San Francisco: Palantir Technologies CEO Alex Karp has sparked debate in the tech world after comparing excessive use of artificial intelligence (AI) tools to addiction, warning that more usage does not automatically translate into more value.
Speaking during a live interview on the sidelines of Palantir’s AIP Con 10 event, Karp used unusually blunt language to criticise what he sees as a growing obsession with maximising AI usage rather than focusing on meaningful outcomes.
“People are just sitting there all day… kind of like an addiction,” Karp remarked, arguing that simply increasing interactions with AI systems does not necessarily improve productivity or business performance.
Shift from AI usage to real value
Karp’s comments reflect a broader shift within the tech industry. During the early phase of the AI boom, companies often encouraged employees to use AI tools as much as possible, treating higher usage — measured in prompts or tokens — as a sign of adoption and progress.
However, that mindset is now being questioned.
Many organisations are finding that rising AI usage is leading to higher operational costs without a corresponding increase in efficiency or output. This has led to a reassessment of how AI success should be measured.
The problem with “tokenmaxxing”
A key issue highlighted in the debate is the concept of “tokenmaxxing” — encouraging heavy consumption of AI tokens to maximise usage metrics.
Karp criticised this approach, suggesting it can lead to wasteful or unproductive interactions with AI systems. Instead of focusing on quantity, he emphasised the importance of applying AI in ways that deliver tangible results.
This perspective is echoed by Shyam Sankar, who has warned that excessive AI usage can generate low-quality outputs, often referred to within the company as “slop”.
“More tokens means more slop,” Sankar said in a recent company discussion, stressing that businesses need systems to filter and refine AI-generated content to extract real value.
Industry concerns over rising AI costs
The concerns raised by Palantir leaders are not isolated. Other major companies are also beginning to question whether increased AI spending is delivering sufficient returns.
Executives at Uber and Amazon have reportedly flagged similar issues. In some cases, companies have even scaled back internal initiatives that encouraged competitive AI usage among employees after noticing inflated usage metrics without meaningful productivity gains.
The shift towards usage-based pricing models for AI services has further amplified the issue, as businesses now face directly measurable costs tied to how much AI they consume.
Where AI adds real value
Karp acknowledged that AI tools, particularly large language models, are highly effective for certain tasks. These include generating reports, summarising information, and handling routine queries.
However, he argued that more complex, real-world challenges — such as supply chain management, industrial operations, and logistics — require structured workflows and human decision-making.
In these scenarios, AI should act as a support system rather than a replacement.
“They are enhanced by large language models. They are not replaced by large language models,” Karp said, highlighting the importance of integrating AI into existing processes rather than relying on it entirely.
Human judgement remains critical
Looking ahead, Karp suggested that many AI capabilities will eventually become widely accessible. What will differentiate successful organisations is not how much AI they use, but how effectively they apply it.
He emphasised the role of human judgement — or what he described as “taste” — in identifying the right problems to solve with AI.
A changing AI narrative
Karp’s remarks signal a turning point in how the tech industry views artificial intelligence. The conversation is gradually moving away from hype-driven metrics like usage and towards more practical considerations such as efficiency, cost-effectiveness, and real-world impact.
As businesses continue to experiment with AI, the focus is likely to remain on achieving measurable outcomes rather than simply increasing usage.