Stop Choosing Sides on AI

Steven Levy, the longtime editor of Wired, edited one of my earliest articles. In it, I explored historical moral panics that came with emerging technologies: the radio would destroy intelligence as fewer people would read newspapers; the telephone would destroy personal relationships as people wouldn't spend time together in person; social media would further push us into isolation and destroy meaningful connections.

We're facing a similar moral panic today surrounding artificial intelligence. The fear stems from the near-limitless power and revenue of tech companies at the expense of stolen work and potential job losses. What I failed to address in that original article nearly 15 years ago is that both sides of the argument can be true.

• The radio did lead to a decline in reading newspapers while also opening up fresh forms of sharing news and entertainment.
• The telephone did lead to teenagers spending more time talking on the phone than in person, while also connecting people across vast distances in a previously cost-prohibitive way.
• Social media did lead to fewer in-person connections and the rise of the loneliness epidemic, while also providing expanded access to people and information.
• Likewise, AI uses stolen data and positively reframes workflows and productivity in previously unimaginable ways.

These are all accurate statements, and none of them are mutually exclusive. Yet, too much of the conversation around AI is that both cannot be true: either AI is bad and no one should use it, or AI is good and everyone should use it. This binary is the wrong framing and misses the nuances surrounding AI — after all, AI is not an all-or-nothing choice.

Today, I want to look at the three biggest problems I see with the current AI discourse:

1. The inability to hold opposing viewpoints simultaneously,
2. The US-centric, fear-based thinking, and
3. The threat of job losses.

I.

The first problem with AI discourse is the refusal to acknowledge that people can hold multiple conflicting viewpoints.

Saying I use AI to improve my workflow immediately labels me. I got some replies to my last article built solely upon this incorrect assumption: by failing to give up my use of AI because of the mass theft of writing, art, and music, I am complicit in the theft and, in one case, mentally unwell. As a result, I'm also labeled as wholly relying on AI to generate everything I publish.

People are multifaceted and can hold opposing views. Yet, instead of discussing these contradictions openly from a place of mutual respect, we often make assumptions: this person is not like me and is therefore on the wrong side of the argument.

Yes, I know AI was trained on stolen data — including my writing — and I've been dealing with that tension for years. At one point, I tried to block AI bots from the site and added a login requirement. But that action went against one of my guiding principles about the internet: information should be free and accessible to all.

After over a year of experimentation, I removed the login requirement and decided that fighting with AI was a losing battle. I don't like that AI models are built with wholesale copies of unlicensed data and hope that the courts respond accordingly. But what I guarantee will not happen is some kind of complete erasure of AI tools. Like it or not, Pandora is out of the box.

Even if a company like OpenAI were shut down by a class-action legal ruling, dozens of open-source models would continue to rise. They'd develop at a slower pace and with less funding, but they would still exist. Like the radio, telephone, and social media before it, AI is here to stay.

Instead of fighting a completely unwinnable fight, I choose to learn how to make the technology work for me. AI does not write for me or generate articles whole cloth. Full stop. It does help me edit, polish, and find connections with other work. AI also automates parts of my workflow that were previously tedious or expensive, completely replacing tools like Zapier.

I wrote the first draft of this article on my phone while mid-flight. In the airport before takeoff, I got into a Mastodon conversation with Elizabeth Tai, a tech writer and former journalist from Malaysia, about this exact topic. She and I share similar views on the subject, and she shared some interesting data we'll get into in the next section.

But as is wont to happen online, people who don't hold dualistic views jumped into the conversation to point out just how wrong we were. It turned a social media conversation (with people on opposite sides of the globe, no less) into an argument that sparked me to write this article.

Before boarding, I connected to the version of Claude Code running on my home computer and asked it to grab the full Mastodon conversation. It also pulled copies of the studies that were mentioned and compiled them into a brief. This entire process took a few minutes, and I was prepared to review everything on the plane.

This workflow was not possible a few months ago. Now AI is supporting my work streams in ways I couldn't have previously imagined. Does this negate the fact that AI models were trained on stolen data and could represent a complete reframing of the workforce? Not in the slightest. But I can hold opposing views at the same time.

II.

The second problem with the AI discourse is that so much of it focuses on US-centric, fear-based thinking.

In our Mastodon conversation, Elizabeth shared that her home of Malaysia is very accepting of AI tools. Data from the Malay Mail shows that 9 in 10 Malaysian workers use AI in their job, compared to only 46% of Americans. Additionally, 90% of Malaysian university students use AI, up from under half in 2023.

Open source models like DeepSeek and Qwen, and tools like Hermes and Open Claw are big accelerators for countries like Malaysia, according to Elizabeth. They make AI far more affordable and accessible, which helps with adoption rates.

Elizabeth concedes Malaysians might not consider the downsides of AI usage enough, but adopt the technology to stay competitive in the global market. Looking at the AI discourse as simply US-centric ignores the broader picture of higher global adoption.

Microsoft's AI Economy Institute published a study that used aggregated, anonymized data from Microsoft products (yes, they are always watching, and that's a topic for another time). What they found is pretty insightful and supports the Malay Mail data:

• 16.3% of the world's working-age population uses generative AI tools, up from 15.1% in the first half of 2025.
• The United States ranks 24th, at 28.3% adoption, despite leading global AI infrastructure investment.
• DeepSeek's open-source (free) model drove outsized adoption in China, Russia, Africa, and other markets with limited access to Western tools — 2 to 4 times higher usage in Africa compared to established platforms.
• The UAE has 67% public trust in AI versus 32% in the US — a 35-point gap that maps almost directly onto their adoption gap.

Elizabeth asked me why I think the US has been slower to adopt and more vocally anti-AI than other parts of the world. I believe that the adoption gap here in the US is fear-based. Our media, politicians, and the foreign bot farms that prop both up have spent decades teaching us to be afraid of things we don't know or don't understand.

New technologies like the radio, telephone, and social media? Be afraid of them. People with beliefs or views (or from places) other than your own? Be very afraid of them.

This fear-based mentality has pervaded every corner of American society. Fear is just as clear in politics as it is in conversations about AI.

When people feel uncertain about their world — or their place in it — they rarely sit with that discomfort. According to uncertainty-identity theory, they gravitate toward groups and ideologies that offer simple answers. The more uncertain someone feels, the more attractive those rigid, us-versus-them frameworks become.

The research shows that uncertainty makes people more likely to support extreme positions. Decades of manufactured fear have made people tribal.

People who fear what they don't know or don't understand are much more susceptible to adversarial thinking — and that susceptibility is exactly what gets exploited, whether the topic is immigration, democracy, or artificial intelligence.

The problem, however, is that when Western viewpoints largely dominate the discourse, it feels like everyone is anti-AI, and that just isn't reality.

Stanford ran a study on AI public opinion that confirmed these details:

• AI optimism is rising globally. The share saying AI offers more benefits than drawbacks rose from 55% (2024) to 59% (2025), even as 52% say AI products make them nervous — two things can be true at the same time.
• 73% of AI experts expect AI to have a positive impact on how people do their jobs; only 23% of the US public agrees.
• 64% of Americans expect AI to lead to fewer jobs over the next 20 years; only 5% expect more.
• The US has the lowest government trust for AI regulation of any country surveyed — 31%, versus a global average of 54%. Singapore leads at 81%, Indonesia at 76%.

I posit that the anti-AI fear pervading American culture right now results from distrust in a government that allied with the tech billionaires. Again, this fear-based thinking may be well-founded for AI (we can hold multiple views at once), but that shouldn't prevent us from learning about the technology or what the rest of the world thinks.

Case in point, in the same week the White House declared the US didn't need a national framework for AI, a Hangzhou court ruled that replacing employees solely to reduce costs through AI is not legally permissible under Chinese labor law.

Presenting a US-centric view as the only view on the subject is the wrong approach with AI, just as it's the wrong approach with any subject.

III.

The third issue with the AI discourse right now is the fear of job losses. Granted, part of this aligns with fear of the unknown, but it's bigger than that — it's an inability to make accurate predictions about the future.

Depending on what you read or who you listen to, we're looking at the end of the workforce as we know it, or the biggest shift in jobs ever, or literally no change in the workforce at all.

Every technological innovation changes the economy and how people work — there’s no reason AI will be any different. Just how and when that change occurs is anyone's guess. The problem is that everyone is guessing and moving towards extreme viewpoints (maybe it has something to do with uncertainty-identity theory, or maybe it's just the attention economy).

Casey Newton at Platformer has been interviewing people in the AI space about potential job losses for three weeks now. So far, here are the summarized takes:

1. Aaron Levie, CEO of Box, says AI agents multiply on top of existing workforces rather than replacing them, creating an increase in capacity without a corresponding drop in headcount.
2. James Manyika, who examines the broader consequences of AI for Google, says about 50% of tasks are now automatable, but fewer than 10% of jobs can be automated.
3. Boris Cherny, the creator of Claude Code, says this is the end of software engineering as we know it, but the start of a 100x increase in "builders" who don't write code but create with AI.

These ideas are all over the place. Will there be zero job losses or total decimation of an entire industry? Who knows! That said, even the Pope is weighing in.

Boris acknowledged that when you follow the exponential job loss models out, things get a little "weird" — that the whole concept of jobs will change. This sounds like an acknowledgement that exponential forecasts have limits and reality will be more tapered. Either way, predicting the future is basically impossible.

That said, I do believe we'll soon see the AI bubble burst.

There's no doubt that we're experiencing a serious AI bubble — just look at the industry's strange circular funding or companies realizing that most AI automation is too expensive to be profitable.

Further, the Microsoft data shows that 65% of workers fear falling behind without AI adoption, yet only 13% receive reward or acknowledgment for their innovations. This tension points strongly to an AI bubble.

So, what happens when the bubble bursts? Again, it's anyone's guess. But, based on previous bubbles, it's likely that frantic funding and spending will die down, fringe startups will die off, and the industry will refocus on its core deliverables. This pattern occurred with the .com bubble at the end of the 90s and again with the real estate bubble at the end of the 00s.

Given the likely scenario of the AI bubble bursting, we should look at what the core deliverable of the technology actually is. "AI" defines so many things that all-or-nothing thinking is inevitable — we can't even agree what we're arguing about!

AI's core deliverable is not generating articles or videos that fill up content platforms. Granted, this is largely how companies position their products and how most people experience the technology. But generative content is clearly not the AI endgame.

OpenAI has shipped tons of generative AI tools: ChatGPT, DALL-E, Sora. Yet, it has scaled back these tools over the years, shifting its focus solely to ChatGPT as a collaborative tool. OpenAI even shut down Sora entirely.

AI is also not, I don't think, even true artificial intelligence, as many of the labs would have us believe. I don't foresee a world (at least in the near future) where we have fully autonomous robots just side by side with humans in the world. I also don't think the research will lead to some breakthrough in replicating human consciousness, not without serious breakthroughs in quantum computing power that the economics just don't support.

Instead, I think AI's core product is actually what's shipping with Claude Code and ChatGPT Codex: agentic tools to help us enhance our current workflows.

Agentic computing is a way to use AI tools like a virtual assistant or a type of employee. You tell the agent what you want it to do, and then based on the permissions you've established, it goes and does the thing without interruption.

I have agents for all sorts of things. At work, I have one that lives inside my inbox that builds expense reports from payment notifications. I have another that looks for regulatory changes that will impact my industry and alerts me via email.

These are merely surface-level examples of agentic computing. Some folks have agents that run 24/7 doing things like finding (and booking) travel deals, writing software, and more.

Now, this might sound horrific to you, and I understand that. It might also seem like travel agents and software developers are in trouble. However, at least in the current state of AI and for the foreseeable future, I don't think that's the case.

Will some companies shift jobs fully to agents? Yes. We've seen it happen with experiments in drive-through technology. At the same time, we saw how quickly the companies had to reverse that decision because agents cannot fully replace humans at scale.

So what's the point of an agent? I believe they enable individuals and small companies to move quickly and nimbly, doing things that would otherwise be cost-prohibitive or downright impossible. As Aaron shared in Platformer, an agent is about scalability.

Sure, I could manually download the receipt and fill out the expense report or routinely read governmental reports instead of having an agent do that. But by taking these tasks off my hands, I have more time to support my team's needs.

The same goes for the agent that created the Mastodon summary for this article — I could have done that work myself when I got home and back to my computer in a few days. Instead, the agent did it all for me so I could get straight to reviewing, reading the studies, and writing — all from my phone.

In the Platformer interview, Boris asks if Casey is more productive than he was before the dawn of AI agents. Casey answers he is, but he's still working the same amount of time. That time is instead used to do other things, like the entire miniseries of interviews on AI and the workforce. Agents don't replace our work; they enable us to focus on more important things.

The biggest concern for potential job loss post-bubble, as Ryan Broderick at Garbage Day writes, is (ironically) likely to occur at companies like Google and Meta that went all in on AI. Google has fundamentally restructured the search engine around AI, which will be very difficult to unwind.

I don't know if AI will cause massive job losses any more than the experts do. But I do know that agentic computing will only continue to grow, despite it being kind of boring.

An all-or-nothing stance toward AI, which involves refusing engagement because of flaws or unethical sourcing, doesn't improve AI's safety or foundations for anyone. Nor does labeling or assuming extreme positions of those who use the technology.

A better use of that energy is not arguing for abolition but understanding how to mitigate the risks — and that requires engagement, not avoidance. Not to mention mutual respect and good faith toward each other's opinions. You don’t have to love AI. You just have to stop pretending it will all magically go away.

Over a decade ago, I wrote about moral panics surrounding new technologies. I got some things right and missed others. What I know now that I didn’t then is that the technology always outlasts the panic.

My thanks to Elizabeth Tai for contributing to this article.