AI Is a Tool: Defining Skill and Agency in the Age of LLMs

The Debate Over AI as a Tool

Whether Large Language Models (LLMs) qualify as tools is a subject of intense debate, centered on the distinction between automation and human-wielded instruments. The argument that AI is not a tool—often attributed to the view that these systems lack the predictability and bounded nature of traditional software—is countered by the perspective that skill, intention, and feedback loops define an instrument's utility. A tool does not cease to be a tool simply because it is complex, unpredictable, or prone to misuse.

The Role of Human Skill and Oversight

Using AI effectively requires a deliberate feedback loop where the human remains the primary agent. Expertise is not rendered obsolete; rather, it is redirected toward knowing what to ask, how to verify outputs, and when to reject or refine AI-generated content. As one perspective notes, the distinction lies between "wielding an instrument with skill and delegating work without adequate understanding or oversight." Without this oversight, users risk falling into "cargo-culting," where they rely on outputs they do not fully comprehend, potentially leading to a long-term erosion of cognitive ability and domain expertise.

The Risks of Anthropomorphic Framing

Critics of the "AI as a tool" framework argue that the industry's use of anthropomorphic language—terms like "go rogue" or "conspire"—creates a false sense of agency. By framing software as an independent entity, we risk obscuring real engineering problems. Technical failures, hallucinations, or unexpected outputs are not evidence of a sentient agent, but rather symptoms of complex, non-deterministic systems that require better guardrails and more rigorous verification processes.

Perspectives from the Community

Discussions among practitioners reveal a divide between those who view AI as a powerful force multiplier and those who see it as a threat to fundamental human skills.

• The Skill Multiplier View: Many users find that AI acts as an "intellectual sparring partner," allowing them to explore complex codebases, challenge their own interpretations, and accelerate learning. In this view, the skill lies in the ability to communicate intent clearly.
• The Obsolescence Concern: Conversely, some argue that the "skill" required to use AI is rapidly diminishing as models become more capable. There is a fear that as models improve, the need for human expertise will shrink, potentially leading to a "decades-long dip in human cognitive ability" as people outsource critical thinking.
• The Agency Problem: A recurring theme is the danger of "outsourcing thinking." As one commenter noted, "It’s a tool until you start outsourcing the thinking to it. Then it’s not a tool, it’s literally the only thing that can work on the codebase."

Conclusion: A New Kind of Instrument

Ultimately, the classification of AI as a tool may be less important than the discipline required to use it. Whether one views AI as a "slot machine" that occasionally yields results or a sophisticated instrument for productivity, the outcome depends on the user's commitment to maintaining control. As the technology evolves, the most future-proof skill remains the ability to define clear objectives, verify results against domain knowledge, and prevent the tool from dictating the workflow.