"It's undermining system reliability": Why AI workslop is the new technical debt
"Workslop isn’t a minor nuisance. It's spreading through systems faster and more quietly than traditional code problems."
One third of AI-generated code is unusable without significant modification. That's the reality in fast-moving engineering environments.
To describe this level of low-quality, AI-produced work proliferating across modern organisations, Harvard Business Review introduced the term "AI workslop" earlier this year.
As individuals struggle to make sense of AI use within modern business environments, workslop (or AI-generated work that “masquerades as productivity” but lacks meaningful substance) can emerge as malformed code, incomplete test cases, and corrupted data pipelines.
Workslop can also be a result of faulty prompt engineering or poor education around proper practices. Unfortunately, workslop isn’t a minor nuisance, nor is it limited in scope. In fact, it is a new form of pervasive technical debt, and it is spreading through systems faster and more quietly than traditional code problems.
The human cost of AI workslop
Picture this: Developers are being asked to debug AI-generated outputs they did not create and cannot fully trust. Analysts are left validating processes they do not understand. Hallucinations lead to unusable information or recommendations.
For early-career developers, the impact of leveraging untrustworthy AI is particularly acute. They are wasting time wrestling with overcomplicated AI-generated code while also missing the opportunity to hone and refine their own development skills.
Middle managers feel the strain too. They carry the burden of extra oversight, additional review cycles, and increased rework. Time that should be focused on strategic direction is diverted to correcting issues that basic validation frameworks should have already caught.
The morale impact of workslop across technical teams is significant, and if misuse continues to scale alongside the current AI adoption rate, it will only grow worse.
The illusion of speed
The biggest challenge in AI adoption is shifting from speed to discipline. Organisations are deploying AI to generate scripts, configurations, and test logic without the protections that traditional software-quality processes provide. The promise was faster delivery. The reality is hidden technical debt that spreads through systems and erodes trust in automation.
AI-generated test cases often fail to run or execute cleanly, failing to offer meaningful coverage. Similarly, generated code may pass an initial review and then break under real conditions because the underlying logic is unsound. The illusion of speed often comes at the cost of stability and trust. No CIO can afford that trade-off.
What we are witnessing is a mismatch between adoption velocity and implementation maturity. Organisations are racing up the AI adoption curve faster than teams are being trained to use it properly. The result is low-quality work being produced, and impeding business, at scale.
Two critical factors to combat workslop
Preventing AI workslop requires getting two fundamentals right: choosing the appropriate tool for the job and applying proper prompt engineering discipline.
The first problem is misapplication. Large language models (LLMs) are powerful, but excitement around AI has encouraged teams to use them everywhere. They are being applied to tasks that would be better handled by deterministic logic, traditional automation, or well-defined rule-based systems. We are misusing the technology too often, and that directly contributes to workslop.
Second, prompting matters. Most teams have never received formal training in prompt engineering. Formulating direct and applicable questions leads to more discerning and targeted responses.
READ MORE: "There’s no micromanagement on the battlefield": What leaders can learn from the British Army
Effective prompting requires structure, clarity and understanding of how these systems will interpret instructions. Ongoing employee education is a core component of combating the workslop problem, and prompt literacy is fast becoming a core technical skill.
This gap between organisational expectations and team capability is where AI workslop thrives. Across industries, organisations are encountering outputs filled with hallucinations, inaccuracies and omissions that proper human review would have caught. It is the consequence of prioritising speed over quality.
Organisations have to invest in the fundamentals – not just in widespread prompt engineering education, but also output validation, hallucination detection, and in helping employees discern when and where traditional methods might outperform AI – to fuel long-term AI-enabled growth.
Fixing this requires more than hype management
Training teams in prompt engineering reduces errors at the source. Teaching hallucination recognition enables faster identification of issues. Building validation frameworks prevents unreliable outputs from reaching production.
Most importantly, smaller models can help humans realise more specific objectives and outcomes even faster – and keeping humans on the loop is essential to validate, refine, and make the most of smaller AI use cases. But all of these require a more intentional and discerning approach to AI use across the board.
The way forward requires honesty about what AI can and cannot do, discipline in how it is implemented and a commitment to quality that may require slowing down to go faster later. This can be difficult when competitors appear to be moving rapidly, but it is crucial for sustainable AI adoption.
Training teams in prompting fundamentals and teaching developers how to determine when traditional approaches are warranted over AI use are tablestakes. Output validation should be mandatory at every stage. And designing frameworks that catch AI-generated issues before deployment are critical to curb unnecessary risks.
READ MORE: Security is tired of alert fatigue: Will AI finally let SOCs get some well-earned rest?
Organisations are finding themselves at a turning point. They can continue accelerating AI adoption without proper frameworks and watch technical debt compound, junior developers miss critical learning experiences, and teams lose morale as debugging AI outputs becomes their primary task. Or they can slow down, implement engineering discipline, leverage AI thoughtfully, and train teams properly.
The rush to adopt AI has outpaced our understanding of how to use it well. That gap is what creates AI workslop. Closing it requires education, validation, and discipline. These are not constraints on innovation. They are the foundation that will make future innovation sustainable.
Recognising and remediating AI workslop before it undermines system reliability and team confidence is not about resisting AI. It is about ensuring that adoption is grounded in engineering principles, trust, and intentional use rather than hype. Organisations that take this deliberate, disciplined approach will be able to move faster later with quality and trust intact.
Joel Carusone is the Senior Vice President of Data and Artificial Intelligence at NinjaOne.
