"AI Psychosis" and the Labor Market

Box founder Aaron Levie recently introduced the concept of "AI psychosis" to describe a concerning dynamic in today's technological landscape. According to Levie, those making strategic decisions about AI adoption and its ability to replace certain job functions are often the least qualified to understand the true nature and complexity of those roles. This disconnect can lead to misjudgments and hasty implementations, with significant consequences for the labor market.

A concrete example of this trend was observed with ClickUp, which recently cut 22% of its workforce, attributing the decision to the introduction of AI agents. This episode is part of a broader pattern of tech layoffs, with projections for 2026 already approaching the total personnel reductions recorded throughout 2025, highlighting a phase of profound transformation and uncertainty.

The Impact of Automation and Technological Challenges

The adoption of Large Language Models (LLMs) and AI agents promises efficiency and innovation, but their large-scale implementation presents significant technical challenges. Replacing complex human tasks with automated systems requires not only a deep understanding of the tasks to be automated but also a robust technological infrastructure. This includes the ability to handle intensive inference workloads, the need for continuous fine-tuning of models, and integration with existing systems.

For companies evaluating the deployment of AI solutions, it is crucial to consider hardware requirements, such as GPU VRAM for running large LLMs, and the latency needed to ensure adequate throughput. The choice between an on-premise deployment and cloud-based solutions is not trivial and depends on factors such as data sovereignty, compliance requirements, and direct control over the infrastructure, which are central for those operating in air-gapped environments or with sensitive data.

Evaluating TCO and Data Sovereignty in AI Deployments

The decision to adopt AI agents to replace human roles must be supported by a rigorous analysis of the Total Cost of Ownership (TCO). This includes not only initial costs for software licenses and hardware (such as servers with high-performance GPUs) but also long-term operational costs, such as energy consumption, infrastructure maintenance, and specialized personnel for model management and optimization. A self-hosted deployment can offer greater control and data security but often entails higher CapEx and the need for specific in-house expertise.

Data sovereignty and regulatory compliance (e.g., GDPR) are crucial factors driving many organizations towards on-premise solutions. Keeping data within their own infrastructural boundaries ensures greater control and reduces the risks associated with data transfer and management by third parties. For those evaluating on-premise deployments, AI-RADAR offers analytical frameworks on /llm-onpremise to understand the trade-offs between control, costs, and performance, helping to navigate the complexities of these strategic choices.

Future Prospects and the Need for a Balanced Approach

The phenomenon of "AI psychosis" highlights the tension between the transformative potential of artificial intelligence and its practical implementation in the workplace. While AI can automate repetitive tasks and increase efficiency, it is essential that deployment decisions are based on a realistic understanding of AI capabilities and human needs. Enthusiasm for automation should not overshadow the need for a thorough analysis of work processes and ethical and social implications.

Companies that adopt a more measured and informed approach, focusing on augmenting human capabilities rather than mere replacement, will likely reap the greatest long-term benefits. This requires investment not only in technology but also in personnel training and the creation of new synergies between human and artificial intelligence, while mitigating the negative impact on the labor market, as suggested by current tech layoff trends.