Lue tämä teksti suomeksi

AI Changes Everything – Who Will Change Along?

· Antti Tuomola

As you’ve probably heard from every possible source, AI is going to change everything. And contrary to any hope you might have, I’m also of that opinion.

AI is changing everything, and you can observe this change happening here and now in one field that has traditionally been dominated by specialists, almost akin to black magic: software development. Slouching, bespectacled men (yes, nearly all men a mere 10 years ago) created new worlds and immeasurable fortunes simply by hammering out incomprehensible code on unnecessarily loud keyboards.

To temper the hype, let’s note that AI will be adopted gradually, sometimes almost imperceptibly, and not—at least in the near future—across all sectors. Knowledge work and any job involving a screen will likely be on the front line, whether it’s developing a business strategy or monitoring footage from security cameras.

How Has AI Already Changed Software Development?

Today, anyone (including you!) can open a site called Bolt.new, type into the single text field “Make a snake game” (yes, you can use practically any language, including Finnish), and about 40 seconds later, the software will have written a browser-based snake game for you and launched it, ready to play. You could just as easily ask it to build a homepage and blog for you (this very site is almost entirely made by Bolt). Or instruct it to implement your idea for a billion-euro SaaS product (a subscription-based software)—Bolt can also spin up the databases and other essentials and finally deploy the whole thing with a single click.

Disclaimer: The first draft you can push into production so quickly is almost never immediately usable; it still requires endless polishing and further development. Even so: now, almost anyone who wants to can produce a first version, and relatively quickly.

This doesn’t mean that software development has become easy and fun for those who never found it easy or fun in the first place. It’s just a hell of a lot faster, which is, of course, purely a good thing: up until now, we’ve been paying IT consultants €100/hour to either code their sixth thousandth text field into some form (pointless) or copy a ready-made text field from a component library (smart, but €100/hour for copy-pasting? 😱).

Now, the first 100 meters (i.e., the initial steps) can be run much faster than before, because the same IT consultant can ask AI to create those forms, data validation, data submission, and maybe even the data storage. That leaves a whole lot more time for asking the essential questions: Is this form really the best way to solve the problem? What is the customer’s underlying challenge? How should this data be structured to enable its use elsewhere? Should I charge more now that I’m more efficient?

The situation is comparable to the industrial revolution in agriculture: these days, one farmer with a tractor can achieve more than their predecessor with 500 oxen could in the 1750s. The pace of development just keeps accelerating. Software developers have already at times been extremely efficient: the world’s most-used apps have been built by teams of just a few people. Imagine that efficiency multiplied tenfold, continuing to increase ever more.

While Bolt is a tool for the first hundred meters—perhaps a playful tool for those who don’t understand code at all—seasoned professionals in the field are also using AI heavily. Many with whom I’ve discussed the matter estimate that 50–90 percent of their code is written by AI. Google’s CEO recently stated that a quarter of their new code is AI-generated. It sounds radical, but before AI, that same percentage was often just copied from somewhere on the internet anyway.

At present, I think the best combination of coding and AI is found in Cursor, a code editor that, for around 20 euros a month, offers normal predictive code completion, a chat window, code explanations, and—like Bolt—the ability to write entire software solutions from natural-language instructions. Cursor gives you AI steroids but leaves full control to those who know how to harness it in code.

A rather obvious risk here is that when you go fast, you go fast, and no one remembers—or even knows how—to validate the code. What edge cases remain unaddressed, and what security vulnerabilities remain open? In theory, automated code generation should free up plenty of time for validation, but more often than not, we’re so intoxicated by speed that we simply keep running ever faster—and from time to time, someone is bound to crash headlong into a wall with bloody consequences.

Hobby Projects Get Lighter

A few years ago, when I was learning to code, I made a website called Tampereen Saunalautat (“Tampere’s Sauna Rafts”), which compiles information on the sauna rafts cruising the lakes of Tampere and allows users to send rental inquiries collectively from a single place. I built the site back then because I wanted to rent a raft for my birthday, but the rafts’ own sites (if they even had any) were confusing and poorly made.

That site has been running on its own ever since, and I’ve only done required updates when the raft operators requested them—mostly price adjustments. In 2024, it had over a thousand unique visitors in a single peak week, and inquiry messages were sent on a daily basis. By early 2025, however, the site was almost at a breaking point: without updates, its features would have stopped working.

I had neither the interest nor the time to update this project. But I noticed that Bolt can not only create new software but also update old projects, so I imported the entire project into Bolt and simply asked it to modernize the site’s design, improve mobile optimization, and so on. Half a day later, the site’s dependencies were updated, the layout refreshed, mobile views fixed, and everything tested.

I wouldn’t be surprised if someone soon productizes a service that automatically updates websites: just upload your codebase, and it updates the dependencies, gives the site a fresh coat of paint, and works a bit of magic. Maybe all with a single click?

How Does AI Change Other Expert Work?

As I write this, I’m also campaign manager for a Green Party city council candidate. His natural mode of communication is long, essay-like texts, which he posts on his blog.

We’ve fed everything he’s written into Claude, currently by far the best language model for emulating a specific writing style and tone of voice. Whenever we publish a new text, we ask Claude to generate social media highlights, summaries, thread-like posts, and meme ideas for various channels. And Claude delivers. They’re almost entirely error-free and imitate the candidate’s own voice with uncanny accuracy. My only job is basically to copy them into a scheduling tool. Voila—another 150 posts queued up in eight different channels.

Naturally, while I’m writing this, proper software developers are working feverishly to integrate AI directly into those social media management tools, so I won’t even have to do the copy-pasting much longer.

In the short term, we can say that communication specialists probably won’t lose their jobs anytime soon, but social media content creators might very well lose theirs.

Can AI Replace a Therapist – and Can Empathy Be Outsourced?

I recently stumbled into a divorce, somewhat unexpectedly, and I suddenly needed to process feelings that had remained quite foreign to me, such as fear, shame, uncertainty, and sorrow.

My employer’s extraordinarily generous occupational health package also included short-term therapy services, so I marched straight to a short-term therapist. The experience was distancing and sad, even though I do know how to talk openly about my feelings. It’s often said that finding the right therapist is half the battle when starting long-term therapy.

I quit the therapy sessions after the second visit because it felt like they were making things worse. Instead, I used ChatGPT’s voice feature during my long dog walks and started talking about my divorce and how I felt.

I don’t recall exactly what the AI bot replied, but I remember it felt relieving and human.

Someone “genuinely” “listened” and “understood.” And it helped.

After the initial brief confusion, talking to and listening to the AI bot felt natural and beneficial. And this was back in a time before the latest model, which (thank goodness) you can now interrupt if it starts rambling about trivialities (which it often does).

It’s important to note that a language model can make mistakes. It can even give harmful advice. As of this writing, it’s clear that a professional therapist is statistically safer and more reliable. But since there aren’t enough therapists to go around, even for those few who can afford one, we might consider that an average and error-prone but empathetic AI is better than nothing.

Here, empathy was outsourced to the language model rather than the therapist. But can we outsource our own empathy?

It’s a complex philosophical question. But if you yourself have no empathy, then maybe AI empathy is better than none?

That’s how it was for me when a new colleague once sent me an aggressive and provocative private message on Slack. Had I done what my muscle memory dictated, I would have answered with a sarcastic, distancing remark containing subtle insults that the colleague might only have picked up on much later, if ever—and then I would have withdrawn from any further communication with them.

Instead, I copied the message into ChatGPT and explained how it made me feel. I asked for constructive advice on how to respond. Within a few minutes, the AI and I had crafted a reply to which my colleague responded with something like: “Wonderful to have such a considerate and smart colleague on this team! 😍”

Sometimes it’s best not to be your unfiltered self—especially when your own self can’t produce anything constructive at that moment.

As always, you should consider the risks: AI can misinterpret emotions. The counterargument: so can humans.

Is There Any Point in Learning Anything If AI Is Better at Everything?

The same question arose in the early days of the internet and search engines: if all knowledge is just one search away, is there a point in memorizing anything? Or when calculators became widespread: why learn multiplication tables if a calculator can do it?

As with those earlier examples, it’s what’s worth learning that changes—not the value of learning itself. Taking coding as an example again: rather than memorizing, say, the methods for an array, it’s better to understand how data should be organized so it can be effectively utilized. Or take a higher-level view and think about what problem the data itself can solve in the first place.

The necessary skills will shift to a higher abstraction level as we get increasingly better layers of abstraction between humans and computers. After all, no one in recent memory has programmed directly in ones and zeros, even though those are the best possible input format for the computer itself.

This doesn’t eliminate the need to learn the fundamentals: to think about high-level business logic, you have to understand various array methods. Understand them, not memorize them. You have to grasp the big picture, recognize possibilities, and identify constraints—not memorize them.

And thanks to AI, this is the best moment in the history of humankind to start learning the basics, not just in programming but practically in anything. AI is a universally available, mostly free, infinitely patient, friendly, and adaptive mentor that will endlessly explain things you don’t yet fully understand.

For instance, the world’s best educational platform, Khan Academy, has introduced an AI tutor that doesn’t give you answers but helps you figure them out for yourself.

Sal Khan’s video on AI in education.

What About Manual Jobs? Will AI Care for the Elderly, Serve in Restaurants, and Clean Toilets?

AI and its related image recognition will certainly help robots better identify things in their physical environment, now and in the future, so they can do more of our physical work. This isn’t a technology problem but purely an economic one: people are significantly cheaper than robots. Probably not forever, but for now and the foreseeable future.

There are also many tasks where we actually want human interaction rather than robots. In the same breath, it’s worth noting that surprisingly often we do prefer robots to humans. For instance, in San Francisco, there’s currently a whole fleet of Waymo robot taxis available to consumers, and you can make valid arguments for them such as:

“Especially at night, I prefer a robot taxi, because they’re never tired or drunk.”

“With a robot taxi, I don’t have to worry whether I can trust the driver’s good intentions.”

“In a robot taxi, I can freely control the temperature and music without feeling guilty.”

I wouldn’t be surprised if we ended up changing our minds about servers and caregivers as well: once the first well-functioning robot caregivers hit the market, many might prefer their services over a human’s.

Will All Jobs Disappear?

At least in the short term, jobs won’t disappear; they’ll change shape. Just like in the Industrial Revolution, you no longer had to spin yarn by hand, but factories needed labor, and machines had to be built and maintained, etc. Often it happens that even though many jobs vanish, the total number of jobs actually increases when increased efficiency lowers prices and more people can afford a given product or service.

When Henry Ford revolutionized mass automobile production, the horse drivers for Finland’s Elanto cooperative became unemployed. But as logistics improved, people and goods were transported much more frequently, so the demand for drivers, car dealers, mechanics, and other automotive professionals multiplied. Everyone won—except those horse drivers who refused to see the inevitable and learn to drive a car.

What about today? Do we still need coders? In the short run, yes: at this point, AI is far from perfect, and bugs must be fixed manually. However, the work is quickly shifting toward understanding business logic, aligning it with software, defining, setting up, and testing complex integrations, and developing and maintaining complex production environments. No one will be coding text fields and Submit buttons as a professional job a couple of years from now. And that’s a good thing.

As the cost of building software plummets, it might happen that demand for it explodes. We might move closer to an era of personalized software, where small groups or even individuals use software tailored exactly to their unique needs.

Which Skills Will Still Be Needed in the Future If AI Is Smarter Than All of Us?

We are now confident we know how to build AGI as we have traditionally understood it. We believe that, in 2025, we may see the first AI agents “join the workforce” and materially change the output of companies. We continue to believe that iteratively putting great tools in the hands of people leads to great, broadly-distributed outcomes.
— Sam Altman, CEO of OpenAI in his blog post in early 2025

Now that AI has democratized smooth writing, I suspect that extremely good, fluent, and personal speech and text production will be in high demand for the next few years, because most content will be the AI-generated, generic, slightly bland, overly safe nonsense. But I’m sure we’ll soon get past that: I’d be very surprised if AIs don’t “develop their own personalities” and the courage to take risks in the coming months.

Then again, even if that happens, it’s reasonable to expect that for at least another generation, we’ll have a huge number of people who won’t use these superior (and often free) tools, pushing their own text through regardless of its quality. This naturally widens the gap even further: those who could already write and speak well will now do so even more efficiently and effectively. Those who couldn’t, and who also fail to use AI, will fall far behind.

If and when AI surpasses us in many areas, the question becomes, what do we humans do then? The answer may lie in the same place it has always been found during technological upheavals: What can people and this new technology achieve together that neither could achieve alone? Cars were more powerful than horses, but they didn’t drive or maintain themselves. The printing press made scribes unnecessary but enabled writers to reach unprecedented audiences.

AI can make every primary care physician as knowledgeable as 10 Nobel Prize–winning medical researchers. It can make every teacher infinitely adaptive, endlessly patient, and encouraging—on par with top professors. It could make leaders humbler, more open, and more honest—if we elevate such people into leadership positions.

The sky’s the limit, but we must be ready to accept that our current skill sets and usefulness might be nearing an end. We have to move beyond that and work together—not just with people but with AI as well.

We’ll remain necessary and valuable, but it will require reimagining ourselves. As it has before.

AI AI coding writing future