The Great Category Error: Why Rich Hickey Says Your AI Agent Isn't a Tool

The Semantic Trap of the 'AI Tool' Moniker

In the current gold rush for enterprise automation, the word 'tool' has been stripped of its technical rigor. Rich Hickey, the creator of Clojure, argues that we have allowed the term to become a platitudinous stand-in for 'anything useful,' effectively masking a dangerous shift in professional agency. While industry players celebrate the rise of autonomous AI SEO, Hickey’s framework suggests we are trading professional mastery for black-box delegation.

'Tools are implements with which you do something, they do not do things for you.'

By labeling an LLM agent as a 'tool,' we imply a level of control and predictability that simply does not exist. We are not using these systems to extend our reach; we are using them to outsource our cognition. This linguistic sleight-of-hand allows organizations to ignore the reality that they are no longer the primary operators of their own workflows.

Deterministic Reliability vs. The Illusion of Competence

True tools are defined by their predictability and bounded function, providing the sensory feedback necessary to build human expertise. LLMs, by contrast, are probabilistic engines that offer an illusion of competence while actively eroding the user's ability to verify the output. The inability to audit AI output, which often defies human verification, confirms Hickey’s suspicion that these systems lack the reliability required of a true tool.

When we rely on these agents, we stop 'trying'—the very root of the word expertise. We become passive observers of a process we no longer fully understand or control.

The Exploitation Architecture of Silicon Valley

If we strip away the marketing veneer, Hickey suggests a much darker definition of 'tool' in the current tech landscape: anyone or anything that can be exploited. The AI titans have successfully rebranded their entire ecosystem to ensure that every participant—from the VC funding the next 'disruptor' to the developer writing the training scripts—is merely a component in their own expansion.

In this architecture, the following entities are being treated as 'tools' for the benefit of AI titans:

• VCs seeking the next high-growth exit.
• Politicians looking for regulatory leverage.
• Wannabe-titan-CEOs chasing market valuation.
• Software developers building on unstable, proprietary APIs.
• Rural towns providing the physical infrastructure for data centers.

This is not a collaborative ecosystem; it is an extractive one. By mislabeling these relationships as 'tooling,' the industry obscures the power dynamics at play.

Reclaiming Expertise in an Era of Delegated Thought

The long-term consequence of this shift is the atrophy of human technique. When we replace sensory feedback and iterative practice with AI-generated shortcuts, we lose the ability to troubleshoot, refine, and innovate. As firms navigate the current infrastructure pivot, they must decide whether to build systems that augment human skill or those that merely replace it.

True progress requires us to reject the convenience of the 'agent' in favor of the precision of the 'tool.' We must demand systems that are deterministic, auditable, and subservient to human intent. Anything less is not a tool—it is a surrender of the very expertise that defines professional excellence.