The AI Daily Brief: Artificial Intelligence News and Analysis - Don't Blame AI for Workslop

Starting point is 00:00:00 In the wake of MIT's 95% study, now everyone is talking about the scourge of work slop. So what is it? Is it actually a thing? And if it is a thing, what do we do about it? The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, quick notes before we dive in. First of all, thank you to today's sponsors, Notion, Blitzy, Superintelligent, and Robots and Penciles. To get an ad-free version of the show, go to patreon.com slash AI Daily Brief.

Starting point is 00:00:35 And if you are interested in sponsoring the show, shoot us a note at sponsors at AIDDailybrief.A.I. Welcome back to the A.I. Daily Brief. This being a weekend episode, it seemed like the perfect time to dig into the latest study that is consuming an outsized portion of media attention in and around AI. Hot on the heels of MIT's endlessly referenced 95% failure rate study, which you'll remember suggested that 95% of AI efforts were failing, and got, just an absolute boatload of press before people realized that it was based on interviews with 52 executives and a public reading of earning statements where if an organization hadn't explicitly reported that AI was contributing to revenue growth, that was considered a failure. It's very clear that people are interested right now in narratives around AI underperformance. And this week, that came home to Roost in and around this discussion of new research

Starting point is 00:01:31 from a team up between Stanford Social Media Lab and Better Up, which, is a company selling workforce training and support that honed in on the idea of AI-generated Workslop. The Harvard Business Review writes AI-generated Workslop is destroying productivity. In Fortune, we get AI promised to revolutionize productivity. Instead, Workslop is a giant time suck and the scourge of the 21st century office. CNBC writes AI-generated Workslop is here. It's killing teamwork and causing a multimillion-dollar productivity problem. And there on the homepage of this company Better Up, Workslop is the new busy work and its cost-tall. millions. Betterup defines Workslop as AI-generated content that looks good but lacks substance.

Starting point is 00:02:11 They say it creates the illusion of progress. Slick slides, lengthy reports, overly tightened summaries, or code without context. Rather than saving time, it leaves colleagues to do the real thinking and cleanup. In a survey of 1,150 U.S. desk workers conducted just a couple of weeks ago, 40% of the people they surveyed said that they had received workslop in the past month. The researchers argue that it takes two hours of average time to resolve each incident of Workslop, leading to a monthly cost of $186 per employee and $9 million annually for a 10,000-person company. Now, I am much more interested in the conversation generally than in the context of this specific

Starting point is 00:02:48 report. Look, ultimately, Better Up is a company that is trying to sell a solution to Workslop, and they are using research about Workslop to justify why you should be hiring them or using their products to solve said problem. No disrespect at all there. I actually think that this sort of approach to marketing is authentic, clear, value-aditive, in the sense that it generates content that you can engage with, whether you choose to come to the same conclusions as them or not, but at the end of the day, it does have a particular motivation.

Starting point is 00:03:14 If it didn't support a story that led to this particular company's products, they wouldn't publish it. But like I said, whereas I have been basically universally hostile to the MIT study and everything around it, this work-slop conversation is, I think, more important, regardless of anything having to do with the study itself. Basically what I think is that where we find ourselves is an inevitable waypoint in the story of AI for work adoption. It's a point that we were always going to have to deal with. I believe, as you can probably tell from the title of this episode, that Work Slop is not in fact an AI problem. I think it is instead a human and organizational problem. Consequently, I don't think the solution is an AI solution. I think that it is a human and organizational solution or set of solutions. Basically, I think

Starting point is 00:03:59 I think where we are is that AI is revealing and exacerbating much more fundamental work issues and that to address the scourge and challenge of work slop, we're going to have to address some very core issues. Now, first of all, let's try to define what we're even talking about with this. Vraser X nails it when they write, everyone talks about AI slot, but nobody agrees on what it actually is. They point, however, to a new paper from collaborators at Northeastern University, Stony Brook University, and Meta, that actually tries to put some definitions around Slop. In short, AI Slop is not things like bad grammar. Instead, it is those ponderous fingerprints that instantly give you a sense that what you're reading is from an AI. As Vraser sums up,

Starting point is 00:04:43 it's verbosity, vagueness, repetition, and incoherence. It's characteristics of writing that people do too. As they put it, the study shows Slop is less about machines versus humans and more about the hidden signals that make writing feel sloppy or sharp. Now, if we expand this type of definition out to the broader set of quote-unquote work slop, which can include not just written words, but also certain types of imagery, slide decks, etc. I think that those same patterns of verbosity, vagueness, repetition, incoherence all follow. And I think the better-ups definition of content that looks good but lacks substance, that feeling of a sort of hollowness also helps ground us as well. So let's talk about what isn't the issue here. And that is,

Starting point is 00:05:24 AI model performance. Now, it is absolutely the case that there are certain hallmarks of AI that tend to tip content over into this realm of slop. We just heard some of them, and often the native generations that come out of certain tools are going to have some element of that sloppness. I also think that there are certain categories of tools that aren't really good enough yet for a lot of the use cases that we want them to be good for, at least not with full autonomy. The creation of slide presentations, PowerPoints, et cetera, for example, I think AI tools can help with, and certainly there are a lot of people using tools like Gamma, but there are real limits to how autonomously you can let those tools run right now.

Starting point is 00:06:02 Still, the point is that contra to what basically every one of these articles tries to imply, both in their headline and in their substance, the issue here is not AI underperforming. Simply put, the models are good enough to generate valuable things. And to put it differently, when the model is not generating work of value, it is in in very many cases, less about the raw capabilities of the model itself, and instead about the context of the person who is trying to get that work out of it. I think at core, more than anything else, what the abundance of work slop is showing is the brokenness of the fundamental incentives of work in most settings.

Starting point is 00:06:43 Specifically, AI is revealing just how much of work is people doing things pretty much just to be seen to be doing things. it is a byproduct of a focus on task execution rather than goal completion. It's an issue, in other words, of where we have intentionally or unintentionally structured work performance to be based on measuring inputs versus outputs. In the settings where work slop is proliferating, my strong guess is that many people who are presenting that workslop believe explicitly or implicitly that what they need to do to be successful in their job, at least in so far as it relates

Starting point is 00:07:18 to retaining their job or getting promoted in their job, is about showing off the raw amount of stuff that they did. And AI is revealing that you can do a lot without accomplishing much. And this is, I think, the core issue. AI can be used into wildly divergent and differing ways, depending on where incentives are at work. If the incentive is to simply show off how much you've done to present an endless parade of slides, an endless set of idea memos, AI will oblige. It can output more than you've ever been able to possibly do without any regard for how useful any of that doing actually is. If, on the other hand, you or your organization are outcome-focused, and the goal is not how much can I produce, but in fact, how efficiently can we achieve our goals, AI can also be extremely good for that

Starting point is 00:08:07 in a way that will tend not produce reams and reams of documents because that would actually run counter to the efficiency of accomplishing the goal. All of which is to say that the first thing, that the first and most obvious way to combat work slop is to shift your organization's incentives from measuring inputs, i.e. how much stuff people do to measuring outputs, how effectively and efficiently did they get their goals accomplished. Now, a related issue is that it's not just the incentives to show that you're doing a bunch that screw all of this up, but also the fact that AI is revealing that huge portions of the work that is assigned is completely unnecessary. This is in many ways the organizational and structural embodiment of the problem that we were just

Starting point is 00:08:52 talking about before, where work is assigned effectively to make it seem like everyone's doing something rather than figuring out how to actually accomplish what it is that moves the organization or company forward. In other words, I think one of the things that Workslop is doing is revealing that the work that was done just wasn't really necessary in many cases in the first place. When I posted about this on LinkedIn, Chris O'Dell provided this scenario. A teacher says, students don't use AI to write this, use your brain. The students respond, but the assignment is unnecessary work, and my brain says AI can do it faster. A boss says, use AI to do this, because the assignment is unnecessary work and AI can do it faster. The employee asks, why are we even doing

Starting point is 00:09:30 this assignment then? And the boss has no real answer. Chris concludes, I think the real power of AI transformations is revealing what's mission critical to actually producing outcomes and what is just extraneous processes enterprises have built up over time. Professor Ethan Malick tweeted about this. He wrote, I think the idea of work slop is not that helpful, as it places the burden of appropriate AI use on workers who are given AI tools and told to increase productivity

Starting point is 00:09:54 without efforts by managers to figure out which processes to change or define what good AI productivity looks like. Putting it more simply, he writes, make more PowerPoints as an organizational incentive is not going to work out well. Pessimistically, he concludes, I suspect workslop will become a way to shift responsibility from workers and managers to AI. See, the AI did bad work. It's nobody's fault,

Starting point is 00:10:15 but the AI that made us send useless documents. Bad AI. Which brings us to a third and related issue that naturally falls out of people having to do all sorts of work that they know to be not really all that relevant for accomplishing the mission or goals of the organization, which is people becoming anesthetized to the tasks that they have to do in a day-in-day-out way. This is the entire substance of the movie office space from 25 years ago. What is it exactly that you would say you do here. Well, Bob, let me tell you about TPS reports. Chatbots are great, but they can only take you so far. I've recently been testing Notion's new AI agents, and they are a very different type of experience. These are agents that actually complete entire workflows for you in your style, and best of all,

Starting point is 00:10:58 they work in a channel that you already know and love because they are purpose-built Notion super users. Notion's new AI agents completely expands the range of what Notion can do. It can now build documents from your entire company's knowledge base, organize scattered information into organized reports, basically do tasks that used to take days and get them complete in minutes. These agents don't just help with work, they finish it. Getting started with building on Notion is easier than ever. Notion agents are now your very own super user to help you onboard in minutes.

Starting point is 00:11:26 Your AI teammates are ready to work. Try Notion AI for free at the link in our show notes. 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 pre-compiles code for each task.

Starting point is 00:11:55 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 Blitzis Blitzie as their pre-I-D-E development tool, pairing it with their coding copilot 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

Starting point is 00:12:29 to AI Native. That's BLITZY.com. Today's episode is brought to you by Super Intelligent. Now, one thing that we are having a lot of conversations with folks about is the fact that for some of you, your fiscal year is coming to an end, and that means two things. One, it means planning and thinking about what you're going to do in the next year. And two, it means using up those last of budgets so you don't lose them. If you are an enterprise that happens to find yourself in that situation,

Starting point is 00:12:56 Super Intelligent would love to help on both fronts. We are moving increasingly towards an annual AI planning model, where we map out how you can create an action map of your organization's agent opportunities that represents an executable backlog of AI and agent use cases that you can deliver on over the course of the next year. Additionally, for those end-of-year budgets, we have worked out deals with a number of partners where we can pre-lock in general implementation packages even before you figured out exactly what use cases are going to require them. If you'd like to learn more about superintelligence agent readiness audits and this new end of fiscal year plan, visit us at B-super.a.i. click get started

Starting point is 00:13:32 and make sure to use the word fiscal somewhere in the description. AI isn't a one-off project. It's a partnership that has to evolve as the technology does. Robots and pencils work side by side with clients to bring practical AI into every phase, automation, personalization, decision support, and optimization. They prove what works through applied experimentation and build systems that amplify human potential. As an AWS-certified partner with global delivery centers, robots and pencils combines reach with high-touch service. where others hand off they stay engaged, because partnership isn't a project plan. It's a commitment.

Starting point is 00:14:07 As AI advances, so will their solutions. That's long-term value. Progress starts with the right partner. Start with robots and pencils at robots and pencils.com slash AI Daily Brief. And so in all of these areas, we have causes and remediations that are not about AI. When incentives are warped towards being seen to do work rather than accomplishing goals efficiently, well, then we have to design new incentive structures that focus on accomplishing goals and prioritize the efficiency with which that gets done.

Starting point is 00:14:37 Secondly, given how much work is actually unnecessary in just those legacy built-up processes, we need to use our new focus on accomplishing goals with efficiency to eliminate and hack apart busy work and unproductive tasks that were generously about legacy processes that never changed, and much less generously, just about work theater, and get those things out of the way. We need to eliminate fake work. And then lastly, we need to get alignment and buy-in between leadership and the teams around that new way of working.

Starting point is 00:15:06 I think that any organization that was dealing with the problem of work slop, if they did these things, would find a massive decrease in that problem very, very quickly. And yet, even with that solved, I do think that there will still be problems with work slop specifically. In other words, the way in which certain AI work gets outputted. And that's because it is not necessarily intuitive in all cases. To know how to use these tools, to understand what a good version of tool usage looks like, or to know where these tools are going to struggle and produce subpar outcomes. A lot of the work here is in the same things that we've been giving lip service to for two years now, but not really accomplishing, which is investing in and helping your team actually know how to use these tools.

Starting point is 00:15:51 TLDR is that I'm sorry, but you are simply not going to get out of the problem of work slop without actually investing in your people. So what does that look like? Just give them some credits for Coursera courses and that'll do the trick, right? Wrong. First of all, I think we need to model what quality outputs actually look like. For example, for people that have a lot of presentations as a part of their job, do they know the difference between a presentation or an essay or an image or whatever it is that they're trying to output that is clearly good versus clearly bad, or the more tricky one, which is often the problem with AI, something that is masquerading as good, when it is actually bad. If people don't have those templates, then they don't know what they're

Starting point is 00:16:32 striving to achieve. Think about this study. We're throwing around this term work slop without actually understanding exactly what it means, which leaves us basically to ask our employees to know it when they see it and figure out how to do something about it, even though we haven't defined it. Step one, we got a model quality outputs versus outputs that are insufficient. Second, people need space and support to learn how to interact with the models to accomplish those good outputs. This is, yes, prompt engineering, but it's also context engineering. And once again, it's about more than a course. It is about space and support and time for people to get the reps and the time on task to actually figure this out. We've recently been running analysis on the now thousands and

Starting point is 00:17:14 thousands of interviews we've conducted with execs as part of our agent readiness audits. And one of the things that comes up absolutely most often as a huge blocker, a huge barrier to actually getting value out of AI is the problem of I don't have time to learn how to use the tool that's supposed to save me time. So many organizations right now are encouraging their people or even mandating their people to use these new tools, but they're not simultaneously saying, and by the way, Thursday afternoon from one to three, do nothing but try to use those tools. They're not creating structured space and support for people to do those things. They're just basically assigning homework on top of their normal jobs. Third, this is one that will be a lot of,

Starting point is 00:17:54 seem completely obvious in retrospect, or maybe it won't because it'll be obviated by agents that just outperform in the future. But there needs to be a culture of AI editing and iteration that doesn't just take the default output, but instead works with it and shifts the burden and balance of time from initial creation to editing, trimming, and perfecting. Just to give you guys an example, as I was preparing for this particular episode, here are the pages and pages of different versions of the cover image that I was exploring in advance of doing the episode. different style, different approaches, different prompting. I would be massively underutilizing these tools.

Starting point is 00:18:30 If I was just taking the first or even the second or third thing that it gave me and not going back and trying to interact with it to get exactly what I wanted out of it. But I don't necessarily think that that's super intuitive to people. The raw outputs are so powerful that I think a lot of folks think, well, that's good enough. Now, one group that has really had to live inside this mindset shift throughout the year is, of course, software engineers. Basically, agentic coding got so good this year, so fast, so performance, so powerful,

Starting point is 00:18:59 that it really was no longer a question of if coders were going to use these tools. It was more about what new challenges did these new patterns of usage create that became the new things we had to figure out how to work with. In other words, the question with agentic coding was not, can AI do everything all on its own? It was whether the new challenges that an AI-mediated process creates are worth it for the value it provides by shifting how much can get output in general. You might remember we talked about this Google Cloud study of 5,000 developers earlier in the week.

Starting point is 00:19:29 They found significant increase in a huge number of desirable areas, including total amount of code output, quality of code output, and a number of other factors like that. But they also simultaneously found an increase in some things that weren't as desirable, like code instability. Rune from OpenAI even joked, I've forgotten how to program as of this past month. I just beg and plead with Codex and GPT5 to do it. Many times it works, but I'm clearly just being lazy. Now, that's obviously a bit tongue-in-cheek, but the point is that developers are in the midst of this

Starting point is 00:19:58 adaptation where they're not viewing it as strictly static. They're seeing their job shift around them and changing their behaviors on that basis. I think basically that everyone's going to need to have more of a manager mindset. They're going to need to be able to think and organize and plan out goals that actually move whatever their role or responsibilities are forward. They're going to have to figure out how to delegate parts of that to AI and agents, and they're going to have to work with the outputs that come back to make sure that they are actually all moving in sync towards whatever that set of goals is. And so to sum up, my argument is not that workslop isn't a problem, but that it is not an AI problem and is instead a human and organizational problem.

Starting point is 00:20:38 I think to change it, we have to think first in terms of organizational structures. We need to change incentives to focus on accomplishing goals, not just being seen to do work. We need to eliminate all the fake work that happens all over the place, simply because it always has, and we need to align our teams around those new ways of doing business. Then, with our teams align, we need to support them. We need to help people model what quality outputs actually look like, give them structured space and support to figure out how to work with these tools to get those sort of quality outputs. We need to encourage editing, reminding people that the first generations are not the end product that they're going to turn in. We need everyone to think differently about their roles and about

Starting point is 00:21:15 their relationships with these new powerful digital assistance and employees they have. And with that, I believe, this productivity-destroying scourge of Workslop can be beaten back and defeated, relegated to the junk heap of history as a frustrating, inevitable, but ultimately surmountable part of the transition between the pre-AI and the post-AI work world. What do you think? What am I missing on this conversation about Workslop? Can we overcome this? Is there too much inherent human laziness that we're never going to be able to work out?

Starting point is 00:21:43 Does all of this just get solved because agents come along and get more performing and more autonomous and cut us out of the equation entirely? Let me know what you think of the comments. For now, that's going to do it for today's AI Daily Brief. Appreciate you listening or watching as always. And until next time, peace.

The AI Daily Brief: Artificial Intelligence News and Analysis - Don't Blame AI for Workslop

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