In 2017, the product team at Airbnb built a proof of concept for a new prototyping tool, one in which a user interface could be built simply by sketching on paper. If you watch the accompanying video, you’ll see that the designer draws some rough shapes on a piece of paper, and then slides that paper under a camera. From there, a computer analyzes the picture, identifies the shapes, and associates them with specific UI components from Airbnb’s design system. Equipped with that information, the computer renders a finished, web-ready prototype in the browser.

When Airbnb shared this prototype, the team said that their work was shaped by this guiding principle: 

The time required to test an idea should be zero.

There’s a lot that’s appealing about this idea. As a mission statement, it’s a fine one. But it’s worth remembering that people—you, me, all of us—live and work in the hours, minutes, and seconds we’re trying to streamline out of our product cycles. And our industry has a terrible track record of asking about the second-order effects of its product decisions. For example, if we could instantly create designs in a browser, what happens next? What jobs would be changed as a result? Who would be impacted? 

In the years since Airbnb’s experiment, it seems we’ve been asking those questions more and more often. In fact, there’s been an explosion of “generative artificial intelligence” software platforms, all of them focused on automating tasks that’ve traditionally been the domain of engineers, writers, or designers—all of them released in rapid succession since 2021. GitHub released Copilot, a utility that can parse plain-language inputs from a user, and suggest working code snippets that fulfill the desired request. The research laboratory OpenAI released ChatGPT, a tool that allows users to ask questions in plain English, and then returns natural-sounding answers. And there are AI art tools like DALL-E, Midjourney, and Stable Diffusion, which can generate stunning images in response to simple text prompts. 

These utilities are remarkable, full stop. By simply typing a few phrases into a chat interface, their users can quickly produce something that resembles code, content, or art. And frankly, I’m awed by how rapidly these tools are evolving. In 2022, AI image generators could barely render a recognizable human face; a few months later, they were producing high-resolution, photorealistic pieces. In March 2023, OpenAI promoted the release of GPT-4, the latest version of its large language model (LLM). As part of the publicity around the launch, they released a video showing GPT-4 generating a simple web page by analyzing a rough, hand-drawn sketch. These tools are evolving fast. It’s hard not to look at their current state and see today’s shortcomings as their floor. 

These utilities aren’t remotely perfect, mind you. Nor are they free from controversy. They’re able to produce such stunning results because they’ve sifted through staggering amounts of training data—of preexisting code, content, or art—often without regard for whether the original work’s license or copyright permitted it. And what they produce often has significant flaws: on its website, ChatGPT notes that it “may occasionally generate incorrect information”—and indeed, the prose it spits out often sounds realistic, but isn’t factually correct.

But putting these shortcomings aside for a minute, I have to admit I’m impressed and terrified by these tools in equal measure. As a designer, I can’t help but wonder what they mean for the nature of my work. And maybe you’re curious, too: after all, when everyone has access to tools like these, what does it mean to be a “designer,” an “engineer,” or a “writer”? 

These tools are inspiring. From an engineering perspective, they’re brilliant; on a personal level, I’ve always dreamed of something like this, where software could instantly translate my ideas into a finished, polished product. But they exist in an industry that’s deeply invested in the idea that, as the Airbnb team said, the time required to test an idea should be zero—and that the number of people involved should be close to zero, too. 

I’m reminded of a blog post by Dave Rupert, in which he shared some early impressions of Copilot, GitHub’s code generation tool. He wrote about how its code suggestions often worked, but weren’t always what he was looking for. However, the net effect was that the software changed how he approached his work: 

My biggest adjustment with using Copilot was that instead of writing code, my posture shifted to reviewing code. Rather than a free form solo-coding session I was now in a pair-programming session with myself (ironically) in the copilot seat reviewing. […] The end goal of programming is working software and the robot can suggest code faster than I can write code.

According to Rupert, the primary responsibility for creating code shifted away from him and to the software. As a result, he found himself adopting more of a supervisory role, spending his time reviewing the code that the tool produced. (And occasionally, rewriting it.) 

There’s a term for this in economics: deskilling. As a new technology is introduced to an industry, it may be able to complete tasks that had traditionally been performed by human workers. And as the technology matures, it can be overseen by fewer workers, which gradually lowers wages and may eliminate jobs. In other words, the demand for skilled labor is gradually reduced—or eliminated altogether. Once the technology reaches a certain critical mass, workers are moved into roles they would traditionally have been seen as overqualified for, while less skilled workers are driven from the market altogether. Every worker is, quite literally, de-skilled by automation. 

As tech workers, we’re far from the first industry to see the nature of our work change, or so quickly. I spoke with Keith Hogarty, a longtime union organizer with the Communications Workers of America. Hogarty has been organizing for twenty years, helping workers from various industries form unions. And he’s worked in the telecommunications industry for decades. 

During our chat, Hogarty told me how he managed to get work installing residential networks during the internet’s early days. The work was, he remembered fondly, really good. “My job was a cushy job,” he remembered. “We didn’t have enough people to do the work: to install jacks for people who needed internet in their houses. Then,” he added, “the smartphone came out. Pretty quickly, my cushy job wasn’t cushy anymore. I was in a fight for survival.” 

I think about Hogarty’s story often. Over the span of a few short years, the easy work of installing fiber-optic networks— working indoors, and getting paid well to do it—quickly became scarce, thanks to the deployment of cellular networks. The easy work dries up; the value of your labor becomes lower over time. 

Technically speaking, software alone can’t displace workers—rather, it takes time, effort, and concerted investment from those who want to devalue our labor. None of this happens by accident. A truly staggering amount of capital has been invested in automation software, much of it on the heels of mass layoffs across the tech industry. But as significant as this moment is, we have an equally powerful response: we must unionize our workplaces.

Excerpted from You Deserve a Tech Union—on sale August 15, 2023 from A Book Apart