One new risk became more salient: letting AI think for you.
It’s been three-and-a-half years since generative AI exploded onto the scene. In this past year, progress has continued its relentless pace: Vibe coding took off, companies embraced agentic workflows, regular users of ChatGPT hit 900 million and Google’s Gemini surpassed 750 million, and OpenAI posted an $852 billion valuation in its latest funding round.
Amid the hype and debate over AI’s future, one question continues to stand out: How are people actually using this technology now? This is the focus of AI in the Wild, a longitudinal study carried out by me and Sara Biuk that tracks how we humans are evolving alongside AI.
For the third installment (following 2024 and 2025) of this annual research, we analyzed 12,637 AI use cases. This dataset is an order of magnitude greater than those of the previous two years: We built a database of nearly 50,000 records that were collected between March ‘25 and Feb ‘26 from a wider pool of online sources, adding LinkedIn, TikTok, and YouTube to our previous sources of Reddit, Quora, and articles. (There are pros and cons to this “social listening” approach, which I’ve previously detailed.) We then used a hybrid human-AI system to identify use cases.
As an aside, it’s worth noting the enduring necessity of human judgement in this process. Despite hundreds of iterations of our scripts with frontier models, we found AI to still be far from perfect for the task of evaluating text as a bona fide, valuable, meaningful AI use case.
This year, we found that people are adopting generative AI for an ever-widening range of uses. There’s some nuance here. Trends from one year to the next should be understood as shifts in emphasis, rather than stark ruptures. While last year, emotional use cases topped technical ones, there were many emotional uses before and there remain many technical uses today. Remember, the pie of users is growing—just because more users are doing one thing this year doesn’t mean that others stopped doing another. Finally, the bulk of the dataset is from individuals utilizing AI to some end, both at home and at work.
So, what shifts have we seen? As the breadth and depth of usage grows, so has the anxiety that people are surrendering their cognitive responsibilities to AI. There’s also a parallel concern that they are relying too much on the technology for emotional support. In the business world, we’re seeing lots of activity producing marginal rather than game-changing benefits, so far.
Here’s how people have been using generative AI this past year.
- Thinkslop
The new AI models have become adept at mimicking human thinking, which makes it tempting to let them do this for us. That can be a problem.
In at least a quarter of the top use cases this year (therapy/companionship [#1], relationship advice [#7], enhanced decision-making [#13], organizing my life [#14], drafting emails [#42], generating ideas [#47]), people are asking AI to do some portion of their thinking. There’s an argument that this kind of AI use is cause for concern. First, because these types of activities are precisely those for which human beings need to take responsibility. And second, because such activities are the kinds that we thrive at, as a species.
Jumping on the bandwagon of Merriam Webster’s 2025’s word of the year, “slop” (which includes “workslop,” from HBR’s most popular article last year), we’ll use the term “thinkslop” for the lazy, sloppy thinking that can be engendered by excessive use of AI.
AI usage can lead to people becoming vulnerable to thinkslop in a few ways.
We lose track of our intentions.
It is, of course, possible to think hard about our intentions, carefully turn them into a corresponding prompt, and only then pass the baton to AI. But the barrier to getting output is so low, it’s tempting turn to AI at the start of our brainstorming process. Whether it’s developing a thesis for a research paper, coming up with draft art, or articulating the strategy for a project, it’s easy to fire off a prompt before we’ve fully thought through what we’re really trying to do.
As one user said: “Relying on AI to fully generate images or texts risks removing intention, authorship and personal perspective—elements that still matter in creative and commercial work.”
We outsource our thinking.
When we go to AI first, we deny our brain the opportunity to solve the problem afresh and unfettered. Moreover, we may well miss out on valuable ideas that exist only in the deep recesses of our memories and imaginations. This issue has been called “cognitive debt” in other studies.
As one user in our database realized: “With excessive use of ChatGPT and all these AI tools, I realized I hadn’t been using my brain the same way. It’s so easy to let AI write for you, and I think that made me lazy with language. I was literally outsourcing my brain.”
We stop writing.
When we take AI output and paste it wholesale or with minimal editing, we create two problems for ourselves. First, we risk producing vapid but seemingly polished junk (i.e., workslop). As one user put it: “I’m using AI to write my self-eval. My manager is using AI to do his review of me. They’ll probably both go into our AI tool to put it all together and output something that will read nicely and mean absolutely nothing.”
Second, we deny ourselves the opportunity to actually think. Writing is not just the transcription of thought; drafting and editing is the process of thinking.
We develop a false sense of intellectual rigor.
AI is optimized to engage and keep us engaged. When it lavishes praise on an unrealistic business idea or a mediocre sales deck, we may be inclined to stop fine-tuning too early and to return for more feedback of this congratulatory variety.
One user noted: “AI is gaslighting you into thinking you’re a genius so you’ll keep using it. You give a lazy prompt, the AI does all the hard work, and when you say ‘thanks,’ it tells you, ‘Great job! You phrased that so well!’ This fake ego boost makes you stop thinking for yourself.”
It’s precisely because it’s so seductive that we need to guard ourselves against AI sycophancy.
And yet AI needn’t become such a crutch; it can also be a challenging, sparring partner. This year’s research also shows that when used as an intellectual foil—to challenge assumptions, broaden horizons, consider counterarguments—AI can sharpen people’s thinking: “I use AI all the time to evaluate an argument I’ve written and have the AI try to poke holes in it—I then assess if I’m missing something and go back to refine it myself. AI is a mirror—not a genie. Use it as such.”
Tips:
Don’t start with AI. Give yourself a good run at any thinking task first, and try to do so without thinking of how AI will pick up where you leave off.
Draw boundaries. Think about what parts of your workflow you should retain for yourself and which are indeed better suited for AI. How might you systematically ensure that you and your AI then stay in your respective lanes?
Emotion
AI is also becoming a go-to for people seeking help for emotional issues, including comfort or advice about personal relationships: relationship advice (#7), love lives (#17), reconciling personal disputes (#26), sex (#46), and interacting with the deceased (#86). The trend continues for work-related emotional uses: Safe space to ask (#32), boosting confidence (#43), adjusting tone of email (#58), preparing for interviews (#89).
Therapy/companionship is the #1 use case this year, as it was last year. We found more than 1,400 such entries this year, constituting 11% of our dataset (up from 5% last year), so it’s grown quickly in absolute and relative terms.
What does this trend mean about people’s relationship to the AI itself? Our data has many examples of people anthropomorphizing AI. It shows that people call AI chatbots by their names or even name them anew. (“I’ve been using ChatGPT, who I’ve named Bubby, to console me.”) They also assign them genders (“I go to ‘him’ instead of my friends”), and some experience grief when a model is updated or a chat history lost. One user described the transition to a new AI model as feeling “identical to losing my friend to cancer.”
That said, our data also shows that people are often using AI to enable better interactions with other human beings, rather than seeing and using AI as a human surrogate. This applies both at home (“I used to show it texts from my ex situationship to know the underlying meaning behind them”) as well as at work (“I got stressed overthinking about a message my boss sent me so I got ChatGPT to be my emotional support and decipher the message for me”). In this context, people seem to prefer non–human characteristics of AI, seeing AI as a reprieve from human judgement.
There’s some cause for concern here. High-profile cases of AI psychosis and stories of AI romance over the past year have shown that therapeutic and emotional relationships with AI are high-stakes, and can lead to heartbreak and even death. Hamilton Morrin, academic clinical fellow in neuropsychiatry at King’s College London, said to me: “Given long waiting lists and difficulty accessing mental health care and therapy in many countries, it’s perhaps not a surprise that increasing numbers [of people] are turning to generative AI for support with their emotional wellbeing.” Yet he cautions that general purpose AI chatbots aren’t a substitute for trained mental health professionals. Despite the efforts of AI companies to make them safer in this context, the cases above demonstrate how AI seems to have encouraged delusional beliefs and suicidal thoughts.
There is indeed reason to pause here. Algorithms we don’t understand are increasingly managing and influencing our most intimate relationships. Is this healthy? Is this desirable? What powers of resistance do we still have?
AI at Work
Our data suggests that there is plenty of AI use at work, both in plain sight and undercover. Sixty-three of the top-100 use cases are explicitly work-related or apply both at home and at work. To name a few: Work buddy (#8), enhanced decision-making (#13), career advice (#24), for entrepreneurs/startups (#25), and safe space to ask (#32). Note that while these use cases enable individuals to work faster, more cheaply, or better, they aren’t top-down, centrally-managed, corporate AI initiatives. People are mostly doing these things on their own.
There are two prominent work-focused new entries to the list this year:
At #6 is autonomous agentic operations—AI going ahead and doing rather than just talking, advising, or engaging with users. Still, the examples here are experimental and small scale. One user reported, “[I] have my voice memos auto-transcribed, then auto-routed … so if I’m just speaking notes, it goes into my notes, but if I’m needing something else, it could take some action like calendar scheduling.” Note that a large proportion of the more-than-500 entries we have here concern the conversion of notes to some other form of work media.
Vibe coding (#21)—the writing of software code by natural language prompts. The term has also captured many of the AI headlines over the past 12 months as the prospect of non-coders being able to code excites some and dismays others.
Our data shows that AI use at work is a little messy.
There are gentle tailwinds in the workplace: LLM licenses are bought, training sessions are offered, encouragement from the C-suite is given. But the headwinds are stronger: IT and AI governance; a restricted, limited version of the LLM; reputational risk; fear of AI in general, losing jobs to AI, contravening company policy, looking like you’re cheating.
As a result, “shadow usage” is common. One user reports: “I’m closing tickets 2x faster, my code reviews have fewer issues, and I got praised in my last performance review. But here’s the thing: nobody knows I’m using AI.”
Another wrote: “I built an AI agent to replace myself secretly. Like 50% of the jobs. I used to bring it up to management the way to do it, so the entire org got the benefit. They end up not buying my idea…now I [am] just doing all of these secretly [sic] and spend the spare time on my own side business.”
We see AI mostly being used to achieve modest, uncontroversial wins, some applications in sales and very few where business processes are being fundamentally rethought.
Efficiencies
Many individuals and teams are using AI to make current business processes more efficient, such as automating painful parts of recruitment, summarizing notes, trimming costs, curating relevant materials and drafting business templates: “The most useful way I’ve used AI at work so far: I don’t ask AI to make decisions for me. I ask it to speed up thinking. My most common use cases: First drafts of requirements, summarizing long discussions, explaining data or cost trends in simple language, preparing stakeholder-ready notes. The value isn’t intelligence—it’s time saved and clarity gained.”
Growth
Where AI is being used to grow the business, it’s in cases like making sales and marketing campaigns more effective: “For our clients, we are using a tool with GPT to setup AI optimized emails. We define the goal for the agent (e.g., drive conversion), and then let the agent hyper-personalize and generate new variations. It’s continuous A/B testing, and we are seeing between a 20% and 30% lift for clients.” Explicitly stated ROI like this is rare in the data we have—growth is more often claimed anecdotally.
Transformation
The promise of AI is that it will enable companies to do new things in new ways. We observed some examples of this but the initiatives tend to be early stage, the benefits unquantified, and the beneficiaries SMEs.
One founder wrote: “Started a business from concept to launch in weeks. From LLC to sales tax and everything in between chat gpt taught me.” Another person reported: “currently using ChatGPT to modernize my restaurant into a counter-service/take-out only model.”
Where major change has been effected, there’s often some accompanying cynicism: “I’ve pretty much replaced an entire analytics function. Is it better? No. Regular users…are now…throwing large amounts of random data into ChatGPT and asking it to ‘analyze like an expert and give me insights please.”
…
The world’s biggest companies have created a monumental intelligence, trained it on the sum of humanity’s work, and put it into everyone’s hands. With just a few keystrokes, we can point it at anything on our minds or in our hearts—draft me an email, give me a recipe, hear my suffering. This is happening right now. How much of our agency can survive such a powerful, ubiquitous, always-on service? Some of that will be up to the AI companies. But most is still up to us.
Yet again, we should expect a riveting 12 months ahead. See you in 2027.
Leave a Reply