A year later, or what do we really use AI for in 2026?

8 min read

In my bubble everyone talks about AI and everyone uses it for work. We use the latest models to write code, documentation, analyses and tests, to plan and to build strategy. It’s easy to believe the whole world looks like this. But is that really the case? What do so-called “normal” people, the ones who don’t work in IT, use artificial intelligence for? So I googled around a bit and came across an interesting study.

I remember that a year ago Marc Zao-Sanders’ study for Harvard Business Review made quite a stir. We were supposed to be using AI for emails, reports and code, and it turned out that most often we simply talk to it. The podium went to therapy and companionship, organizing one’s life, and finding purpose. In June this year a new edition of the study came out (already the third!), “How People Are Really Using AI in 2026”. So I decided to check which of last year’s conclusions survived the year.

What 2025 showed

A quick reminder. In 2024 the ranking was led by the uses everyone expected, such as generating all sorts of ideas, searching for specific information and editing text. In 2025 the entire podium was taken by personal matters. Search dropped from third place to thirteenth, because instead of asking a chatbot for keywords we started treating it like a teacher, and Google itself started answering with AI summaries. Programming held up well, with generating code in fifth place and improving it in eighth.

The conclusion was simple. AI had stopped being just a work tool.

Therapy still in first place

First place didn’t change. Therapy and companionship lead the ranking for the second year in a row. That’s probably the most important finding of the whole study. Talking to AI about what’s bothering us wasn’t a passing fad. It’s becoming a lasting trend. The reasons are the same as a year ago. The model is always with us, available at any hour, never loses patience, doesn’t judge and doesn’t cost as much as a visit to a specialist.

Place 2024 2025 2026
1 Generating ideas Therapy and companionship Therapy and companionship
2 Therapy and companionship Organizing one’s life Solving technical problems
3 Searching for specific information Finding purpose Fun and nonsense

Fixing, fun and agents instead of finding purpose

Organizing one’s life and finding purpose dropped off the podium. Their place was taken by two categories that sound much less lofty, solving technical problems and something the author simply calls “fun and nonsense”, meaning play with no purpose at all. Right behind the podium came writing fan fiction and stories. Generating ideas, learning support and creativity also fell.

How should we read this? AI has probably become ordinary. A year ago it was a bit like a new, fascinating companion you ask about the meaning of life. Today it’s more like a search engine or a smarter assistant. Something stopped working, so I ask the model. I’m bored, so I fool around with it. That’s simply what a technology looks like once it has become part of everyday life.

From my perspective, though, the most interesting newcomer in the top ten is something else, autonomous actions by agents. Is the much hyped “year of agents” actually materializing? A year ago models mostly answered questions. Today they increasingly carry out multi-step tasks on their own, for example fixing code, running tests and using tools. In programming you see it every day, and HBR notes that companies are increasingly willing to build such agentic workflows.

The scale is impressive too. According to HBR, ChatGPT has about 900 million regular users and Gemini more than 750 million. Generative AI is no longer a niche productivity tool, it has become a mass technology.

A note on the method

Like last year, the study is based on what people write online. This year it covered 12,637 use cases, selected from almost 50,000 posts collected between March 2025 and February 2026. There are more sources than before, though. Reddit, Quora and articles were joined by LinkedIn, TikTok and YouTube. That alone can shift the results. People write about work on LinkedIn and about fun on TikTok.

HBR describes the results as an ever wider range of uses, and I think that’s how they should be read. A yearly ranking is more a snapshot of changing behavior than a final verdict on what AI is best for. That’s why I would treat changes in position as a direction rather than a precise measurement.

Wider use doesn’t mean better

The ranking also shows growth in categories that give me mixed feelings, namely relationship advice and astrology and tarot. Maybe it’s loneliness, maybe we feel lost or are looking for comfort. It’s a bit scary, because that’s when it’s easy to trust too much, get bad advice, give away private information, become emotionally dependent, or make ill-considered decisions without taking responsibility for them.

What’s more, psychologists see it too. The American Psychological Association reports this year that more and more patients come to therapy with experience of talking to chatbots. My view hasn’t changed. Talking to a model can help sort out your thoughts, but replacing a real conversation with a human being with it is probably not the best idea. Especially in moments of crisis, it shouldn’t be your only support.

The takeaway for now

AI hasn’t picked one role. It’s a teacher, a mechanic, a playmate and, more and more often, a coworker that carries out tasks on its own, all at once. On top of that it’s something like a therapist and a confidant. That last part is a bit scary, and maybe even a bit funny, because we all talk about privacy, and at the end of the day we send our most personal matters to the servers of big corporations without batting an eye.

If you’re building a product or rolling out AI at a company, you should keep this in mind all the more. People have already woven these tools into their daily habits, often very differently from what the companies selling them assumed, and differently from what we think. They will probably use your product differently than you expect, too.

What this means for developers and founders

If you build products, this ranking is worth treating as a kind of map. It shows what people turn to AI for, and they do it mostly through general-purpose chatbots that nobody designed specifically for these purposes. A conversation in a difficult moment, help when something stops working, fun, planning the day. Each of these needs is a potential product. The agents in the top ten also show that people increasingly expect AI to do something for them rather than just give advice.

What seems most important to me, though, is what the ranking doesn’t say outright, and I dare say stereotypical IT people won’t necessarily find their footing here quickly. Almost all of the top uses concern the situation of a specific person, their emotions, their problem, their day. Everyone has access to a language model today, so the advantage will go to the app that understands the problem, addresses it well and remembers a specific user’s history as well as possible. It will also matter to measure well whether the issue actually got resolved, because time spent in the app says nothing here. With products like these it’s easy to mistake engagement for addiction.

One more thing. People also come to AI in their weaker moments (the growth of relationship advice and tarot is hardly a coincidence), so anyone building such a product takes on considerable responsibility. For me that means collecting only the data that is really needed, and a clear path to a human when things get serious. I see the biggest opportunity in products that take one of these everyday needs seriously and handle it better than a general-purpose chatbot.

And how is it for you? Has what you talk to your AI about changed over the past year?

Sources

All posts