Introduction to Agents
The term "agent" is being used to describe a wide range of technologies, from simple scripts to sophisticated AI workflows. However, there is no shared definition of what an agent is, which can lead to confusion and disappointment for customers. Companies are marketing basic automation as advanced AI, which doesn’t just confuse customers but also invites disappointment.
The Problem with Agents
We don’t necessarily need a rigid standard for agents, but we do need clearer expectations about what these systems are supposed to do, how autonomously they operate, and how reliably they perform. Most of today’s agents are powered by large language models (LLMs), which generate probabilistic responses. These systems are powerful, but they’re also unpredictable. They can make things up, go off track, or fail in subtle ways—especially when they’re asked to complete multistep tasks, pulling in external tools and chaining LLM responses together.
Reliability Issues
A recent example of the reliability issues with agents is the case of Cursor, a popular AI programming assistant. Users were told by an automated support agent that they couldn’t use the software on more than one device, but it turned out that the policy didn’t exist. The AI had invented it. This kind of mistake could create immense damage in enterprise settings. We need to stop treating LLMs as standalone products and start building complete systems around them—systems that account for uncertainty, monitor outputs, manage costs, and layer in guardrails for safety and accuracy.
Building Complete Systems
Some companies are already moving in the direction of building complete systems around LLMs. For example, AI21 has launched Maestro, a system designed for enterprise reliability that combines LLMs with company data, public information, and other tools to ensure dependable outputs. These measures can help ensure that the output adheres to the requirements expressed by the user, obeys the company’s policies regarding access to information, respects privacy issues, and so on.
Cooperation Between Agents
For the agent model to work, different agents need to cooperate without constant human supervision. Google’s A2A protocol is meant to be a universal language that lets agents share what they can do and divide up tasks. However, A2A still falls short. It defines how agents talk to each other, but not what they actually mean. Without a shared vocabulary or context, coordination becomes brittle. We’ve seen this problem before in distributed computing, and solving it at scale is far from trivial.
Conclusion
In conclusion, while agents have the potential to revolutionize the way we work and interact with technology, there are still significant challenges to overcome. We need to develop clearer expectations for what agents can do, build complete systems around LLMs, and develop protocols that allow agents to cooperate effectively. With continued innovation and development, agents can become a powerful tool for improving productivity and efficiency.
FAQs
- Q: What is an agent?
A: An agent is a technology that can perform tasks autonomously, but the term is often used to describe a wide range of technologies, from simple scripts to sophisticated AI workflows. - Q: What are the challenges with agents?
A: The challenges with agents include reliability issues, lack of clarity around what they can do, and the need for cooperation between agents. - Q: How can we overcome the challenges with agents?
A: We can overcome the challenges with agents by developing clearer expectations for what they can do, building complete systems around LLMs, and developing protocols that allow agents to cooperate effectively. - Q: What is the A2A protocol?
A: The A2A protocol is a universal language that lets agents share what they can do and divide up tasks, but it still falls short in terms of defining what agents actually mean. - Q: What is the future of agents?
A: The future of agents is promising, but it will require continued innovation and development to overcome the challenges and realize their potential.
