The AI Daily Brief: Artificial Intelligence News and Analysis - Where AI Is Right Now: 15 Charts in 15 Minutes

Starting point is 00:00:00 Today on the AI Daily Brief, 15 charts and 15 minutes that share where AI is right now. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. Welcome back to the AI Daily Brief. All right, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, Blitzy, and Superintelligent. And to get an ad-free version of the show, go to patreon.com slash AI Daily Brief. Today we are doing a fun little episode. This week I was traveling to a conference to do a couple of things,

Starting point is 00:00:36 including this kickoff session called Where AI is. The idea was to give people 15 charts that told the state of AI in 15 minutes. For the event, I had to stick really closely to that. For our purposes, it'll probably be slightly more than 15 charts, but you get the idea. For those of you who are trying to catch up your friends or colleagues or family on where AI is right now, hopefully this will be a useful episode. And we got to kick off with the big theme of the moment, which has to be acceleration. One of the things that's extremely important to note about AI is that it is not just moving fast.

Starting point is 00:01:08 It keeps getting faster. It was already the fastest growing technology when you look at things like fastest to 100 million users, which Chat ChbT reached in five weeks, beating the previous fastest which was TikTok at eight months, but it's done nothing but accelerate as you can see from these charts. In fact, we got a major inflection point this year where when deep research and reasoning models launched, ChatGBTGBTWT went from 400 to 800 million users in just a couple of months. I don't have it in here, but another chart that could represent this right now is Anthropics revenue.

Starting point is 00:01:39 They started the year at $1 billion. It took about a quarter to get to $3 billion. And then it took a month to get to $4 billion, another month to get to $5. And there is no end in sight. It's not just revenue and it's not just users. It's actual usage as well. On their earnings call this quarter, Google shared that they had jumped from processing $480 trillion tokens in May to $980 trillion tokens. in July. That's 104% growth more than doubling in just two months. This is representative of

Starting point is 00:02:06 everything that I've seen recently, which is that even in this broader acceleration, there has in just the last couple of months been an even more profound inflection. Now, because of this, for those of you who are worried that maybe the market pricing of compute and compute-related companies is unsustainable. Fear not. Between June of 24 and May of 25, we saw nearly 4,300% growth, in total tokens consumed per week. This year, JPMorgan also estimates that the shortfall between supply and demand for data center capacity will actually increase from last year. They also anticipate that shortfall being at around 10 gigawatts for the foreseeable future.

Starting point is 00:02:45 And yet, even as this demand is increasing and even as we're facing shortages and compute, the costs just keep coming down. The cost of inference specifically has come down precipitously with no end in sight. As it does so, it opens up more use cases, which is a good thing, because recently Anthropic has actually had to throttle devs who were literally just running Claude 24-7. As bullish as it is, in fact, that cost is coming down even as demand goes up. There is still a bit of a short-term misalignment. Part of what we're seeing, I think, with charts like this one, Google's inflection, is the rise in agentic coding. And what people are

Starting point is 00:03:19 discovering is that some of the most interesting use cases aren't just sitting there, having an assistant help you. It's spinning up lots and lots of agents that can do things in the background, i.e. Ambien or background agents that go off and do their work on their own. Basically, the most exciting use cases are the ones that are taking up the most tokens, and even in a world where cost is coming down, demand is rising faster. It's important to note that the inflection is not just in consumer AI. Ramp keeps track of the share of U.S. businesses that have paid subscriptions to AI, and they saw a huge jump between Q424 and Q125. Well, it took about two years to go from 5% to 25% by the end of Q4. Between Q4 and Q1, the percentage jumped from around 25 to 42% of businesses with

Starting point is 00:04:05 paid subscriptions to AI. No, no coincidence, that inflection came in the wake of reasoning models, which of course opened up all sorts of new use cases, including agentic coding. Now, given that I keep using that word, let's talk about agents. Regular listeners will have heard this one before, but we are rapidly moving right on past the agent experimentation phase into the agent deployment phase. In KPMG's Q2 Pulse survey, they found that the percentage of businesses that had achieved a full deployment of some agent, i.e. that were not just piloting, jumped from 11 to 33%. In other words, enterprise agent deployment's 3xed in a single quarter. Part of the reason for that is that agent performance just keeps going up. You've almost definitely

Starting point is 00:04:45 seen this chart from meter about the time horizon of different software engineering tasks that LLMs can complete. They found that over the course of the last couple of years, agent performance had been doubling around every seven months, but that more recently it seemed like that number was actually about 70 days. And we've talked before about why their methodology might not be perfect. They're basically measuring how long an LLM can complete a task at 50% success, but the point is it's a consistent methodology and it shows this up into the right graph. What's more, while I didn't include it in this one, they recently released a follow-up that showed that across a variety of benchmarks this pattern was holding as well. I've mentioned agentic coding a number of times, and that is definitely,

Starting point is 00:05:22 and fairly definitively the first big breakout use case. You see it in the literally billions of net new revenue generated in a single year, but you also see it in the success of individual companies. After spending like a decade getting to 10 million in ARR, Replit spent just a handful of months getting to 100 million in ARR. Lovable recently announced that it had achieved the 100 million ARR milestone as well in just eight months, making them the fastest company to ever do so. One of the things that I think is profound about agent decoding is that this is an example not just of something where AI and agents are rewriting how professionals do their work. it's also democratizing access to that particular skill set.

Starting point is 00:05:59 Vib coding remains the odds-on leader for AI Buzzword of 2025 because it's totally changing how people interact with code across a variety of positions, not just software engineers. Now, interestingly, a lot of the different studies and surveys that I see seem to validate that agentic coding is the leading enterprise use case as well. In this study from Iconic, which was focused on the companies that are actually building AI, 77% of them were using AI or agents for coding assistance. After that, content generation, documentation, and knowledge retrieval, and product and design were the next highest use cases, and then basically everything else was in a big clump in the middle.

Starting point is 00:06:33 And the big takeaway from these use case charts to the KPMG chart is that agents aren't the next big thing. They are here. Another recent survey of business leaders had 66% of leaders saying that agents were increasing productivity, helping with cost savings, helping improve the speed of decision making. Now, this is of course not to say that agents are perfect. there are still lots of challenges in deployment, but these things are happening fast. Speaking of challenges, let's talk about enterprise challenges. One of the interesting questions is what agents are actually being built for. In that same KPMG survey, one of the questions was how much your agent strategy is

Starting point is 00:07:08 focused on efficiency or productivity gains versus focused on new revenue. The majority of companies said that they were equally focused on efficiency and opportunity, with 46% reporting that, but then it was about 2 to 1 for the remaining audience, with 36% saying that they were mostly focused on efficiency, and 18% saying they were mostly focused on new revenue. Interestingly, no one was exclusively focused on efficiency or new revenue opportunities. They were all doing some hybrid. In another survey, employees and executives were asked how their companies were using AI tools. The three buckets the survey gave were deploying, reshaping, or inventing, deploying being basically supporting adoption of productivity enhancing

Starting point is 00:07:46 tools, reshape being redesigning and to end workflows and processes, and invent being building an innovating entire new business models to drive growth. Already a full half of companies were thinking beyond just deploying assistance into their staff to instead really think about redesigning and reimagining how people worked and even nearly a quarter thinking about how to reinvent fundamentally what they do. That said, the speed of which agents are coming online is creating at least some amount of mismatch with where agent startups and agent companies are building. Stanford recently mapped agent development into four automation zones, a green light zone, a red light zone, a low priority zone, and an R&D opportunity zone. The red light zone were areas where there was

Starting point is 00:08:28 lots of opportunity for automation, but where workers didn't really want automation. Green light is, of course, where there was both opportunity for and demand for automation. Low priority is where it's hard and people don't want it. And the R&D opportunity zone is where there's higher desire for automation, but capabilities lag. Now, as we get deeper into agent, were going to have to have this sort of conversation around matching and aligning current workers with their new digital colleagues. One of the other things that this same Stanford study did was look at what sort of relationship people wanted to have with AI and agents. 45.2% for example said that they wanted an equal partnership with digital employees.

Starting point is 00:09:05 Now, these are obviously very nascent sort of studies and I think that they should be viewed as a snapshot in time, but asking this type of question feels important going forward. alongside these changes there will of course be a shift in the skills that matter. Some skills that are highly compensated now will become much less in demand from humans because AI does it so well, for example, analyzing data or information, and other types of skills are likely to become more valuable, things like communication, training, and teaching others. We're also starting to figure out what sort of skills we need as agent and digital employee management becomes a key part of what enterprises have to do. A recent research survey, for example, found that executives thought that soft skills like

Starting point is 00:09:41 decision-making, collaboration and teamwork, and logical reasoning, were the most important to effectively build, manage, and capture the potential of agents. Now, speaking of executives, another challenge that they face is that there is often right now a bit of a gap between what employees think and what executives think about their AI strategy. In a study from writer last December that surveyed 800 employees and 800 executives, 75% of executives thought that their company had been successful in adopting AI over the previous 12 months, as compared to only 45% of employees who said that. I think the takeaway for me is that leadership can't just communicate their vision for AI. They have to get people bought in. And part of them getting bought

Starting point is 00:10:18 in is, I think, facing head-on some of the types of challenges that go with that, with one of the big ones being technical and data readiness issues. According to an economist impact survey, only 22% of organizations said that their current architectures were fully capable of supporting AI workloads. Beyond that, there are challenges around things like who can access data, how to deal with data silos and a lack of integration. These are the nitty, gritty types of issues that are going to hold AI and agent deployments back and are going to become a bigger focus for companies this year. So where does this leave us? What's the big picture? The TLDR is that change is even bigger and coming faster than you think. There are basically an infinite number

Starting point is 00:10:54 of charts that I could have chosen to represent that, but the one that I think is really, really notable comes from the recent announcement of ChatGPT agent. What you see on the screen here is how often different models did as well or better on a specific type of task than a human did. Here's how OpenAI summed up the results. On a benchmark designed to evaluate model performance on complex, economically valuable knowledge work tasks, chat GPT agent's output is comparable or better than that of humans in roughly half the cases. In other words, right now no one is claiming AGI, no one is claiming ASI, no one is claiming superintelligence. And yet already, the first version of a general chat GPT agent. In Open AI's estimation, yes, so take with a grain of salt.

Starting point is 00:11:35 They think that the agent can do better than humans in roughly half the tasks. Like I said, the point is, change is even bigger and coming faster than you think. Today's episode is brought to you by KPMG. In today's fiercely competitive market, unlocking AI's potential could help give you a competitive edge, foster growth, and drive new value. But here's the key. You don't need an AI strategy. You need to embed AI into your overall business strategy to truly power.

Starting point is 00:12:02 it up. KPMG can show you how to integrate AI and AI agents into your business strategy in a way that truly works and is built on trusted AI principles and platforms. Check out real stories from KPMG to hear how AI is driving success with its clients at www.kpmg.org.us slash AI. Again, that's www.kpmg.org.comg.com slash AI. 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.

Starting point is 00:12:46 The Blitzy platform provides a plan, then 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-I-D-E development tool, pairing it with their coding co-pilot of choice to bring an AI-Native STLC into their org. Blitzy is providing a limited time, 30-day free proof-of-concept for qualifying enterprises. The team will provide a 5x velocity increase on a real development project in your org. Visit blitzy.com and press book demo to learn how Blitzie transforms your STLC from AI-assisted to

Starting point is 00:13:24 AI Native. That's BLITZY.com. are a regular listener, you will have heard about Super Intelligence Agent Readiness Audits at this point. But I wanted to tell you today about the full suite of Agent Readiness products that go beyond just the initial Readiness Report. Over the last six months, Super Intelligence has built out an entire Agent Planning Suite. We help you move from discovery to planning to implementation. After you've completed your agent readiness audits, we help you double-click on your most important use cases with what we call our use case planning reports. These reports are going to help you understand what sort of technical preparation you need to do to be ready for a use case,

Starting point is 00:14:03 what challenges you might face in implementation, and whether you should be thinking about building, buying, partnering, or some combination. After that, you can even get a spec document in what we call our technical blueprint that gives either your developers or the developers of the partner you work with what they need to build exactly the agent that you're looking for. If you want to learn more about superintelligence agent planning suite, we've built a custom GPT to answer your questions. Just go to bit.ly slash superagent. that's BIT.L.L.Y. slash super agent, all one word. And if you have any questions, the agent can even help you book an appointment with our team. All right. And so with that, we have our 15 charts that express the state of where

Starting point is 00:14:43 AI is. I think we actually did it in a little faster than 15 minutes. But I'm now recording the second part of this episode after having done this presentation. And what was interesting about it is that the presentation led directly into a panel discussion with a number of leaders who are working on AI transformation both inside their companies, as well as more broadly as consulting and professional services partners. There were four big themes to the conversation that I wanted to follow up, and the reason that I think this is interesting to pair with these 15 charts is that this was a four-day conference put on by KPMG that included not only a lot of folks in the KPMG organization who are working with a huge array of different clients, but also some of those enterprise partners

Starting point is 00:15:24 and clients themselves. What that means is that this was a pretty interesting and representative section of the type of challenges that enterprises are dealing with when it comes to figuring out how to actually make sense of all of those trends that we were just discussing and turn that into practical action inside their organizations. So as I said, let's talk about the four big themes from this follow-up panel discussion. The first was a real honing in on this question of efficiency versus opportunity. And there were kind of a couple different places that this conversation went. On the one hand, there's no doubt that when it comes to the long term what people are really excited about and already starting to think about, is what new interesting opportunities AI is going to hold?

Starting point is 00:16:03 What was fascinating about this conversation is that it's very clear that many organizations are really taking this chance to go all the way to their core and foundation and really ask what purpose do we serve in this new world? Basically, rather than just thinking about opportunity as flashy new product line kind of opportunity, it's a moment for broader re-evaluation of the paradigms in which they operate on a more fundamental level. To put it in consulting or professional services terms, instead of just asking what new consulting products can we offer or can we offer our existing products to a new clientele at a cheaper price that matches their budget, additionally, the questions being asked are things like, what is the purpose of consulting

Starting point is 00:16:42 in a world where people have access to this much cheap intelligence? A second really interesting part of the conversation was almost a recalibration and a coming back. There was a reminder from some of these panelists, that even if an organization is doing this big re-evaluation and trying to think broadly about where they fit in this new and novel world, they shouldn't get so distracted by that that they don't take advantage of the efficiency gains that are just sitting there waiting for them. The point being that if you can do what you do currently better, fast, or cheaper, it's probably worth taking the time to figure out how to use AI and agents to do that, even if you know that it's also going to change over the next few years. So really, the conversation went in two totally

Starting point is 00:17:21 different directions. One is an even farther, deeper, more profound exploration of what opportunity really means. And the other was a reminder to not get lost in that sauce and just do the damn productivity work that'll make you work better in the short term as well. The next big theme was about org chart breaks. And this came from a broader conversation around what the challenges of actually putting agents into practice were. The big theme for this part of the conversation was the extent to which roles are being redefined across the organization. In other words, one of the first implications of AI, especially as soon as enterprises moved beyond just simple efficiency and productivity gains from assistance, was that in some cases, the way that the org chart

Starting point is 00:18:00 had been previously structured was ceasing to make sense in exactly the same way. And in fact, I'm going to bring in the third theme here, because I think it blended together a little bit, which is how companies are thinking about AI agents effectively as software, or are they thinking about them as closer and more akin to digital employees. Now, as you might imagine, the more that organizations and enterprises are thinking about agents as digital employees, the more implications there are for those org chart breaks. One of the things that all of the panelists noted is that there's this entire new management discipline around human-to-agent relationships, agent-to-agent relationships, and none of that stuff has precedent that's easy to pull from.

Starting point is 00:18:40 KPMG Steve Chase referenced Nvidia CEO Jensen Huang's argument that in the future IT will be the HR for agents, and he basically argued that, in fact, it would be HR that sees their role expand to incorporate this entire new set of relationships. Another big theme from this was the challenge of employee capacity. We discussed how for a little while there in late 23 and throughout 2024, enterprises seemed like they were prioritizing and adding additional emphasis to upskilling and learning and development and people work in general. But then as soon as agents came online, that started to get shunted back to the side

Starting point is 00:19:16 as leaders started to ask how these digital employees could restructure their organizations. What's clear right now is that there is going to be a totally new skill set around orchestrating digital employees, managing digital employees, integrating digital employees with human employees, and there aren't really great resources for teaching people how to do that at the moment, which I think got to maybe the biggest theme, the undercurrent for a lot of the conversation, which is the leadership gaps that are so endemic as this change happens. The panelists were not painting with a broad brush in the sense that they weren't saying, that every organization was facing some big leadership gap, but what they were noting was how in far

Starting point is 00:19:54 too many cases, leadership seemed to be treating AI and agents strictly as a software consideration, rather than as an actual change management project. Now, as you saw from those statistics from writers survey from December, there is in many cases a big gap between what leaders think of their AI strategy and what employees think of their AI strategy. That gap, I believe, is having negative consequences on the organization where it exists. It's being manifest in slower movement, underperformance, even in some cases, according to other studies, intentional sabotage. It's very clear even just looking at those numbers that leadership needs to be viewing this as an organizational development challenge, as a change management challenge, not just as some new tooling. We have recently hit the point, which was completely inevitable, where Wall Street is cheering on those efficiencies in terms of CEOs and leaders talking about how many people they've been able to let go and replace with agents.

Starting point is 00:20:46 I believe that that creates not only big challenges for organizations in the form of employees who are more concerned than ever about what the future actually holds for their particular role, but it also creates opportunity. It creates opportunities for leaders to plant their flag and actually take a stance on how they want to design their company for the future and get their employees bought in on that vision, or even better give their employees agency to help shape that vision. I strongly believe that the companies who navigate this transitional period well will not just be those who do a good job of implementing the newest models as fast as possible.

Starting point is 00:21:20 It won't even be those who do the best job of getting their data on MCP servers. In many case, the difference will be how well leaders articulate a vision of what they think it means to serve their customers in this new environment, what it means to run their business in the context of all these new capabilities, and the future they see for their people, their human people, in those pictures. Hopefully we start to get more stories of companies that are doing that well so we can scream about it from the rafters and give other companies a template to follow as well. Still, overall, it was a super interesting conversation, which, as I said, I think helps contextualize

Starting point is 00:21:55 how all these trends are rubbing up against the reality of enterprises that they operate today. Anyways, guys, hopefully this gives you a little bit more insight into what's happening out there in the wide world of applied AI. For now, though, that's going to do it for the AI Daily Brief. Appreciate you listening and watching, as always. Until next time, peace.

The AI Daily Brief: Artificial Intelligence News and Analysis - Where AI Is Right Now: 15 Charts in 15 Minutes

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.