{"id":16751,"date":"2025-12-09T16:20:26","date_gmt":"2025-12-09T15:20:26","guid":{"rendered":"https:\/\/haimagazine.com\/uncategorized\/how-do-we-really-use-ai-not-the-way-you-might-think\/"},"modified":"2025-12-18T17:14:15","modified_gmt":"2025-12-18T16:14:15","slug":"how-do-we-really-use-ai-not-the-way-you-might-think","status":"publish","type":"post","link":"https:\/\/haimagazine.com\/en\/ai-news-2\/how-do-we-really-use-ai-not-the-way-you-might-think\/","title":{"rendered":"\ud83d\udd12 How do we really use AI? Not the way you might think"},"content":{"rendered":"

We tend to talk about AI as a tool for getting work done. Conventional wisdom says it helps developers, speeds up writing, and keeps documents organized. But is that really how people use AI models day to day? The latest OpenRouter report<\/mark><\/a> suggests that… not really.<\/p>

OpenRouter is a platform that supports over 300 models from 60 providers. More than half of its traffic comes from outside the United States, which gives a global snapshot of behavior. The report doesn\u2019t include conversation content\u2014only metadata. The report\u2019s authors analyzed over 100 trillion tokens generated by hundreds of models and billions of interactions, which makes it possible to see patterns, how long interactions last, and the types of tasks that repeat millions of times.<\/p>

What does this data tell us about AI?<\/p>

Fiction and coding<\/h4>

The biggest surprise is that half the prompts are about creating fiction, narratives, or simulations. Everything from dialogues with characters and worldbuilding to text-based games and literary experiments.<\/p>

Importantly, this isn’t a niche. It’s mainstream.<\/p>

In practice, it\u2019s about more than just fun. AI is becoming a space where people design experiences, try out relationships and explore ideas. This side of things rarely comes up in discussions about productivity, even though it\u2019s growing faster than most other kinds of interaction.<\/p>

It’s also worth noting that how people use AI depends on the type of model. In open source, roleplay interactions dominate, while closed models are heavily focused on technical and professional tasks. That suggests the market is clearly starting to split. Open models are for experimentation and interaction (because they aren’t moderated as strictly), while premium models are chosen where stability and quality matter.<\/p>

On the flip side, programming became the fastest-growing category in 2025. Early in the year, it accounted for about 11% of model usage. By the end, it topped 50%. And this isn’t about writing short snippets of code. The classic “fix my bug” prompt is an exception. What dominates are long sequences, often over 20,000 tokens, so basically entire repositories. The models analyze architecture, spot problems and suggest refactors. They’re becoming real participants in the software development process.<\/p>

Claude’s been leading this category, handling over 60% of programming questions for most of the year.<\/p>

\"\"
Anthropic.<\/strong> Used mostly for programming and tech tasks (over 80%). Source: OpenRouter<\/figcaption><\/figure>
\"\"
Google.<\/strong> Wide range of uses, from legal and scientific to tech and general queries. Source: OpenRouter<\/figcaption><\/figure>
\"\"
xAI.<\/strong> Most token usage centered on coding, and by late November, tech, roleplay and academic use were on the rise. Source: OpenRouter<\/figcaption><\/figure>
\"\"
OpenAI.<\/strong>\u00a0Over time, there\u2019s been a shift toward programming and tech, with a big drop in roleplay and casual chats. Source: OpenRouter<\/figcaption><\/figure>
\"\"
DeepSeek.<\/strong> Mostly used for roleplay chats and casual interactions. Source: OpenRouter<\/figcaption><\/figure>
\"\"
Qwen.<\/strong> A strong focus on programming tasks, with roleplay and scientific categories that change over time. Source: OpenRouter<\/figcaption><\/figure>

Agents are getting more common<\/h4>

As tasks get more complex, the way models work is changing too. More and more interactions are going to reasoning-focused models\u2014the kind that can work through a problem step by step. In practice, that means using extra tools more often, working with very long texts, and following instructions split into multiple stages. So AI is starting to actually plan and carry out tasks, not just generate single answers.<\/p>

This shift toward agency makes interacting with AI feel more like a collaboration than a conversation. Most models are no longer just tools that react to simple commands. Instead, they work like agents that can assess the situation, make decisions and come back to you with the next steps.<\/p>

China is getting stronger<\/h4>

The report also shows a geopolitical shift. Models from China have seen the biggest jump in market share. Over the past year, they went from about 1% to nearly 30% during peak weeks. DeepSeek, Qwen and Moonshot AI are growing the fastest. What\u2019s more, the second most-used language is Simplified Chinese. It\u2019s a clear sign that a new hub for model development is emerging right before our eyes, and it\u2019s going to shape both competition and regional expectations around AI.<\/p>

The global landscape of AI use is shifting too. Asia’s share of total usage has risen from 13% to 31%. That’s another sign the market isn’t driven solely by North America. We’re seeing more local applications, products and preferences.<\/p>

Loyalty through effectiveness<\/h4>

Another interesting finding in the report is about retention. It turns out models build loyalty not when they\u2019re being promoted the loudest, but when they\u2019re the first to solve a real, previously unsolved user problem. When that happens, a cohort forms that sticks with the model for months. This effect appeared with Claude 4 Sonnet and Gemini 2.5 Pro. In some user groups, both models kept around 40% of users even after 5 months of use, which is much longer than for later releases. So sticking with a proven model can matter more than the version number.<\/p>

The report also says that retention doesn\u2019t have to do with brand loyalty, but with the fact that the model is baked into a workflow. Users don\u2019t want to switch to other models, since they\u2019d have to change how they work\u2014rebuild their workflow or rewrite their prompts. That explains why the models that were first to meet a group\u2019s real need gain an advantage that\u2019s hard to shake later.<\/p>

Price isn’t a (big) factor<\/h4>

OpenRouter\u2019s data also show that, in practice, demand barely moves when costs change. A 10% price cut boosts usage by only about 0,5%. Pricier models maintain high usage if they offer a quality edge. Cheaper models grow alongside them, especially for tasks that chew through lots of tokens. The result is a diverse market where different models play different roles instead of replacing one another.<\/p>

The market is also starting to split into four segments: premium leaders, high-performance budget models, niche models, and the long tail. That last group includes hundreds of less popular models that don\u2019t scale big, but stick around thanks to narrow, specialized use cases. They\u2019re not chasing the mass user\u2014they\u2019re built for the specific needs of small groups, often for highly specific tasks. Together, they make up a broad, though scattered, slice of the market that shows just how diverse user expectations can be.<\/p>

How is AI changing the world?<\/h4>

The OpenRouter report shows that AI isn\u2019t evolving according to simple assumptions. It’s not just a productivity tool or a simple “question-and-answer” system. It’s becoming an interactive medium, a partner in technical workflows, and the foundation for new kinds of agent apps.<\/p>

At the same time, the market is more diverse than you\u2019d guess from conversations about individual models or the performance charts and quality claims around new releases. It\u2019s clear you build an edge not with buzz, but by solving a specific problem. AI isn\u2019t changing the world according to its own plan (because for now, it can’t have one), nor according to the plans of model creators. It’s changing it through how people use it. And people need two things: help and connection. So performance matters, but it\u2019s the \u201chuman\u201d side of AI that largely determines its popularity.<\/p>","protected":false},"excerpt":{"rendered":"

The latest look at OpenRouter data shows people use AI far less for writing emails or summarizing documents\u2014and far more for chatting and creative role\u2011playing.<\/p>\n","protected":false},"author":465,"featured_media":16613,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"rank_math_lock_modified_date":false,"footnotes":""},"categories":[813],"tags":[],"popular":[],"difficulty-level":[38],"ppma_author":[892],"class_list":["post-16751","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-news-2","difficulty-level-medium"],"acf":[],"authors":[{"term_id":892,"user_id":465,"is_guest":0,"slug":"kmironczuk","display_name":"Krzysztof Miro\u0144czuk","avatar_url":{"url":"https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/10\/awatar-2.png","url2x":"https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/10\/awatar-2.png"},"first_name":"Krzysztof","last_name":"Miro\u0144czuk","user_url":"","job_title":"","description":"Od lat zajmuj\u0119 si\u0119 nowymi technologiami w biznesie, edukacji i codziennym \u017cyciu. W centrum mojej uwagi pozostaje cz\u0142owiek \u2013 i to, by technologia wyr\u00f3wnywa\u0142a szanse, zamiast tworzy\u0107 bariery."}],"_links":{"self":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/16751","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/users\/465"}],"replies":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/comments?post=16751"}],"version-history":[{"count":1,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/16751\/revisions"}],"predecessor-version":[{"id":16752,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/16751\/revisions\/16752"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/media\/16613"}],"wp:attachment":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/media?parent=16751"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/categories?post=16751"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/tags?post=16751"},{"taxonomy":"popular","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/popular?post=16751"},{"taxonomy":"difficulty-level","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/difficulty-level?post=16751"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/ppma_author?post=16751"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}