LLM agents: buzzword or innovation breakthrough?

In the rapidly evolving AI landscape, few terms generate as much confusion—or excitement—as "AI agents." These systems promise to transform how we interact with technology, but defining exactly what constitutes an agent remains surprisingly difficult, even among experts.

Key Points:

• Definition spectrum: At its simplest, an agent could be just a chat interface with an LLM. At its most complex, an agent approaches AGI with abilities to persist, learn, and work independently on problems.

• Technical architecture: Most experts agree that agents typically involve an LLM running in a loop with tool use capabilities, making dynamic decisions through a form of planning.

• Value proposition: While often marketed as human replacements, current agents typically augment rather than replace humans—complete automation remains elusive for tasks requiring genuine creativity and nuanced decision-making.

The Identity Crisis of AI Agents

The challenge in defining AI agents reflects a fundamental tension in the field. As Matt Bornstein of Andreessen Horowitz points out, "Agent is just a word for AI applications… anything that uses AI kind of can be an agent." This definitional ambiguity isn't merely academic—it has real implications for how products are developed, marketed, and priced.

When we look beneath the marketing hype, what distinguishes agents is their ability to perform complex planning and interact with external systems. The problem? Modern LLMs already do both these things to varying degrees. The line between a sophisticated prompt and an "agent" has become increasingly blurred as foundation models grow more capable.

What makes this particularly interesting is the gap between current capabilities and future potential. Andrej Karpathy, formerly of OpenAI, has suggested that truly useful AI agents represent "a decade problem"—while much of what we see today consists of "weekend demo" implementations that generate confusion through overpromising.

The Economics of Digital Labor

Perhaps the most revealing aspect of the agent conversation is how it's reshaping business models. Some startups pitch their AI agents using value-based pricing: "This replaces a $50,000/year employee, so we'll charge you $30,000/year." This approach might seem compelling initially, but economics suggests that prices eventually converge toward the marginal cost of production.