In this talk, Toju Duke introduces the concept of "AI workslop," the extra time and effort employees must spend correcting AI-generated errors. Citing a BetterUp/Stanford Social Media Labs study of 1,150 U.S. workers, he quantifies the problem: roughly 40% of workers experienced workslop in the past month, averaging two hours per incident and costing about $186 per employee per month, amounting to an annualized $9 million in lost productivity. Duke explains how AI workslop stems from cognitive offloading—over‑reliance on AI that erodes memory, analytical thinking, and problem‑solving skills—creating a cyclical loop where workers must redo work that AI supposedly saved. He illustrates the issue with high‑profile failures: Deloitte Australia’s AI‑generated government report containing fabricated quotes and citations, leading to reputational damage and refunds; New York lawyers fined for using ChatGPT‑generated legal arguments that were inaccurate. Duke stresses that while AI holds great potential, misuse leads to back‑firing outcomes, including wasted ROI, employee unhappiness, and even rehiring of displaced staff. He concludes with a practical five‑step framework for responsible AI use at work: (1) Ideate – define the problem and brief before prompting; (2) Elaborate – outline key points and goals; (3) Collaborate – use AI to expand ideas and iterate prompts; (4) Validate – fact‑check, proofread, and ensure accuracy; (5) Iterate – refine prompts and outputs, aligning with brand voice. By following this workflow, organizations can mitigate AI workslop, preserve cognitive skills, and realize genuine productivity gains.
Chapters
Have you heard the term AI workslap? It's a new term that refers to the heartache and the extra work that AI gives to employees and colleagues when they have to fact-check and correct the mistakes and errors found in the output of AI models. So we have a recent study from BetterUp and Stanford Social Media Labs that investigated about 1,150 US workers on how they work with AI the product of its work or the quality of the work and the overall employee dynamics. You know, an AI work slop is basically just thinking about the illusion that AI brings to the table where, you know, it provides a lot of slick slides and lengthy reports and amazing summaries and quotes sometimes without context. And rather than saving time, it leaves employees to do the real thinking. And fixing. So what are the hidden costs of work slop? About 40% of US DEX workers received work slop last month, which equated to about 2 hours of average time it took to resolve each incident, and it cost about $186 per month per employee for these incidents. Overall, the annualized, um, revenue for that, or the annualized cost for that, is $9 million. So rather than AI being a collaborative tool, it's being used as a clutch, and it's causing a lot of dissonance within employees, a lot of unhappiness, lack of respect, and it's really negatively impacting employees' dynamics. So over time, long-term reliance on AI can cause cognitive offloading, according to psychologists, where you're actually dependent on a technology and a tool, um, and over time it affects human cognitive abilities such as, you know, memory retention and analytical thinking and problem-solving skills. Um, with WalkSlop, it's making it a bit different because we're not just over-relying on technology to solve a problem for us and offloading our cognitive abilities on technology, it's now a cyclical effect where it's making us do the work again and wasting our time. So let's look at a few examples of work slop. We have a very recent news story that has been all over the news recently from Deloitte Australia where they used an AI system to produce a government report and charge the government $440,000. $1.2 billion, and it was full of incorrect information, you know, from quotes that never existed before and judicial citations that never existed before, and also references that were non-existent. This is a well-known problem with large language models, or LLMs, and we know that, and it's caused a lot of reputational damage for Deloitte in Australia, where they've had to partially refund the government and of course apologize for the problem. We also have stories about, you know, a few other examples like New York lawyers who used ChatGPT to build their judicial cases and they were all incorrect and they've been fined. And of course they've probably lost some face. So I've been talking about using AI appropriately at work and for personal use. It's really important to leverage on the benefits of AI. AI still has potential. I am still a very strong believer in the potential for AI because it is a tool, it is technology, and it's really meant to advance our work, our productivity, our skills, and save us time. That's the whole point. But if we do not use it appropriately, then it's going to backfire. And now we're hearing stories of companies who initially fired humans and their human workers to replace them with AI, retracing their steps and rehiring them back. We hear stories of developers and coders who highly depend on AI coding assistants, working more slowly. Um, we hear stories of companies who were one of the first to implement AI or are implementing AI and working with AI, but they're not seeing any return on investment. All of these stories is because people are using AI inappropriately. So here are a few tips on how to use AI appropriately when using it at work. One, you want to ideate. So think about the problem that you're trying to solve. Think about the idea that you're trying to bring on the table. Create an idea first and a short brief first before you create a prompt. Do not go to AI first. That's just being lazy. That's cognitive offloading and it's going to backfire. The very second thing you want to do is elaborate. So write down a few bullet points, build the body of the project or the, or the goal that you're trying to achieve. Try to understand what it is that you're trying to achieve. The next thing you want to do is collaborate. So now you can go to your AI tool, your large language model, your chatbots, you know, your video creator, whatever it is that you're using AI for, you can use it to expand your ideas. You can add those prompts then and keep on, you know, fine-tuning and iterating on those points and prompts to come up with a proper solution to the problem that you're trying to solve. The fourth thing you need to do, which is very important, is validate the output, validate the work, look at the accuracy, fact-check, proofread, do the work that you really don't want to do. Using AI still saves you time if you use it appropriately. And the last but not the least is iterate. So based on the output of the results that you have seen, you can now put in more prompts. You can work on the body of your work. You can iterate again and improve it so that it comes up with the best results. And you also want to check that, you know, the results of this work that you're doing represents your brand. Your brand represents your tone of voice and actually sounds like a human. So remember, let's try to avoid AI work slog and use it more appropriately.
