A six-month old paper called “AI as Normal Technology” is having a resurgence moment this week, likely due to the recent conversations about whether AI has real measurable value.
AI as Normal Technology (by Arvind Narayanan and Sayash Kapoor) argues that while AI will profoundly transform society like electricity and the internet did, this will take decades, not months. This paper has the rare distinction of being dunked on by both AI accelerationists (who think it undersells the exponential progress) and AI skeptics (who still think it gives AI too much credit).
I have a third viewpoint: if AI really is “normal technology” that takes years to fully integrate into organizations, then having a solid foundation of stable infrastructure—while “boring” by AI hype standards—is actually your best strategic play.
In my past seven months at Citrix, I’ve had hundreds of conversations with customers, partners, colleagues, and industry folks about the future of work, enterprise IT, and the impact of AI. I can say with 100% certainty that most online conversations about AI in the enterprise do not represent the actual conversations happening in real life.
So let’s look at the conversations that are actually happening, which of course depends on who’s doing the talking.
CISOs are terrified, not of missing out, but that they’re going to be in the news. Their nightmare is the headline, “YOUR company suffers breach after employee feeds data to ChatGPT.” In this context, a CISOs job is to say no, lock things down, and prevent a disaster. Traditional DLP breaks down when AI has perfect memory. You can’t rely on simple regex rules to determine what’s sensitive to an AI that has perfect memory. Traditional security models break when workers can screenshot anything & everything and feed it to an AI which has perfect memory.
CEOs are having FOMO. Their peers talk about cutting headcount by 30% thanks to AI. Their golf buddies brag that they’re no longer hiring human workers into IC1 or IC2 level positions. They’re reading McKinsey reports about trillion-dollar opportunities. In board meetings, they’re being asked about their “AI strategy.” While they don’t exactly know what to do (lean in, hedge, or balance both), they know for sure that they don’t want to be the Blockbuster in the Netflix documentary.
Workers are using tools like ChatGPT a lot. Like, a lot a lot, for as much as they can. They really don’t care if IT restricts them or has corporate tools, they just use their preferred tools in their preferred ways anyway. This is happening whether the company likes it or not, and is where the real AI transformation is happening / will happen, as I’ve previously discussed here, here, here, here, here, here, here, and… here.
Workers are building shadow workflows because they’ve discovered AI makes their jobs easier. As Casey Newton pointed out in his recent Platformer newsletter, even high school students are finding ways around restrictions because the tools are just too useful to ignore.
This isn’t a technology problem. It’s the same tension we’ve seen with every technology wave: the enterprise wants control, leadership wants transformation, and workers just want to get their jobs done.
Narayanan and Kapoor’s paper argues that AI will follow the same adoption pattern as electricity or the internet. Powerful! (But slow.) Transformative! (But incremental.) If they’re right, this will have massive implications for enterprise IT. (Eventually.)
Think about electrification of factories, which I’ve written about before. The first factories just swapped steam engines for electric motors, but kept the same layout, processes, and workflows, and therefore saw minimal gain. It took decades of incremental changes before the entire manufacturing process was transformed due to electrical power distribution. What I didn’t write in that article was that throughout those decades, the foundational infrastructure had to be reliable, sold, and boring to provide the technological foundation for that transformation to happen.
That’s where we are with AI in the enterprise today. Everyone’s looking ahead for the massive, revolutionary use cases while missing what’s right in front of them, that AI is already seeping into every crack in the organization through whatever channels are available:
While the impact of these actions is revolutionizing how your workers do their work, the techniques they’re using aren’t even that revolutionary. (Dare we say, they’re… normal?) This is exactly why you need normal, boring, reliable, foundational infrastructure more than ever.
I’ve been working in IT for over three decades and have experienced this exact cycle several times. Each time, the pattern was the same:
My first exposure to this was with the web application boom in the late 1990s. Back then, offices ran on boring old Windows apps. Then one day the web emerged as a new platform for computing, with new abilities to deliver apps without managing endpoints, app updates that only require touching a few servers instead of thousands of endpoints, access to critical applications from any device (not just Windows), from any location (not just the office), over any connection (not just the local network), all with better security (no data stored on endpoints).
This. Sounded. Freaking. Awesome.
Except for one little catch: to get all these benefits of web apps, companies & vendors had to rewrite all their existing Windows applications from scratch for this new platform.
Oh. Dang. That was a lot of work, a lot of money, a lot of time, and a lot of risk.
In fact, solving this back then was what enabled Citrix to really take off, as Citrix WinFrame offered the best of both worlds—companies could keep their existing Windows applications, workflows, and operations, and Citrix provided a modernization wrapper which allowed them to be managed centrally, accessed from any device, with office-like performance even over slower remote connections, and better security since no data was stored on the endpoint.
Since them, similar versions of that story played out around virtualization, the cloud, enterprise mobility, and SaaS.
So when I look to the AI future and I think about how enterprises need to prepare, I just come back the same things they’ve always needed through every one of these transformations:
None of this is sexy. None of it is “AI-first.” And that’s exactly the point.
There’s “AI strategy,” and then there’s “IT infrastructure strategy to support that AI strategy.” Every executive team is debating their AI strategy right now. (Should we build custom models? Should we buy copilots? Should we wait for agents? What do we tell our workers? What do we tell our investors?)
But if AI really is “normal technology” that will be adopted gradually over many years, these are the wrong questions. The right questions are:
These aren’t AI questions; they’re the same infrastructure and security questions you should be asking about any new technology. The “AI-ness” of them might make them seem more urgent, but if you boil off the hype you’ll see they’re the same questions you’ve been asking (and answering) for decades.
I personally believe AI will be transformational. I’ve written extensively on this blog about AI agents operating desktops, the future of human-AI collaboration, and how work itself will be redefined.
But I also know enterprises. They move slowly and deliberate. They have legacy systems from the ’90s that still run critical processes. They have compliance requirements that won’t magically disappear. They have workers who need to get things done today, not in some theoretical future.
This is where infrastructure partners like Citrix matter more than ever. While everyone else pivots to become “AI-first”, we’re doing what we’ve always done: providing the secure, reliable infrastructure that makes work possible, regardless of what that work looks like or who’s doing it.
Even if AI transforms everything overnight (it won’t), you’ll need stable and flexible infrastructure to handle the chaos. And if AI diffuses slowly over years and decades like “normal technology” (it will), you’ll need stable and flexible infrastructure even more. Either way, you need stable and flexible infrastructure. (a.k.a. “the boring stuff.”)
The irony is that in a world where everyone’s chasing the next shiny AI object, being boring might be the most strategic position of all. While others are pivoting and re-pivoting, you’re making sure the work gets done, day after day. Securely. Reliably. Boringly.
At the end of the day, enterprises don’t need another AI strategy or transformation initiative. They just need the infrastructure to handle whatever comes next—normally.
Read more & connect
Join the conversation and discuss this post on LinkedIn. You can find all my posts on my author page (or via RSS).
My upcoming public talks
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。