front-end developers used to spend hours creating images to produce rounded corners on the web. Now this can be accomplished in with one line of CSS code. When programmers recognize problems that are solved over and over, they create reusable libraries and frameworks that can radically improve the speed and reliability of software construction. Programmers are so fanatical about improving the state of their work that they often join world-wide communities of volunteers to build and maintain software that eliminates programming work.

If programming keeps getting “easier” then why isn’t there less demand for software developers? Because as creating software gets more efficient, we find new uses for it and tackle new problems that were too hard and too expensive to do before. Now we have autonomous vehicles and too many online social networks.

If Computers Could Program

So the nature of programming is that it eliminates repetitive work. Let’s focus back on the question: Can we teach a computer to program? Can a computer take some very basic input from a human and create a new program as a human programmer would?

Here’s our thought experiment: Tom wants to start a business selling books, so he goes to a computer and says, “build a program to sell books.” The computer recognizes this command and, with the precision of a team of master programmers, creates and deploys a new book-selling application in one second. Tom logs into his new online store. There he needs to fill out some information. He needs his business’s bank information. He needs some way to tell the application what books is he selling and whether they are physical or can be downloaded. What shipping providers is he going to use?

This is getting involved, but this isn’t programming, right? This is just using and configuring a program.

Even if we could teach a computer to write this program it wouldn’t matter. Having computers write programs that solve problems with existing solutions isn’t useful. If someone came to me today and said they want to sell books on the web (and that is all the information they gave me) I would go online, find an existing program, and use that. Again, this is the nature of programming – after we write a program we can run it and copy it as many times as we want at very little cost.

The only point where a “programming computer” becomes useful is when it requires very little human thought and input to work. In other words, the computer would need Artificial Intelligence fill in all the logical gaps to create the program. Tom says, “make me money” and the computer decides that a book store would be profitable. It gathers information on book buying habits, creates a store, sets up a bank account. The computer writes 1,000 teen vampire books and decides to sell them exclusively on the Kindle. It brokers a deal with Amazon for this exclusivity and gets a bigger share of the profits. The store nets over $1,000,000 in the first six months of operation.

At this point we’re talking about sentient and omniscient computers that can create new solutions to problems that humans have not thought of. If we ever get here, a decline in programming jobs will be the last thing on our minds.

Programming is Not a Good Candidate for Cheap Outsourcing

So if humans are going to continue programming for now, why can’t we ship this work to low-wage workers overseas as my teacher suggested in 1998? First, adding programmers to a software project has rapidly diminishing returns. Second, for a software project to succeed, the ideas that motivate it and the ideas in the program must be closely aligned.

For most software projects I would not expect, for instance, a team of 1000 programmers to produce software any better or faster than a team of 5 programmers. Having more developers means that companies can work on more projects, but it does not allow them to finish any one project faster. Furthermore, having 1000 less experienced and less skilled programmers on a project will never compare to 5 experienced and skilled programmers. The ideal programming team is small, motivated, and competent.

The bottom line is, no large organization is ever going to achieve the ideal of a handful of talented, like-minded people communicating well. (53:03)

Small groups of focused, really smart people can do great things in short periods of time – things that larger groups of less motivated, less experienced, less talented people can never do.

It’s not like, ok it takes these 5 people in room a year to do this but if I have 300 people and 3 years, I could equal them. No, will never equal them with 300 people if you don’t have the right five… (58:00)

– John Siracusa on Accidental Tech Podcast #55

My experience may be skewed since, as a consultant working in the US, if I’m working with client that tried to build something cheaply with an offshore team they probably have been burned. Still I’ll present some anecdotal evidence to give an idea of how bad things can get for these companies.

In one case, a friend of mine was working with an offshore team based in India to create a software product that gave paying users access to content. After spending a year and more than $300,000 on the product, they had to abandon it. This was a devastating blow to this small company and they had to give up on their growth strategy and change their business model to survive. My friend later described the product to me and I told him that all of the functionality could be found in a free product called Wordpress. He installed it in a weekend, picked out a theme, and added the company logo. They successfully use that product today, but the company is still recovering from the financial impact of the failed project.

I had another friend whose company invested $80,000 in a product that was lead by a US-based consultancy and outsourced to cheap overseas workers. The product failed to work consistently (due to bizarre architecture choices and unmaintainable programming practices). The problem that the program addressed was trivial and the incumbent team created a reliable solution in a couple days.

Why We Still Have Local Developers

Over and over I’ve seen large teams of cheap programmers fail to deliver what small, motivated, and competent teams of three to five programmers are capable of. There are teams that can produce software in a matter of weeks that another team could never create.

This phenomenon can’t be explained by looking at the intelligence, skill, or intrinsic motivations of the programmers involved. I’ve been on projects where I was floored by what the team delivered and I’ve been on projects where the programmers had no way to help a project succeed.

When companies work with cheap, offshore labor they create a situation where it is impossible for programmers to succeed. These companies expect to give very little input about the problem they are solving and believe that the sheer number of programmers they have at their disposal will produce results. The relationship between the client and the company doing the programming work becomes adversarial as they argue about exactly what the original requirements said and whether those requirements were technically delivered.

More programmers is rarely better because as team sizes increase the number of possible communication connections increases exponentially. A 3 person team has 3 connections, a 5 person team has 10 connections, and a 15 person team has 105 connections. Think 15 person software teams are rare? Keep in mind that I’m not talking about the programmers on a team, I’m talking about everyone that participates in the process of building software: Executives, Project Managers, Programmers, Designers, Marketers, Copywriters and Quality Assurance engineers.

Given that programming is the practice of synthesizing ideas into solution that can be reperformed by a computer, any distance, noise, and “game of telephone” effects between the problem and the solution are devastating. When a competent programmer knows the full scope of a problem, nothing can rival their productivity. When the knowledge of a problem is overseas and several employment hierarchies away from a programmer, a solution may be unattainable.

I’ve leaned on the word “cheap” to describe the type of outsourcing that was foretold in 1998, but low-cost is not what makes these teams ineffective. The fundamental problem lies in the relationship that hiring companies create with these providers: a relationship where the company’s employees dream of a product and then expect programmers to translate these dreams into “code”. In this relationship, the ideas are valuable and “coding” is a commoditized service.

I’ve worked with extremely effective offshore teams, but these teams were selected for their expertise, not just for their price. These programmers were integrated into the development process as much as any employee and were involved in designing the product, not just coding it into existence based on a set of requirements.

Successful products are built in partnership with developers with the understanding that ideas are cheap and execution is critical.