Transcription
AI was supposed to be this mass paragon of productivity where all of us would be working two-day weeks and sitting on the beach for the rest of the time, right?
A new study out of HBS shows why AI power users are actually working more, not less. And we're exploring what it means for our sense of AI's impact on society as a whole. Welcome back to the AI Daily Brief.
Today we are talking about a study from a couple of HBS professors that was published in the Harvard Business Review. The study is ongoing and comes from Aruna Ranathan and Shingchi Maggi. The TL;DR is that AI, in fact, in a lived context at least so far, is not reducing work. Instead, it is increasing and intensifying it.
There is both good and bad news in this. Or, really, a better way to put it might be that it is neither good nor bad a priority, but creates different types of challenges and opportunities. Maybe the best news is that we're finally deep enough into the impact of AI that we can start to respond to what's actually happening rather than what we think will happen in the future.
So, first, let's talk about this study. Some studies go broad. This one goes deep. The researchers effectively embedded with a 200-employee technology company from April to December of last year. In terms of level-setting the type of company this was, it was not a company where AI use was mandated, but it seems like it was the type of company where employees were going to be positively inclined towards AI in the first place.
Now, the research started from the premise that I think is a common feeling: that AI can save us a bunch of time by making tasks faster. And of course, anyone who's ever used it to help produce a slide deck or generate an image or crunch some data knows that on a single-task level, that's absolutely true.
There is, however, an interesting conundrum that this brings up. If AI lets us do less work, does it also imply that we're less valuable? Will our employers need less of us or fewer of us? This is pretty quintessential to the fear of AI job displacement. And the response to the GenSpark Super Bowl ad showed those fears. While some people took it in the spirit that it was offered, that AI can make our work lives better, many people thought that this was basically Ferris Bueller telling employers that they didn't need people anymore.
Now, here's the interesting thing. For anyone who's been eyeballs-deep in this agentic shift of the last couple of months, the Claude Code Opus 4.5, Codex 5.2, now 5.3 moment that we've been tracking throughout the year on this show, the feeling could not be more the opposite. The new power of having actual productive agents at our disposal has made people feel like they are leaving tons of value on the table. The concern shifts from "Am I valuable?" to "Am I doing enough?"
And interestingly, even in the pre-shift paradigm, this seems to be closer to what these HBS researchers found. They identified three main forms of work intensification that were exhibited among those using AI.
The first they call task expansion. Because AI can fill in gaps in knowledge, they write, workers increasingly stepped into responsibilities that previously belonged to others. Product managers and designers began writing code. Researchers took on engineering tasks, and individuals across the organization attempted work that they would have outsourced, deferred, or avoided entirely in the past. And this, they identify as part of where the intrinsic reward of AI comes from. They write, generative AI made those tasks feel newly accessible. These tools provided what many experienced as an empowering cognitive boost. They reduced dependence on others and offered immediate feedback and correction along the way. The researchers found that while many of these things started as experiments, they ultimately accumulated into a meaningful widening of job scope.
A second category of work intensification, they identify as blurred boundaries between work and non-work. The way they describe it is because AI made beginning a task so easy, workers slipped small amounts of work into moments that had previously been breaks. Many prompted AI during lunch, in meetings, or waiting for a file to load. Some, they write, describe sending a quick last prompt right before leaving their desk so that the AI could work while they stepped away. I know more than a few of you are furiously shaking your heads, knowing exactly how that feels.
The last major category of work intensification they identified as more multitasking. And basically, the idea is that people were doing a bunch of things at once. They discussed manually writing code while AI generated an alternative version, running multiple agents in parallel, or reviving long-deferred tasks because AI could handle them in the background.
So, TL;DR: the work intensified. And there are very clearly some very good things about this. First of all, people can clearly achieve way more than they did before. That means that organizations can move farther, faster. Also, as the researchers identified, the rewards for expanding your capabilities are not just the satisfaction of knowing you helped your organization. There are intrinsic rewards of feelings of new capabilities and mastery in new areas.
For me, I think the biggest one, which the authors don't actually discuss at all, is that this is a fundamental reminder that the aggregate amount of work to be done is not some fixed state. It can always expand up to accommodate more capacity to do the work. And that, I believe, in key ways, changes the calculus around long-term AI job disruption. My belief has always been that in the short term, you will of course see organizations use AI to cut costs and do the same with less. I think in many cases, they will be rewarded in the short term by markets who like that cost-cutting. The winning organizations, however, will be those who use AI to dramatically expand what they do. They will not be focused on doing the same with less, but doing more with the same, or way more with a little more. They will be thinking in terms of new product lines, new revenue streams, new categories and markets to expand into. The winners will view AI not as an efficiency technology, but as an expansionary, opportunity-creating technology. And this points in that direction.
Now, as one total aside, by the way, the so-called SaaS apocalypse that's happening right now may actually have some interesting impacts on that conversation as well. Given that companies are not only not being rewarded for just cutting costs, they're not even being rewarded for staying on the same revenue trajectories. They need to show how they can fundamentally compete for the long term or suffer massive multiple compression. Anyways, we're not strictly talking about efficiency versus opportunity technology, but that's just an interesting observation I have in the background.
So, overall, I think what the researchers are finding is actually net positive. However, like I said, it's probably ultimately neither strictly positive nor strictly negative a priority, and instead just demonstrates what the real challenges will be rather than the challenges we had previously imagined.
And there are certainly real new challenges that the researchers also identified. As people expanded the tasks that they could do, there were spillover effects to other people who had previously done those tasks, who now had new types of cleanup work. For instance, they write, engineers in turn spent more time reviewing, correcting, and guiding AI-generated or AI-assisted work produced by colleagues. These demands extended beyond formal code review. Engineers increasingly found themselves coaching colleagues who were "vibe coding" and finishing partially complete pull requests.
The other big challenge, and certainly the one that these researchers are most focused on, was sort of a "frog boiling in the pot" effect, where people didn't even realize how much less downtime they had and how much expectations around speed of execution had increased without them even realizing. The authors wrote, "Some workers described realizing, often in hindsight, that as prompting during breaks became habitual, downtime no longer provided the same sense of recovery. As a result, work felt less bounded and more ambient, something that could always be advanced a little further." They also note that over time, the AI rhythm raised expectations for speed, not necessarily through explicit demands, but through what became visible and normalized in everyday work. Many workers noted that they were doing more at once and feeling more pressure than before they used AI, even though the time savings from automation had ostensibly been meant to reduce such pressure.
Now, the authors provide a couple of different ways for organizations to think about how to build these new challenges into their AI practice. They talk about intentional pauses, sequencing, and human grounding as some of the new management strategies that organizations might need to put into place.
But hold aside how we respond to the challenges of this shift. It's very clear that the shift is here and happening. One cannot throw a rock at AI Twitter right now without hitting a post like this one from OpenAI President Greg Brockman. "Feels like such a wasted opportunity every moment your agents aren't running." Alli K. Miller writes, "Now, before every long meeting, I'm forced to ask myself what I want Claude Code or Claude Chrome to do for me during that time. Parallel work can be exhausting, unclear what the best approach is." Simon Willis wrote a whole blog post about the research and said, "This captures an effect I've been observing in my own work with LLMs. The productivity boost these things can provide is exhausting. I'm frequently finding myself with work on two or three projects running parallel. I can get so much done, but after just an hour or two, my mental energy for the day feels almost entirely depleted. I've had conversations with people recently who are losing sleep because they're finding building yet another feature with 'just one more prompt' irresistible."
Again, I'm sitting here shaking my head as I think about watching the minutes creep by every night as my bedtime gets later and later as I try to just push a little bit more. The point is that everyone is feeling like this.
And according to a new agentic coding trends report from Anthropic, it certainly seems like this is going to accelerate. This report also came out over the last couple of days. And while nominally and in some ways it's about software engineering and the software development life cycle, it is clearly about much more than that now, as agentic coding has infiltrated everything.
And two of them, I think, set the grounding for who is actually implicated by these trends. Trend seven is that non-technical use cases expand across organizations. That coding capabilities will democratize beyond engineering, that domain experts will implement solutions directly, and the productivity gains will extend across entire organizations. This is obviously happening and it's happening right now. And it means that as we think about how agentic coding is going to change in 2026, the implications are not just for the engineering department, but for all of us.
Relatedly, trend five is agentic coding expands to new services and users. One of my predictions for 2026 was that I thought that we would actually see "vibe coders" hired specifically to work on non-engineering issues. Basically, internally deployed "vibers" to help people in different parts of the organization use software to solve their problems. To get an early preview of what that might look like, go check out on Lenny's podcast, Lenny Rachitsky's recent conversation with Lazar Javanovich. Lazar is a full-time "vibe coder" at Lovable, and I think paints a bit of a picture about how this might look in organizations in the future.
In any case, it's quite clear that the shift from assisted AI to agentic AI is exacerbating some of these feelings of needing to be always on and not doing enough and wanting to always have agents running in the background. In short, all of us are now managers, and we are all feeling the sting of a big, highly capable team that's being underutilized because we haven't gotten it together to tell them what to do. That is going to get worse, not better, based on Anthropic's trend number two: single agents evolving into coordinated teams. The rise of OpenClaw right now is giving us an absolute preview into what this is going to look like. Anthropic's specific prediction is that multi-agent systems will replace single-agent workflows, and that is just happening right now. I've got a thread going on Twitter right now that's basically the Patrick Bateman American Psycho scene where they show off their business cards, except instead, we're showing off the mission controls we've all coded to handle our multiple OpenClaw agents.
So what does this all add up to? First of all, for those who are worried about AI creating mass job displacement, I think in the long term, this certainly puts some evidence in the column that our market system will expand to accommodate all of this new work that is capable. I think that's good news. And while I don't think that we should be Pollyannaish about the potential impacts on job displacement in the short term, I think that overall, that's good news.
But I do also think that these researchers are right to point out that this new capability enhancement is bringing new types of human organizational challenges, ones which very much need to be dealt with and probably need new structures put in place to deal with. But as I said at the beginning, I'm so glad that we're finally in a spot where we can start responding to what's actually happening rather than just our future predictions.
Big thank you to the researchers for doing this important work. And that's going to do it for today's AI Daily Brief. Appreciate you listening or watching, as always. And until next time, peace.