By Lina Irawan 
Among the most prominent voices in this reevaluation is Tobias Lütke, the cofounder and CEO of e-commerce giant Shopify. Only last year, Lütke was a vocal proponent of AI integration, publicly stating to his employees that using AI was "a baseline expectation." He urged staff to first prove they "cannot get what they want done using AI" before requesting additional resources, effectively making AI utilization a prerequisite for efficient work. This directive underscored a corporate culture keen on maximizing AI’s potential at every turn. However, Lütke’s current stance reflects a growing disillusionment. In a recent interview on The Knowledge Project podcast, he expressed frustration over a new problem: employees producing unexamined emails and code generated by AI without taking personal responsibility for the often-sloppy results.
Lütke colorfully referred to these subpar, AI-generated deliverables as "slop grenades" that employees "toss at each other." He unequivocally labeled this phenomenon as "definitely a bad thing." His concern stems from the ease with which AI can be allowed to "go nuts," generating copious amounts of text or code that, while appearing coherent on the surface, ultimately creates more headaches for those who receive and are tasked with reviewing the work. He provided a clear example: AI, in his view, should aid in synthesizing key points for an email, making communication more concise and effective. Instead, it’s being used to generate "a big missive" – a lengthy, often verbose communication that wastes the recipient’s time. "You don’t really read it, and now it has to be reviewed by your colleagues, and they are like, ‘this doesn’t look right’," Lütke lamented, pinpointing the core issue: "You’re just letting AI do the work for you." This suggests a delegation of critical thought and responsibility, rather than an augmentation of human capabilities. The essence of the problem, as Lütke articulates, is the transfer of cognitive load and quality assurance from the individual to the recipient, leading to a net loss in organizational efficiency despite the superficial appearance of accelerated output.
Shopify’s experience is not isolated. Luis von Ahn, CEO of the language-learning platform Duolingo, has similarly backtracked on his initial, aggressive embrace of AI. Last year, Duolingo announced an ambitious "AI-first" strategy. This meant not only evaluating employees on their AI usage but also actively replacing human contractors with AI solutions and only considering an increase in human headcount if a team demonstrably failed to automate the required work. The company’s vision was to leverage AI for scale and efficiency across its content creation and operational processes. However, in May, von Ahn admitted to Fast Company that he had gotten "carried away" by how well AI performed in initial writing demonstrations. He found that while AI could generate content, it ultimately couldn’t match the unique creativity and nuanced understanding of Duolingo’s human employees, especially when scaled. "We may need to write 1,000 different stories for people to learn a language, then you’ll find that 20% of the things were just pure slop," he explained. This realization highlighted a critical distinction: AI excels at generating content, but consistency in quality, creativity, and contextual appropriateness remains a significant challenge, necessitating rigorous human oversight. Von Ahn concluded, "Whenever we scale a lot [of] things with AI, we have to really be careful that slop doesn’t get through."
The concerns articulated by Lütke and von Ahn are indicative of a broader phenomenon that researchers have now formally termed "workslop." This term encapsulates the issue of polished-looking AI output that, despite its superficial presentation, ultimately drags down overall productivity because it necessitates extensive revision, correction, or outright rejection. It’s a deceptive form of inefficiency, where the speed of generation is offset by the time required for human intervention to ensure quality and accuracy.
A recent comprehensive study by BetterUp Labs and Stanford’s Social Media Lab shed light on the pervasive nature and detrimental impact of workslop. Their survey, conducted this year among 962 American full-time desk workers, revealed that over half (52.7%) reported sending workslop to colleagues. Alarmingly, the prevalence of sending workslop was significantly higher in organizations that actively encouraged or mandated AI use, suggesting a direct correlation between adoption strategies and the problem’s escalation. The receiving end was equally affected, with over a third (38%) of respondents reporting that they had received workslop. These recipients estimated that dealing with and revising this low-quality, AI-generated output cost them, on average, 3.4 hours per month.
What truly differentiates workslop from conventional low-quality human work is its insidious nature: it often looks legitimate and well-structured on the surface, masking its underlying flaws. This deceptive quality makes it harder to immediately detect and leads to more time wasted in review. Survey respondents provided vivid examples of workslop, including meticulously formatted emails that contained broken or incorrect links, or code snippets that, while syntactically correct, were unnecessarily complex, inefficient, or contained logical errors that would only surface during testing or integration. These examples underscore the fact that AI can mimic form without fully grasping function or context, leading to output that is technically plausible but practically useless or even harmful.
Beyond the direct time cost, the study also uncovered significant relational impacts. The perception of competence and trustworthiness among colleagues takes a hit when workslop is involved. Employees reported viewing colleagues who sent workslop as less competent and less friendly. This erosion of professional regard can have serious implications for team cohesion and collaboration. Furthermore, the survey found that among those who had received workslop, over a third (36%) reported actively wanting to avoid working with those colleagues in the future. This suggests that workslop isn’t just a productivity drain; it’s a relationship inhibitor, capable of fostering resentment and undermining team dynamics.
The problem of workslop appears to be intensifying. The 3.4 hours per month spent on cleanup in the current survey represents a notable increase from last year’s findings, when 40% of respondents reported encountering workslop and spent an average of two hours per month reworking it. This trend indicates that as AI adoption grows, so does the burden of managing its imperfect outputs. The financial implications are substantial: the researchers calculated that the time spent revising workslop translates to approximately $186 per month for a single employee. Scaled up, this lost productivity could cost an organization with 10,000 employees as much as $9 million annually. Such figures highlight that while AI promises immense efficiency gains, its misuse or uncritical application can lead to significant hidden costs that undermine the very productivity it aims to enhance.
The emergence of "workslop" compels a re-evaluation of AI integration strategies. The initial impulse to simply "use AI" across the board, without clear guidelines, training, or a culture of accountability, appears to be yielding diminishing returns and creating new forms of inefficiency. Tech leaders are now recognizing that AI, while powerful, is a tool that requires human discernment, critical thinking, and responsibility. The shift from an "AI-first" mantra to a "human-in-the-loop" approach is becoming increasingly vital. This involves leveraging AI as an intelligent assistant or co-pilot, designed to augment human capabilities, automate mundane tasks, and accelerate initial drafts, but always with the expectation that human oversight, expertise, and a final stamp of approval are non-negotiable. The goal is not to eliminate human effort but to reallocate it towards higher-value tasks, critical review, and creative refinement, ensuring that the promise of AI-driven productivity does not devolve into a flood of unexamined "slop grenades" that ultimately hinder progress.
