The Free AI Lunch is (Almost) Over: We Are at Wide Adoption, Shallow Use

This article was first published in June 2025 as part of an AI column series for GoSeeko. Almost a year on, the observations have held up with remarkable fidelity. A few things have changed — and those are noted — but the core findings remain largely intact.

Recently, I ran a quick, random survey to check how people are dealing with AI. The questions were simple:

Do you use AI? (Yes/No)
If yes, did you start using it directly or take a course first? (Direct/Study)
Do you use AI for work or hobby? (Work/Leisure/Both)

The results were surprising (for me). I was under the impression that most people had already taken to AI. And that many were studying or taking at least free courses on AI. While 62% of the respondents were using AI in some form, all of those who were using AI had just started using the tools without any formal training or study. That's 100%, which is unprecedented in terms of adapting to a new technology. That's how accessible and intuitive Generative AI tools are. Use cases were also a surprise for me: 82% of respondents who are using Generative AI are using it for work. I had assumed that most would have started exploring it for leisure or creative or hobby purposes and I think my perception was based on the fact that most social media posts related to Generative AI are all creative or entertainment related. Of course the survey led to several follow-up conversations on the type of tasks that AI can do well and how efficient the outputs are.

Where the numbers stand today

The Stanford HAI AI Index 2026, published in April 2026, offers the broadest current picture. In 2025, 58% of employees globally reported using AI at work on a semi-regular or regular basis — with India, China, Nigeria, the UAE, Egypt, and Saudi Arabia all exceeding 80%. Country-level adoption varies considerably and correlates strongly with GDP per capita; the US, perhaps surprisingly, ranks 24th globally at 28.3%. At ground level, Gallup's Q4 2025 Workplace AI report, published January 26, 2026, tells a more granular story: 46% of US workers used AI at least occasionally, 26% used it frequently, and daily use stood at 12%. Crucially, overall adoption flatlined between Q3 and Q4 — growth is now happening in depth among existing users, not in new converts. Nearly half of US workers still report never using AI at work at all. Wide adoption in the headlines, shallow and uneven use on the ground. Which is almost exactly what my small survey found a year ago.

But it is important to note that 38% of the respondents in my survey are not using AI and also have no plans to use it in the near future. For most of them there is no resistance or aversion — they are just not interested in what AI has to offer. And this is an important insight that is not obvious if we consider typical news and trends that general media, social media, marketers and AI companies push out to us. The impression created in the broader world is that AI is taking over everything, most people are using AI and so on. The purpose of such 'news' and trends is to create an anxiety or artificial need on account of being left out.

Even among those respondents who are using AI, they are not using it extensively. They are using it for repetitive, boring, or tedious tasks or for fun and exercising judgment judiciously, evaluating AI outcomes closely.

There are however significant sections of people who are not adopting or exploring AI mainly out of fear or inhibitions about loss of ability, loss of skills, privacy concerns, all of which are valid reasons.

Some people tend to do things, then adjust or improve based on what they learned from doing. After some time they try to get a formal understanding of what they are doing, how it is relevant. After some more time, if they are still interested in the subject matter, they try and get structured knowledge (training, or an education). Finally, after a lot more time, they get to a point where they can think deeply about the subject matter, its philosophy and can ask and answer existential questions.

Some people tend to think about things, then share by speaking or writing, adjusting and improving based on feedback, discussions with others and further thought. After some time they get an education, learning from more learned people, comparing new knowledge with what they thought. After some more time, if they are still interested in the subject matter, they work in a structured manner in a real-world situation to test what they have learned for years. Finally, after a lot more time, they get to a point where they can apply and think deeply about the subject matter, its philosophy and can ask and answer existential questions.

What I have described is called a binary. In everyday language, a binary is two clear and opposite parts or positions. And obviously I am defining thinking and doing as opposites. This is not true in all cases, but I am defining it as such. In reality, human beings can have other natures as well, but most of them are usually a combination of thinking and doing. I have used a binary to show two ends of a range.

AI is on a spectrum defined by the binary of Adoption and Rejection. There are those who are exploring it and there are those who are not. In between these two extremes (binary points) lies a range that defines the spectrum. A simple spectrum would read something like:

Adoption < Adaptation < Exploration < Neutral > Inhibition > Resistance > Rejection

But a life without AI is as good a life. Just as a lifestyle without air travel or air conditioning or antibiotics can lead to as good a lifestyle as any other. These are just choices.

My experience is that for every change, I go through a cycle of four stages: Exploration, Literacy, Education and Study. Exploration is the stage where I am winging it, doing stuff and gathering practical information, learning from it and doing stuff again. Pure experience. Literacy is when I consciously set out to find some structured knowledge, inform myself about knowledge that is already available on the subject. The purpose of literacy is to provide meaningful information for decision making usually related to further interest. Content or material that aids literacy is not about detailed academics or theory but overview in nature, providing links between theory and practice, technology and everyday life and so on. Next comes education which is really a deeper dive into the theory and procedures and knowhow so that I can build serious skills related to the subject and use it with some level of proficiency in everyday life and even aim for mastery of some areas. Finally comes study. The simplest of the words but really much more complex. Study is the stage when I finally have sufficient knowledge and experience to look at the change or subject at a philosophical level, to judge whether it is good for me, for society, or for existence itself. These stages are not exclusive nor are they dependent on duration of time. They can happen simultaneously or in sequence and as quickly or as slowly as one wants. The pace really depends on the individual.

The free window is closing — and it is visible now

When I wrote this piece in June 2025, AI was free in a meaningful sense. You could sign into ChatGPT, Claude, Gemini or Grok and use capable models without any payment, any limit that mattered in practice, and without being nudged toward a subscription every few exchanges. That was a genuinely unusual moment in the history of technology.

That moment is not over, but it has measurably narrowed. The architecture of access has shifted from open exploration to managed funnels. ChatGPT's free tier today gives you access to GPT-5.3 — a capable model — but caps it at 10 messages every five hours. Hit that limit and the system automatically downgrades you to the mini version: faster, lighter, noticeably weaker on anything complex. Claude operates similarly — a free tier exists, but extended sessions, larger context windows, and the more capable models sit behind a paywall. Gemini follows the same pattern. Across the board, the free experience is now designed to demonstrate what the product can do, not to let you actually do your work with it consistently.

The pricing architecture has also become deliberately layered. Take ChatGPT as the clearest example: Free, Plus at $20/month, Pro at $200/month. Free gives you a taste of the flagship model with tight rolling limits. Plus gives you 160 messages every three hours before fallback kicks in. Pro claims near-unlimited access and priority compute. The most capable reasoning modes — the ones that genuinely handle complex, multi-step problems — are reserved for the upper tiers, or throttled so aggressively for free users that they are effectively unavailable during peak hours. Advanced voice, extended memory, the ability to run deeper analysis across long documents: these are now differentiated products, not features everyone gets.

This is rational business behaviour and no one should be surprised by it. AI infrastructure is expensive. The question was always when, not whether, the funnel would tighten. What is worth noting is that for a few years — roughly 2022 to 2025 — the window was genuinely open. Anyone willing to explore had real access to real capability. That window served an important purpose: it seeded adoption, built intuition, and let people develop actual opinions about what AI could and couldn't do for them. The people who used that window are in a meaningfully different position from those who are only now starting to explore, behind paywalls and usage caps.

At this moment, AI is still relatively free. In the history of human civilisation, technologies have been rarely accessible to the common people. And when they have been, people have rarely understood its importance, its impact, its uses. AI is different because it is an unpredictable, evolving technology and companies still need people to use it so that they can get a handle on what it can be used for. But by the time most people realise the importance of participating in the free access, the freedom would have passed — big business has already found the way to control and monetise access. Your window to explore AI, evaluate its usefulness, weaknesses, risks, and figure out how to work with it, is closing. I would say 6–9 months.