Once you understand how to do something, it’s not exciting anymore. It’s just a matter of doing the right thing over and over again the right way. Most computers are better at that than most humans. Many of the inventions that changed the world are based on making the boring, needful things easier and faster — grinding grain, making cloth, assembling cars, doing laundry, milking cows.
When automation is first introduced it’s always disruptive because someone had a job doing that, and now they don’t. It’s also often more expensive than the human work… for a while. But as the automation matures, it becomes more reliable and less expensive, and we switch over to it.
In some ways, it’s harder to see this in computers, because much of the displacement happened before we remember it. Almost no one currently working in software remembers manual switchboard calls, but hundreds of thousands of women once had work routing communications. “Taking a letter” is something we only see in old movies and episodes of Mad Men, and we would find it surprising if any of our managers couldn’t type themselves, at least slowly.
The way we got to Progressive Delivery in software is by breaking all the silos and rigid roles that were so natural to waterfall deployment. If there is one big release, it makes sense to have a release manager to make sure that the manifest is correct and all the checklists are properly checked. We can democratize who can do a release because we have automation backing them up, in the form of automated testing, continuous integration, sophisticated source code management, and integrated security checks. Those were all jobs, and they have been automated.
What do we automate?
When I think about automation replacing work, I have a Jetsons-like image of a robot that performs functions for me, like an anthropomorphic housekeeper in a frilly apron, or a machine that does my hair and makeup in one flash. But it really looks more like a shopping website, a QR code, a delivery driver, or a work ticket. We’re not ever replacing work, we’re redistributing it. I don’t have to shear, spin, weave, and sew for my family to have clothes. But someone does. And even the machines that spin and weave require design, construction, and maintenance. Garment manufacturing requires humans to touch each piece of clothing multiple times — we just don’t think about it much because it’s not something in our field of vision.
A very tiny percentage of “tech people” write machine language. They have created an automation chain that everyone further up the stack relies on. We don’t have to think much about how to talk to a chip, or even our operating systems. In turn, there are people further up the stack than us who don’t think about the work we do.
The thing I want you to understand about automation is that it’s not magic, it’s just a different, sometimes more-efficient, way to do the same work. When Google predicts the next couple words of a sentence you’re writing in email, it’s not generating that out of nothing, it’s basing it on the millions and millions of emails that it has processed.
Nothing comes from nothing. Automation comes from experience. So what do we automate? We automate work that falls into the following categories…
Work we value but don’t enjoy
There are lots of things that need to be done, but doing them is tedious, or physically stressful, or just not enjoyable. When we can, we offload this work to machines, like laundry and dishwashing, or telephone routing. We want it done, but we don’t want to do it.
Frequently, low-paid labor is an intermediate step in this automation. We stop paying humans for this kind of thing when we can get adequate results for a lower cost than wages. Food harvesting is an excellent example of this – there are some foods that we harvest automatically, like wheat and corn, and even oranges. There are some foods where it is still cheaper for the producer to pay humans to do it, like asparagus or apples, because the food is fragile, or hard to harvest in some other way. As wages rise, it becomes more economical and reasonable to invent and run machines that can do the work.
In a technology sense, we try to automate anything that we can describe well, because the act of description is very close to the act of automation. Setting out to build automation for something we’ve never done before is an exercise in frustration (or iteration, depending on your optimism). How can you write procedural steps for something if you haven’t discovered the steps? How can you optimize something undefined? Even if it’s work we don’t enjoy, we need to do it a few times to understand what we’re automating.
Work that needs to be very precise
Everyone cares about banking transactions being accurate, because it involves money, which is the easiest way to describe value. So banks are motivated not to lose track of even fractions of percentages, and people are motivated to be sure they keep all the money they have earned. Humans, even the most accurate humans, don’t always do things exactly right. So we have automated things like calculation and transfer, because it must be precise. Money is also easy, because unlike, say, an appendix, it has a very clear definition and form.
We also don’t hand-draw circuits anymore, we have software to help us design and etch the chips that everyone counts on every day. By applying the efforts of many people to make an automation as perfect as possible, we end up with an automation that we can treat as finished and more accurate than any single human.
Things that are complex
We automate things that are complex. This seems counterintuitive, because it’s difficult to describe complexity accurately, but it’s one of the only ways we can manage the very large abilities we’ve gained from abundance. No single human is picking out the ads you see in your social media feed. Instead, they are the result of data aggregation and analysis that identifies you as part of a demographic, and ad buyers choosing to target your demographic, and ad bidding houses that require milliseconds of response time to load the ad in your page as it’s rendering. The act of assembling the “front page” of any given news site is orders of magnitude more complicated than when I was running printed paper through a waxing machine to do manual paste-up, which in turn was so much more complicated than setting text in a movable type printer.
Complexity becomes invisible to us as it’s automated, because automation is an abstraction layer. An abstraction is a simple way to refer to something you’re trying work with. “CNN.com” is an abstraction for dozens of scripts and references that come together to create the audiovisual experience of a news website. But trying to understand what each of them does and when and how is not important to our ability to reference “the news”. And automation doesn’t really apply to the complexity of constructing and curating what counts as “news”.
Work that crosses domains
We automate work that crosses domains. Many of our inefficiencies at work and at home occur in the gaps between one action and the next. A laptop does not stay in perpetual motion between inception and us. In the factory, there might be delays in the manufacturing if any of the thousands of parts is delayed. In transit, it waits in warehouses and shipping depots until it is loaded into the next transportation mode. It sits in the IT closet or on our doorstep until we get our hands on it. At each of these domain crossings, there is a lull, an idleness. Automation is how we keep things from getting lost in those lacunae.
Progressive Delivery 