Can AI Really Automate 57 Percent of Work?

Today on the AI Daily Brief, can AI really do 57% of all work? Before that in the headlines, just in time for Black Friday, ChatGPT introduces shopping research. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, Robo, Robots and Pencils and Blitzy. To get an ad-free version of the show, go to patreon.com slash AI Daily Brief, or you can subscribe on Apple Podcasts. To learn about sponsoring the show and lock in your rates before they go up for the new year, send us a note at sponsors at aidailybrief.ai. Lastly, due to a number of requests, we're keeping the AI ROI benchmarking study open for just a couple more days. If you are interested in getting the full report, go contribute a handful of use cases and you will get it when it's out in a couple of weeks. But with that, let's dive in. Welcome back to the AI Daily Brief Headlines Edition, all the daily AI news you need in around five minutes. As we head into the holiday season, will you be using ChatGPT as your personal shopper? That's basically the pitch for OpenAI's new shopping research feature, but this is a lot more advanced or in-depth than you might imagine. On the one hand, it does the basic stuff you would imagine, like comparing items and prices to help users find the best fit. but it's really a much more involved, step-by-step, deep research for shopping sort of process. One of the things that OpenAI noticed was that a ton of people use ChatGPT as a way to, in their words, find, understand, and compare products. Sometimes that's about just finding what their options are. Sometimes that's about fitting the options to their needs or preferences. And rather than just showing some shopping-related results when people are searching for those things, shopping research is an entire experience purpose-built for that sort of discovery. Indeed, they make it clear that this is not necessarily just for your everyday simple stuff. They write, For simple shopping questions like checking a price or confirming a feature, a regular chat GPT response is quick and all you need. But when you want depth, comparisons, constraints, trade-offs, shopping research takes a few minutes to give you a more detailed, well-researched answer. So once you get into the experience, which you can automatically select or which can be recommended to you, after you're prompted, it's going to give you a set of follow-up questions. Some of those might be about price. Some of those might be around preference. Some of those might be around the way that you're using it. Some of those will be around distinct features. And after doing a first round of initial thinking, the experience might then ask you to look at a set of different options, giving them a thumbs up or a thumb down or comparing them. So as a way to test this experience, I tried it looking for a robot for my four and a half year old for Christmas. He really wants robots that actually do stuff. And it took me through a whole process experience where it asked a bunch around what types of specific actions it wanted to take. It showed me a number of options and asked whether this was directionally correct or not. And it made a selection that wouldn't necessarily be what, for example, Google search would come up with. The best overall was actually this random little toy robot vacuum, because I had mentioned that he really liked when these things actually did things, not just moved around and talked. Now, interestingly, and as makes sense from what I just said, shopping research isn't just a simple system prompt. OpenAI actually used reinforcement learning to train a version of GPT-5 mini to perform high-quality product research. In their internal benchmarking of product accuracy, the model outperformed the full size of GPT-5 thinking. Initial impressions of this were pretty positive. Olivia Moore from A16Z said, tested ChatGPT's new shopping research feature and I'm pretty impressed. The UI is adaptive to what you're searching for. It also asks you to rate products while conducting the search to better refine options. In my opinion, the results are quite good, with detailed justification for each product. Arthur Lee writes, I used it to buy my wife a new hair color dye since she was concerned with the chemical in the current product she uses. It surfaced a product I would not easily have found. Jonathan Rumer thinks that this is interesting not just from a consumer perspective, but from a business-to-business perspective. Now, obviously, the feature is rolling out just in time for Black Friday, which is expected to be one of the first big tests for AI-assisted shopping. And indeed, Perplexity and Google have also rolled out their own tools for shopping in the lead-up to this big consumer event. Perplexity has relaunched their own version of the feature and made it free for U.S. users. They wrote, AI assistants are at their best when they scale users rather than replace them. They understand intent, remember preferences, and act as extensions of how users would approach a task on their own. Shopping is where an AI assistant can have an outsized impact. Now, the features are very similar. Perplexity's version can also answer specific questions about products, use its memory to hone in on user preference, and present a series of options on product cards to dial in the right choice. But there is also integrated checkout. Thanks to a PayPal partnership, Perplexity users will be able to complete their entire digital shopping trip without leaving the app. Overall, it's clear the AI labs are trying to make a splash with shopping assistance this holiday season, and some industry observers think it'll work. In October, Adobe predicted that we'll see a 520% surge in AI-assisted shopping this year. Adobe Analytics found that AI-based traffic to leading retail sites was up 1,300% last year. That was, of course, off an extremely low base, but their accompanying survey found that 53% of shoppers were considering using AI for their shopping this year. Those users largely expected to use AI for recommendations, deal-finding, and gift inspiration. Candidly for me personally, this is one of the least mentally stimulating use cases for AI, but it's also one where it's super obvious why it can be very useful. And I completely anticipate that despite it not getting me out of bed in the morning, it's going to be a feature set that I use very frequently as the capabilities come online. Next up, we move over to a very weird one. NVIDIA seems to be getting defensive as their standing in the industry faces new challenges. Now, this month has, of course, seen a string of questions about NVIDIA's market dominance. First we had Michael Burry short thesis which claimed the useful life of GPUs was overstated But then much bigger than that was the news that Gemini 3 was trained on Google TPUs which had analysts taking those chips seriously as an alternative That was followed up by news that Meta might be buying a bunch of TPUs from Google all of which has NVIDIA feeling the competition in a way that they just haven't before. And despite all of these narratives seeming a little overblown, NVIDIA appears to be taking them very seriously. Earlier in the week, there were reports of a memo circulating on Wall Street that addressed the bear case for NVIDIA, including certain claims of outright fraud. Unlike Enron, the memo read, NVIDIA does not use special-purpose entities to hide debt and inflate revenue. Now, at first, it seemed unlikely that the largest company on Earth would circulate a memo refuting arguments made on X and Substack, but NVIDIA-focused journalists confirmed the memo had come directly from the chipmaker. Then on Tuesday, NVIDIA stock dropped by 6% intraday on news of that deal with Meta. This was their largest drawdown since April. Around midday, the NVIDIA X account posted, We're delighted by Google's success. They've made great advances in AI, and we continue to supply to Google. NVIDIA is a generation ahead of the industry. It's the only platform that runs every AI model and does it everywhere computing is done. NVIDIA offers greater performance, versatility, and fungibility than ASICs, which are designed for specific AI frameworks and functions. This instantly struck everyone as super weird and unnecessarily defensive. When you're the biggest player in the world, you ride above all this stuff. you don't worry about market analysts getting all excited about Google for like five minutes. And this was especially surprising coming from NVIDIA, where CEO Jensen Huang has frankly been a master of PR throughout the boom. In fact, I can't really remember a time where he set a wrong foot in hundreds of podcasts and public appearances. Some think this probably wasn't an errant social media intern. New York Times tech reporter Mike Isaac wrote, everyone is dunking on NVIDIA comms for this statement, but you do not tweet a post like this unless someone at the top got very mad at Google's announcement and said, we need to do something. I don't know, man, all very weird. But I will say that on Polymarket, the odds of Alphabet surpassing NVIDIA in market cap this year has surged 20x just this month. Lastly today, HP is the latest company to flag AI-related job losses, announcing significant layoffs alongside disappointing earnings this week. HP said they expect to reduce headcount by between 4,000 and 6,000 by 2028, representing around a 10% reduction of force. The earnings presentation said the plan was to, quote, drive customer satisfaction, product innovation, and productivity through artificial intelligence adoption and enablement. CEO Enrique Lórez said, two years ago, we started to do some pilots on how AI could help us to drive these things. What we have learned is that we need to start from redesigning the process. And once we know how the process could be redone using agentic AI, it can really have a very significant impact. Now, all that is true, and that's going to be the case for basically every company. The more you go back and redesign from the ground up, the better results you're going to see. But the question here is whether these layoffs are actually about AI or whether that's just a convenient boogeyman. So are these particular AI layoffs truly about tech adoption or just garden variety cost-cutting with a new cover story? This one seems like it might be a little bit of both. HP has been on a downward trajectory for many years. They just finished up a previous cost-cutting initiative announced in 2022, which aimed to cut around 6,000 workers over three years to save $2.2 billion. Those layoffs clearly weren't to do with AI worker replacement. given that ChatGPT was released a week after that announcement. And while HP is definitely using AI to help with design and customer service, these layoffs feel like they would have happened even if there had been no AI. The announcement came alongside an earnings report that fell short of expectations. Top-line revenue grew by just 3.2%, printer sales are down by 4%, and profit margins have been hit hard by tariffs. HP was already in the middle of a major restructuring with both personnel cuts and moving their manufacturing out of China. CEO Lourdes again remarked that HP doesn't have a lot of choice in using AI to cut costs, stating, it's something we have to do to make sure the company stays competitive. Regardless of the truth, there's going to be a lot of chatter like this from Elections Joe on Twitter who writes, either we ban AI or implement UBI. I can't really see any other outcome at this point that doesn't involve insane unemployment rates. That was liked 8,000 times. And ultimately, when it comes to the politics of this, I don't know that it matters that I think laptop mercenary who responded is right when they say, this is all excuses for layoffs that would happen anyway. Something of a preview of what's going to come next year, but for now, that is going to do it for the headlines. Next up, the main episode. Hello, friends. If you've been enjoying what we've been discussing on the show, you'll want to check out another podcast that I've had the privilege to host, which is called You Can With AI from KPMG. Season one was designed to be a set of real stories from real leaders making AI work in their organizations. And now season two is coming and we're back with even bigger conversations. This show is entirely focused on what it's like to actually drive AI change inside your enterprise and has case studies, expert panels, and a lot more practical goodness that I hope will be extremely valuable for you as the listener. Search You Can With AI on Apple, Spotify, or YouTube and subscribe today. Meet Rovo, your AI-powered teammate. Rovo unleashes the potential of your team with AI-powered search, chat, and agents, or build your own agent with Studio. Rovo is powered by your organization's knowledge and lives on Atlassian's trusted and secure platform, so it's always working in the context of your work. Connect Rovo to your favorite SaaS app so no knowledge gets left behind. Rovo runs on the Teamwork Graph, Atlassian's intelligence 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 teammate powered by Atlassian Get started at ROV as in victory O Small nimble teams beat bloated consulting every time Robots & Pencils partners with organizations on intelligent, cloud-native systems powered by AI. They cover human needs, design AI solutions, and cut through complexity to deliver meaningful impact without the layers of bureaucracy. As an AWS-certified partner, Robots & Pencils combines the reach of a large firm with the focus of a trusted partner. With teams across the U.S., Canada, Europe, and Latin America, clients gain local expertise and global scale. As AI evolves, they ensure you keep peace with change. And that means faster results, measurable outcomes, and a partnership built to last. The right partner makes progress inevitable. Partner with Robots & Pencils at robotsandpencils.com slash AIDailybrief. 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, then generates and precompiles 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. Welcome back to the AI Daily Brief. Maybe the biggest question from a macro perspective when it comes to AI is what its actual impact on work will be. How much of current work it can automate, what the new patterns of human AI interaction are going to mean, what new types of work it unlocks, and how it reshapes business on the other side. And of course, what all of that together means for jobs in the economy. And yet at the same time, for most of the three years since ChatGPT launched, these kind of conversations have been largely theoretical. Increasingly, however, various stakeholders across the AI ecosystem are actually sharing research that is grounded in something. In the process, hopefully giving us a better sense of how things are playing out in actuality, not just in our imaginations. Two bits of research that we're going to look at today. One comes from Anthropic and was focused on estimating AI productivity gains from Claude conversations. Now, Anthropic has an ongoing initiative called the Anthropic Economic Index, which is persistent and ongoing research that's meant to understand AI's impact on the economy. They share a bunch of information on the Economic Index website and are continuously publishing new research, trying to understand how AI is and will impact different sectors and jobs. This new just-released research was basically meant to move from questions of usage to questions of impact. As they put it in their announcement tweet, the Anthropic Economic Index tells us where Claude is used and for which tasks, but it doesn't tell us how useful Claude is. How much time does it save. Now, if you have heard me absolutely yammer about the AI ROI benchmarking study, you will know that this is of keen interest to me right now. I think we desperately need better understanding of how AI is actually impacting things in the real world, and we need better, more persistent and ongoing benchmarks to help us understand how that changes over time. By the way, due to popular demand, we're keeping the survey open for just a couple more days. So if you have not contributed yet and you want to get the full readout, you can go to roisurvey.ai. See how smoothly we integrated that right there? In any case, back to Anthropic. The research team sampled around 100,000 conversations. Then they used Claude to estimate how much time was saved for each conversation. Basically, they generated two core estimates for each task in those conversations. A time estimate without AI, in other words, an estimation of the hours a human professional would need to complete that particular task without using AI. And then, of course, the time estimate with AI, which is how long it would take if you were using AI augmentation and assistance. Now, there is a lot of really interesting analysis around just how difficult it is for humans and AI to estimate task duration, but basically they did a bunch of work to vet Claude's estimates against much stronger real-world evidence, things like data sets of thousands of real-world software development tasks gathered from JIRA tickets, in order to make sure that Claude's estimation wasn't wildly off. Overall, they found that the model predictions have meaningful correlation with real-world outcomes, making them useful for comparing one task to another or tracking changes over time. They then organized things into categories of tasks and estimated a task time, an hourly wage, a corresponding task cost, and an estimate of the time savings between using AI and not using AI. So, for example, for post-secondary vocational education teachers, on the task developing curricula and planned course content and methods of instruction, they estimated four and a half hours as the task time, $33 as an hourly wage, and 96% time savings. for executive secretaries and executive administration, preparing invoices, reports, memos, letters, financial statements, and other documents. They estimated that at 1.2 hours at an hourly wage of $37 and estimated 87% time savings. Across all of the different tasks in their sample, on average, they would take about 90 minutes to complete without AI assistance. And they found that Claude speeds up those individual tasks by about 80%. This is, of course, the big banner headline, 80% time savings across tasks represented by 100,000 conversations. Now, there is a lot of nuance here. They point out that task length varies dramatically across different occupations. Food preparation tasks, installation and maintenance tasks, and transportation tasks take 20 to 30 minutes on average, as opposed to, for example, investment-related tasks, which take humans two hours legal tasks that average 1 hours etc They also find that time savings are highly uneven across occupations And while the average was about 80 and the median was 84 there were very significant outliers For example they say the task of checking diagnostic images only shows 20% time savings. In this case, they say because it's a task that can already be done quickly by experts without AI assistance. On the other end of the spectrum, compiling information from reports sees 95% time savings. Now, from there, they get really macro and found that if you assumed it would take 10 years for AI to reach universal adoption across the U.S. economy and using current models, i.e. AI not improving, Claude's estimates would imply an annual increase in U.S. labor productivity of 1.8%. Now, while that might not sound huge, that would nearly double the current long-term growth rate and would achieve some of the highest growth rates in our history, including in the post-war period as well as in the late 1990s. Still, I want to stay a little bit farther off the macro part and focus in on the task-level understanding, and bring it then over to the McKinsey study. Their report is called Agents, Robots, and Us, Skill Partnerships in the Age of AI. And the goal of the report in many ways was to move away from rudimentary job loss type analysis to try to use the atomic unit of a skill as a better way to dissect the likely impacts of AI and agent adoption. One of the goals was to figure out which skills are most likely to change it in what ways, as well as which are most or least exposed to automation and understand the potential economic impacts of the skill disruption and change that's going to happen. So some of the big banner headline statistics, McKinsey estimates, and this is the one that's going to be running around, probably will end up in the title of this show, 57% of US work hours they estimate right now are automatable with today's tech. If companies redesign work around agents, they see the possibility of $2.9 trillion in annual value by 2030. They break different occupations into seven different archetypes based on the potential roles of people, agents, and robots, by which they mean embodied in physical AI. For example, in the people-centric category, that's future work that's done mostly by people. It includes things like registered nurses, psychologists, and firefighters and represents 34% of the current U.S. workforce. On the other end of the spectrum, in the agent-centric category of occupations like accountants, software developers, and lawyers, where big chunks of that work will be done by agents, that represents around 30% of the workforce. The group who will have a very clear mix where humans will lead teams of agents, and their estimation includes sales reps, secondary school teachers, and HR specialists, and represents 21% of the workforce. And then there's a whole bunch more around the robots, which is also really interesting. But for our purposes, you get the point that they're trying to understand exposure of different job categories based on these skills within those job categories to AI and automation. Unsurprisingly, the fastest growing skill is AI fluency, which is up almost 700%. One of the big things that's interesting about the McKinsey report, and that I think merits even more consideration, is that they find that 70% of skills appear in both automatable and non-automatable work and will, in their estimation, be evolving skills. So in the skills change index, they break things into enduring skills, which are things like social and emotional skills that are the most future-proof versus evolving skills that, in their estimation, are very rarely disappearing but are changing meaningfully in how they're applied. The examples they give are things like writing becoming prompting and editing, encoding becoming architecture and debugging. Now, one of the things that both organizations found is a debunking of the idea that automation comes for low-wage work first. Anthropic found that, quote, tasks associated with higher-wage occupations tend to take more time and thus offer the biggest savings from AI. And for McKinsey, their agent-centric archetype with the highest automation potential includes roles averaging $70,000 a year, which is on the high end. One of the other interesting common insights is that both studies are suggesting that even as we get these big gains, like 80% reduction in key task time, we are dealing with new types of bottlenecks. In this case, human bottlenecks like coordination and supervision. Part of the re-architecting of systems is going to be to speed up those human processes that could get in the way of the overall gains. This harkens back to something I've talked about a lot, that when we move to a new technology paradigm, we're not trading existing problems for no problems, we're trading existing problems for a different set of problems that hopefully have benefits overall relative to the old problems that we had before. For those who are thinking about their personal careers, McKinsey's skill change index, while inherently oversimplified as all visualizations are, does provide an interesting visual way to think about what skills to focus on. Think about a four-quadrant chart where the y-axis represents growth in demand from low to high on the top, and the x-axis is exposure to automation once again from low to high. As you are thinking about the skills you want to develop, low exposure to automation and high growth in demand is a pretty valuable quadrant to be in. Now, high growth in demand and high exposure to automation doesn't mean don't do it, but that's a lot of the areas that are going to see the biggest transition in what those skills mean. Writing, coding, and accounting are all going to exist in the future, but look very different than they do now. Where a lot of this nets out is just more work that's needed on the new patterns of interaction with how human AI hybrid teams are going to work together. I think in addition to this task-based analysis of ROI that we're going to see a ton of in 2026, new patterns and templates for hybrid workforce collaboration are going to be a big theme as well. As I said at the beginning, a lot of these conversations have been by default theoretical for the past several years. And I think it's very exciting that we're moving into the time where they can be based in actual research and data from patterns of usage and impact. There is obviously much more in both of these reports than I could get into in this episode. and so I hope you go check them out individually as well. For now, that is going to do it for today's AI Daily Brief. Appreciate you listening or watching as always and until next time, peace.