What is an AI Agent? Even a16z Admits Confusion

The Murky Definition of "AI Agent"

Overused terminology, stripped of genuine meaning, has long been a feature of the technology sector. Currently, the term “AI agent” – and its variations like “agentic” – stands out as a prime example.

Lack of Consensus on Meaning

Surprisingly, a clear definition of what constitutes an AI agent remains elusive. Even professionals with strong software engineering backgrounds, employed by leading venture capital firm Andreessen Horowitz (a16z), acknowledge the absence of a universally accepted understanding.

a16z's Attempt at Definition

Guido Appenzeller, Matt Bornstein, and Yoko Li, three infrastructure investment partners at a16z, recently explored the concept during a podcast episode titled “What Is an AI Agent?” They sought to formulate their own working definition.

Significant Investment in AI

a16z has demonstrated substantial commitment to the AI landscape, backing prominent companies like OpenAI and Anysphere (Cursor). Reports indicate the firm is aiming to raise a $20 billion fund for further investment in the sector, as revealed by sources to Reuters. Previously, a16z VCs predicted that “every white-collar role will have an AI copilot,” with some roles being fully automated by AI agents.

The Spectrum of "AI Agents"

Appenzeller notes that a wide range of startups are leveraging the current hype by labeling their products as AI agents. He describes the most basic form as “just a clever prompt on top of some kind of knowledge base.”

From Simple Prompts to Worker Replacements

Initially, these so-called AI agents primarily provided pre-defined responses, such as in IT help desk scenarios. However, more recently, companies have begun positioning them as potential replacements for human workers.

The AGI Requirement

Achieving true human worker replacement, Appenzeller argues, necessitates AI software approaching the level of Artificial General Intelligence (AGI). This requires the ability to “persist over long periods of time” and “work independently on problems.”

Current Limitations

Both Appenzeller and Li concede that such capabilities “doesn’t work yet.” The development of reliable AI agent technology has proven to be a more challenging undertaking than initially anticipated, as highlighted by Jaspar Carmichael-Jack, CEO of Artisan, a sales AI agent company.

Technical Hurdles Remain

Significant technical obstacles must be overcome before AI can genuinely replace human workers. These include establishing persistent long-term memory (and managing associated costs) and mitigating the issue of AI hallucinations. No organization desires an employee – whether human or artificial – prone to inaccurate recall or fabrication.

A Working Definition: Reasoning and Autonomy

During their podcast discussion, the a16z team arrived at a practical definition for current capabilities. Li characterized an AI agent as a reasoning, multi-step Large Language Model (LLM) incorporating a dynamic decision tree.

Beyond Task Completion

She clarified that an AI agent isn’t simply a bot executing tasks on demand. It must also demonstrate the capacity to make independent decisions regarding task execution, such as selecting prospects from a database, determining email recipients, and composing the emails. Similarly, it could write code and autonomously determine its placement.

Automation and Human Roles

Regarding the potential for widespread human replacement, the VCs agreed that AI agents could automate certain tasks currently performed by humans, mirroring the historical impact of automation. However, they also suggested that increased productivity could lead to companies employing *more* human workers, not fewer.

The Importance of Human Creativity

Bornstein expressed skepticism about a future where humans become unnecessary, given the current state of AI agents. He emphasized that the technology industry, particularly from its Silicon Valley perspective, sometimes overlooks the prevalence of jobs requiring human creativity and critical “thinking.” He questioned whether complete human replacement is even “theoretically possible.”

Marketing Hype and Confusion

Bornstein attributes much of the current confusion to the exaggerated claims made by AI agent companies, often driven by marketing strategies, business models, and pricing considerations.

A Reason for Caution

Ultimately, if even those with access to the most advanced AI agent applications remain skeptical of the most ambitious promises, it suggests a healthy dose of caution is warranted for everyone else.