Agent

An AI agent is a system that plans and executes multi-step tasks, often using tools, memory, and feedback loops beyond a single prompt-response.

Software that can decide next steps and use tools to finish a job, not only chat once.

Agents combine planning, tool interfaces, memory, and control loops. Reliability depends on tool sandboxing, evals, and human-in-the-loop design—not anthropomorphic branding.

Agent products promise multi-step work. Buyers need clear boundaries between chatbots, workflows, and autonomous agents.

• Multi-step research assistants
• Ops workflows with tool calls
• Developer agents under supervision

• Can complete tasks requiring multiple tools
• Reduces manual glue work when reliable

• Error compounding across steps
• Security risks from tool access
• Harder to evaluate than single-turn chat

• Calling something an agent does not make it autonomous or safe
• Agents still need permissions and audit logs

How does Brel tag AI Agents?

When company materials emphasize multi-step tool-using systems beyond a single prompt-response product.

• NIST AI RMF

• Survey literature on LLM-based agents (arXiv)