AI EncyclopediaAgents & protocols
AI Agent
An AI system built from an LLM, a set of tools it can call, and enough autonomy to plan and take a sequence of actions toward a goal - as distinct from a system that only answers a single prompt.
In plain English
A chatbot answers your question. An agent can go do something about it - search a database, call another program, write and run code, check the result, and try again if it failed - stringing several steps together on its own toward a goal you gave it once.
Technical explanation
Anthropic's engineering guidance distinguishes 'workflows,' where multiple LLM calls are orchestrated through predefined code paths, from 'agents,' where the LLM dynamically directs its own process and tool use. Building an agent requires at minimum three components: the underlying LLM, the prompt (including its goal and constraints), and an action space - the tools it is given to act in the world. Anthropic's own experience across many customer deployments found the most successful implementations used simple, composable patterns rather than complex specialised frameworks.
Why it matters
Agentic systems are the current frontier of applied AI - moving from 'answer my question' to 'complete this task' - and are why 2024-2026 has seen a surge of interest in tool use, planning and standards like the Model Context Protocol for connecting agents to real systems safely.
Real-world example
Anthropic's December 2024 guide 'Building Effective Agents' - based on work with dozens of customer teams building production LLM agents - found the highest-performing systems were built from simple, composable patterns rather than heavyweight agent frameworks, a finding its authors say has held up since publication.
Common misunderstanding
That 'agent' has one settled definition industry-wide. In practice the term spans a spectrum: some teams mean a fully autonomous system operating independently over extended periods, others mean a much more constrained, predefined workflow with an LLM step in it - the label alone does not tell you how much autonomy a given system actually has.
Something wrong here?
Every entry is hand-researched and hand-written by Nakoda. If a fact is stale, a source has changed or a definition needs sharpening, tell us and we will check it.