What People Are Actually Using AI For Right Now

This podcast is sponsored by Google. Hey folks, I'm Amar, product and design lead at Google DeepMind. Have you ever wanted to build an app for yourself, your friends, or finally launch that side project you've been dreaming about? Now you can bring any idea to life, no coding background required with Gemini 3 in Google AI Studio. It's called vibe coding and we're making it dead simple. Just describe your app and Gemini will wire up the right models for you so you can focus on your creative vision. Head to ai.studio/build to create your first app. Today on the AI Daily Brief, what 100 trillion tokens tell us about real-world AI usage and before that in the headlines, could we be getting GPT 5.2 this week? The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. Alright, friends, quick announcements before we dive in. First of all, thank you to today's sponsors, Gemini, robots and pencils, Blitzie, Robo, and Super Intelligent. To get an ad-free version of the show, go to patreon.com/aiDailyBreath or you can subscribe on Apple Podcasts. In either case, it's just $3 a month for ad-free. And lastly, if you are interested in sponsoring the show, send us a note at [email protected]. Welcome back to the AI Daily Brief headline edition, all the daily AI news you need in around five minutes. And of course, we are kicking off the day with the recap of the weekend's rumors around open AI's code red response to Google. It appears that the first drop of code red will be GPT 5.2. The Vergeston Warren is on the understanding from his sources that GPT 5.2 is earmarked for release on Tuesday. The release date is, of course, still subject to change due to anything from server capacity issues, two leaks from rival labs. And interestingly, Warren sources said that the model was originally slated for this month. So even before code red, it was going to come sometime in December, but that it was being fast-tracked because of the pressure of Gemini 3. And as if OpenAI weren't dealing with enough from the pressure from Gemini 3 and skepticism in the markets, new data from sensor tower also suggests that ChatGee BT user growth has slowed down. According to sensor tower, only 7 million new monthly active users were added last month. That compares to 40 to 60 million being added per month over the summer. What's more, growth was just 6% between August and November. Bloomberg also reported that investors are backing companies tied to Google's AI ecosystem and turning away from Bettslink to OpenAI. Before the release of Gemini 3, their basket of OpenAI exposed public stocks was up 125%. That's now down to 74% since Gemini 3 was released. The basket that was exposed to Google was at around 110% year-to-date when Gemini was released and has now searched to 146%. There is also even some chatter that OpenAI stock has fallen marginally in private markets. Although this one I think we need to have even a little bit more skepticism around, as this signal is really hard to tell in these non-public markets. Regardless, altogether, the stakes are very clearly high for the next iteration of ChatGee BT, but the buzz is that the model could live up to the hype. On December 6th, the Matt Schumer tweeted, "The model landscape is about to be shaken up again." Reporting from last week suggested that GPT-5-2 was ahead of Gemini 3 on internal testing, and on Friday, model leaker and suspected insider, I rule the world, posted a fairly clearly fake benchmark card that went viral. Now, on the one hand, I think most people assumed that this was a nano-banana creation, but still it seems to me like the general sentiment is to think that OpenAI might be right back in this after their next model drop. The betting markets are also going haywire. On Polymarket on Friday, in the market for which company would have the best AI model by the end of 2025, Google was at 87% while OpenAI was at just 10.5%. Keep in mind that 10.5% was already a fairly big jump from where it had been just a couple days earlier. Over the weekend, OpenAI jumped to 25%, although they've now fallen slightly back to 18%. In the coding specific market, however, OpenAI completely flipping things at the end of last week, going from 12.4% to Anthropics 85% on December 5th to now sitting at 75% compared to Anthropics 19% as of this morning December 8th when I'm recording. AI Breakfast wrote, "The insiders know, "and sure enough, it appears that users "exclusively bet on OpenAI-related markets "are loading up an anticipation of the GPT-5 to release." Still, as much as people may be focused on the new models, efforts to improve the user experience could end up being even more impactful. The Vergegan reports that the focus will shift away from, quote, "flashy new features" and towards improving the chatbot speed, reliability, and customizability. And certainly, it's not hard to find evidence for the need for that as well. Also over the last week, we've seen a number of tweets like this one, with users showing links to integrated apps for target, Spotify, and Peloton in response to completely unrelated queries. And initially, in response, OpenAI went the strategy of saying, "Actually, these aren't ads." Head of Chat GBT, Nick Turley wrote, "I'm seeing lots of confusion about ads rumors in Chat GBT. "There are no live tests for ads. "Any screenshots you've seen are either not real "or not ads. "If we do pursue ads, we'll take a thoughtful approach. "People trust Chat GBT and anything we do "will be designed to respect that." Unfortunately for them, a lot of people felt like Benjamin de Cracker, who wrote, "It's not an ad if we just keep repeating that it's not an ad. "He shared an image of a recommendation "to connect to target to shop for home and groceries "on a conversation that seems like it was about a computer "issue and said, "You guys literally "announced a partnership with target right before this. "You're handling this very badly and people are noticing." A few hours later, Chief Research Officer Mark Chen took what I think was probably the better tact and acknowledged that being told a shop at target in every session feels a lot like advertising even if it isn't an ad unit that OpenAI specifically sold. Chen wrote, "I agree that anything that feels like an ad needs "to be handled with care and we fell short. "We've turned off this kind of suggestion "while we improve the model's precision. "We're also looking at better control "so you can dial this down or off "if you don't find it helpful." Benjamin de Cracker, whose post I was just mentioning, responded, "Thank you for taking this seriously, Mark." Point of all this is, OpenAI clearly has a lot of work ahead of it, but also there is lots of excitement about how they might respond. Buko Capital summed it up. OpenAI's code red is bullish, not bearish. It's an admission that they were overeating, getting beaten needed to focus. That's what great teams do. All eyes on how they execute code red. And so we'll just quickly go through a couple of other headlines before we move over into today's main episode. The first is another big thing that people are talking about, which is more departures from Apple. Last week, we learned that senior VP of machine learning and AI strategy, i.e. their head of AI, John Giann Andrea, would be leaving the company. A few days later, meta-secured the services of Alan Dye, Apple's head of UX design. By the end of the week, Apple announced that their general counsel and head of government affairs would also be moving on. Compounding with over a dozen departures from Apple's AI team, we're talking about a major loss of talent in Cupertino. Now, Bloomberg's Apple correspondent Mark German reports that senior VP of hardware technologies, Johnny Shouji, is considering leaving in the near future. German says that Shouji, who he considers to be one of Apple's most respected executives, recently discussed leaving the company with CEO Tim Cook. And while the other departures kind of felt necessary, particularly around Giann Andrea, for this one, it is hard to find a silver lining. Shouji, as German writes, was the architect of Apple's prized in-house chip efforts. And frankly, Apple's M-Series chips have been one of the few unambiguous bright spots for the company over recent years. Twitter user Nicholas wrote, "Shouji has had AI-capable chips and hundreds of millions of devices for years, and Apple software teams still haven't put them to use outside the camera app. I imagine he wants to build chips relevant to AI today." Now, German wrote that, differently than the other executives, Tim Cook has apparently been working aggressively to retain Shouji, an effort that he said included offering a substantial pay package as well as the potential of more responsibility down the road. One scenario floated internally by some execs involved elevating him to the role of chief technology officer. Basically, things just continued to be a mess over there, and we still feel very much in the part before they get things straight. Lastly, today, a couple meta stories. The first is that they have acquired an AI device startup called LimitList to further their wearable strategy, or perhaps to cut off the wearable strategy for others. LimitList was a part of the wave of AI wearables that launched last year. Their device was a small pendant that recorded the user's conversations throughout the day and delivered an AI-generated summary. Now that segment, of course, so far has fallen flat, and multiple companies have now been acquired for their talent, leaving their devices to fall by the wayside. Here again, the LimitList pendant will no longer be sold, although the device will still be supported for at least the next year. Subscriptions will be canceled and existing device owners will have access to the unlimited plan for free. Other services, including their rewind software that records desktop activity and meetings will be sunseted immediately. Now, meta doesn't seem to be acquiring LimitList for their hardware. Instead, the team will join Reality Labs, which produces the meta-ray bands and other AI-enabled smart classes. People are trying to figure out the signal in this one. Is the story meta stocking up on talent in the wearable space because of their high conviction in their lead there? Is it them trying to cut off talent to competitors because of their lead there? Not totally clear. And so what's more, when it comes to AI wearables, that is a category that continues to be in the let's call it pre-product market fit stage. Lastly today, meta's chatbot will now provide up-to-date news content under multiple new media deals. On Friday, meta announced deals with CNN, Fox News, USA Today, people ink and more. Meta said the deals would, quote, "improved meta-AI's ability to deliver timely and relevant content and information with a wide variety of few points in content types." One of the stories that has been muted in 2025, relative to where I think people thought it was going to be, is the story of AI platforms versus copyright holders? But I imagine we'll get a lot more of that in 2026. Indeed, with Proplexity facing a pair of new lawsuits from the Chicago Tribune in the New York Times, arguing that Proplexity's web crawlers have intentionally ignored or evaded technical content protection measures, we have yet another example of where this is going to be fought out in courts in the coming year. Now, that is longer than we can get into in this particular episode. So for now, we will close the headlines and move on to today's main episode. AI Changes Fast. You need a partner built for the long game. Robots and pencils work side by side with organizations to turn AI ambition into real human impact. As an AWS certified partner, they modernize infrastructure, design cloud native systems, and apply AI to create business value. And their partnerships don't end at launch. As AI changes, robots and pencils stays by your side so you keep pace. The difference is close partnership that builds value and compounds over time. Plus, with delivery centers across the US, Canada, Europe, and Latin America, clients get local expertise and global scale. For AI that delivers progress, not promises, visit robotsandpensals.com/AIDailyBreathe. This episode is brought to you by Blitzy, the Enterprise Autonomous Software Development Platform with Infinite Code Context. Blitzy uses thousands of specialized AI agents that think for hours to understand enterprise scale code bases with millions of lines of code. Enterprise engineering leaders start every development sprint with the Blitzy platform, bringing in their development requirements. The Blitzy platform provides a plan that generates and pre-compiles code for each task. Blitzy delivers 80% plus of the development work autonomously while providing a guide for the final 20% of human development work required to complete the sprint. Public companies are achieving a 5X engineering velocity increase when incorporating Blitzy as their pre-IDE development tool, pairing it with their coding pilot of choice to bring an AI native SDLC into their org. Visit Blitzy.com and press get a demo to learn how Blitzy transforms your SDLC from AI assisted to AI native. Meet Robo, your AI-powered teammate. Robo unleashes the potential of your team with AI-powered search, chat, and agents or build your own agent with studio. Robo is powered by your organization's knowledge and lives on Atlassians trusted and secure platform so it's always working in the context of your work. Connect Robo to your favorite SaaS app so no knowledge gets left behind. Robo runs on the team work graph at Atlassians intelligence layer that unifies data across all of your apps and delivers personalized AI insights from day one. Robo is already built into JIRA, Confluence and JIRA service management standard, premium, and enterprise subscriptions. Know the feeling when AI turns from tool to teammate, if you Robo, you know. Discover Robo, your new AI-teamate powered by Atlassian. Get started at ROV as in victory, o.com. Today's episode is brought to you by my company, Super Intelligent. Super Intelligent is an AI planning platform. And right now as we head into 2026, the big theme that we're seeing among the enterprises that we work with is a real determination to make 2026 a year of scaled AI deployments, not just more pilots and experiments. However, many of our partners are stuck on some AI plateau. It might be issues of governance. It might be issues of data readiness. It might be issues of process mapping. Whatever the case, we're launching a new type of assessment called plateau breaker that, as you probably guessed from that name, is about breaking through AI plateaus. We'll deploy voice agents to collect information and diagnose what the real bottlenecks are that are keeping you on that plateau. From there, we put together a blueprint and an action plan that helps you move right through that plateau into full scale deployment and real ROI. If you're interested in learning more about plateau breaker, shoot us a note, contact at bsuper.ai with plateau in the subject line. Welcome back to the AI Daily Brief. Today, we are looking at what people are actually using AI for right now. In other words, beyond our suppositions and our guesses, is there a way to see these specific types of applications that are driving AI adoption? And last week, we got a study that was trying to do exactly that. The study comes from a team up of OpenRouter and A16z. A16z, of course, being a prominent venture fund, an open router being a startup that provides a unified API that gives developers and users access to hundreds of different LLMs through a standard API gateway. So to provide a little bit more background on who OpenRouter is, the service offers a near complete range of proprietary and open source models being served on a range of different infrastructure. They serve 25 trillion tokens monthly across 300 models to 5 million end users. One of the big use cases for OpenRouter is consumer facing AI apps. So basically, developers can use OpenRouter to automatically route requests to the most efficient or appropriate model. It also provides failover services in case service of a favored model goes down. So not hard to imagine how you would use this if you were a startup. Most startups that are providing some sort of consumer or business interface for using AI are trying to abstract away all the details of which model you're using and things like that. And so OpenRouter gives them an alternative to plugging into just a single model. Instead, they can get access to the full suite. It's more redundant. It has potential cost efficiencies. That's the sort of idea here. Now, individual users can also make use of OpenRouter, but that definitely tends to be for extreme power users. By way of example, users can plug their OpenRouter API keys into cursor and get full access to models without needing to handle multiple sets of keys. The study they released last week is called the State of AI, an empirical 100 trillion token study with OpenRouter. In the abstract they write, we analyzed over 100 trillion tokens of real-world LLM interactions across tasks, geographies, and time. And findings underscore that the way developers and end users engage with LLMs in the wild is complex and multifaceted. Now, one more note on the methodology before we dive in. While 100 trillion tokens is absolutely nothing to sneeze at, and is a very meaningful and reasonable sample size to start to infer some patterns, the caveats are that one that's somewhere between a 10th and a 15th of the number of tokens Google Gemini was serving per month before the release of Gemini 3. So while 100 trillion is a lot, it is still a fairly limited sample size overall. The second thing to note is that this pattern of usage is concentrated around people who are building things. So if you did a study like this across all the end users who are using ChatchyBT and Claude and Gemini and things like that, it would probably look a little bit different. So with that out of the way, let's look at what they actually found. There were a few different things that stood out to me. The first, which just absolutely defined the year, is the balance between reasoning versus non-reasoning tokens completely shifted over the course of the year. Remember, it was only at the beginning of December of 2024 when OpenAI's '01 became broadly available. Since then, and over the course of 2025, reasoning model token usage went from basically negligible to now over 50% of tokens consumed. OpenRouter calls this a full paradigm shift, and I think that this is absolutely a key part of the story of AI in 2025. Now, of course, part of what reasoning models open up is more autonomy and agenda capabilities. And while not as dramatic as the growth and reasoning, some indications of that are also starting to show up in the data. They write that the share of requests that invoke tools rose steadily throughout the year from around 0% at the beginning of the year to 15% now. Overall, and this will be surprising to no one who is listening to this show, the dominant use case, by far, has become programming. Early in 2025, programming was around 11% of usage, and now it is over 50%. We are coming up towards end of the year episodes, and I think any accounting of 2025 has to start with the fact that the dominant and most important phenomenon of this year in AI was the rise of AI coding. That sudden, surprisingly, then, is showing up in token consumption in the study. Now, there are some other ways that we see coding as the major use case showing up in the study. The average number of prompt tokens per request, in other words, the average prompt length, grew about 4x over the course of the year, from around 1.5,000 tokens to 6,000 tokens. OpenRouter translated it for us saying, the median request is less, write me an essay, and more, here's a pile of code docs and logs now extract the signal. Now, the next thing that is notable, and in some ways, a lot of the study is a tale of two use cases, is that the other use case that dominates is roleplay, basically everything in and around, chatting with AI in a fantasy context from innocent to not so safe for work. That is particularly true for open source models, where roleplay and/or creative dialogue, as they put it, accounted for more than 50% of OSS usage. Now, actually, before we look more at that, let's look at the patterns of open source versus close source overall. Another big story for this year, at least among developers building AI applications, has been the rise of open source models, and specifically Chinese open source models. OpenRouter notes that by Q4 of this year, open weight models had reached about a third of overall usage, but they also noted that they've plateaued this quarter. Now, this makes sense intuitively, given that this quarter, we've seen some major advances in the closed weight models, like Gemini 3, GPT-5-1, and both Sonnet and Opus 4.5. Still, the landscape looks really different than it did last year at this time in terms of the composition of these two types of models, which makes sense when you remember back that the first big story in AI of this year was the deep-seek moment. Indeed, the rise of Chinese open source models is one of the big phenomenons that OpenRouter noted. They grew from around 1% to as many as 30% in some weeks. In understated fashion, OpenRouter notes release velocity and quality make the market lively. And really, what they're saying and what these numbers are showing is that for developers in 2025, open source models in general, but particularly Chinese open source models, became a major contender when it came to choosing what models you were going to use for your applications. Indeed, it turns out that it's not really any there or it's a both-and. OpenRouter writes, "If you want a single picture of the modern stack, closed models are for high-value workloads, and open models are for high-volume workloads. And as they point out, teams are using both." Now, going back to the breakdown of what people are using open source models for, over 50% of it is role-play and creative dialogue. Now, I think a lot of people are interpreting this as developers using the open models for use cases that clearly have a lot of demand, but which fall outside the bounds of what closed source providers want their models being used for. It is notable, though, that over the course of the summer, programming also became a big part of open source consumption and now sits at between 15% and 20% of usage. Indeed, when it comes to the Chinese open source models, programming and technology, in aggregate, are now ahead of role-play, which is down to 33%. Basically, the current crop of Chinese open source models is being seen as viable for pretty much every type of use case. One last note from their highlight summary that I think is interesting, they observe what they call a Cinderella glass slipper effect for new models. Basically, when a new model gets released, tons of people come in and try it, and the people who persist create what open router calls a foundational cohort who resists substitution even as newer models emerge. Basically, they create a foundation in a base group for that model moving forward. So what are other people's observations of the study? Teng Yan, who runs the chain of thought AI newsletter, noted a couple things. One of them, which he called out specifically, was the division of different models by different usage. He writes Anthropics Claude is used for over 80% of programming in almost zero role-play. It is the serious work model, while deep seek is the entertainment king with two-thirds role-play traffic. He also noted that although people are willing to try new models, as he puts it, quote, a model that's the first to nail a painful workload creates near permanent lock-in. Early 2025 cohorts of Claude Forsonin and Gemini 2.5 Pro still retain 40% to 50% of users six months later, while every later cohort churns. Relatedly, he points out demand is wildly price and elastic. Users happily pay 10 to 50x more per token for Claude or GPT-5, if it saves them 10 minutes of debugging. Being cheap is nowhere near enough. Going back to this idea of different models for different uses, he noted that there is a new medium-sized model sweet spot in the 20 to 70 billion parameter range. TokenBender points out that while this study is super useful for understanding the breakdown of different open source model usage, we probably shouldn't extrapolate their patterns overall, because OpenRouter is a less preferred option for the closed model providers. Most people were focused on the use cases. And on Chaudry writes, OpenRouter reported what everyone building tools already knows. AI usage is mostly long-running coding job with tool calls. J. Little writes, her deep seek was good at role-play, but didn't think 80% of the use would be that low. Sean Chauhan writes, role-playing in creative writing is 52% of open source usage. While VC's fun productivity, humans are using AI to write fanfiction in debug code. The market gap versus reality gap is hilarious. I don't know if that's totally fair. If, for example, you look at the internet, it's not like the fact that there is massive amounts of adult content doesn't mean it's also super useful for productivity. Although it certainly does suggest that there's probably capital opportunities that aren't being taken advantage of, because of particular norms and morals. One sub part of the conversation was about how Grock dominated total consumption charts. But this is potentially a little bit dismissalable and where the limits of this study show up most to me. Grock made tokens available for free for some time on OpenRouter as part of a promotion strategy, which was obviously successful as a way to get people to try it, but which warps the model results at least a little bit. One really interesting reflection came from Brian Contano, who actually got meta on the success of OpenRouter in general. Brian writes, I really thought Cursor and OpenRouter would not become big. Cursor is just a fork of VS code. OpenRouter is just a wrapper on top of model APIs. I was very wrong. I'm realizing that my baseline visceral skepticism of scaffolds and wrappers needs to be unlearned. The AI market he continues is special in its sensitive differentiation. It's easy to switch between providers but evaluating any model or provider is sensitive. Small changes in input cause large changes in output. This is true at the prompt level and at the model level. GPT-5 versus Quad 4.5 as inputs to write my code will yield vastly different results. So buyers in a sensitive differentiated market have the following problem. It's easy to switch between providers and the models are always getting better. In addition, because this market is so new, none of the models are sticky yet. This might change with memory, et cetera. So you end up needing wrappers and scaffolds to do your work over time. Otherwise, you lose out on optionality in a rapidly changing provider market. I keep expecting one model to win, but this hasn't ever really happened. Tangyan again made this point as well. There is no single best model. The top 10 models by volume are from eight different labs. So overall, this is a super interesting study that while focused on a particular audience of app developers and power users in a relatively limited number of 100 trillion tokens still shows some of the big changes that we've been feeling throughout the year. If you wanna check out the study for yourself, you can find it at openrouter.ai. It's on a banner right on top of the website. Thanks to the team there and at A16Z for putting this all together. For now, that's gonna do it for today's AI Daily Brief. Appreciate you guys listening or watching as always. And until next time, peace. (upbeat music)