What makes a system agentic?
“What makes an AI system 'agentic'?”
What this tests
- Whether the candidate has a precise definition beyond "it uses an LLM"
- Understanding of the observe → reason → act → evaluate loop
- Awareness that autonomy is a spectrum, not a binary
- Whether they connect agency to cost, risk, and the need for control
Answers by level
Read the beginner answer first and notice what is missing.
A system is agentic when the model decides the control flow at runtime. In a plain LLM call or a fixed workflow, the code decides what happens next; in an agent, the model observes the current state, chooses an action (usually a tool call), receives the result, and repeats until it judges the goal met or a limit stops it. The distinguishing property is that the sequence of steps is not known before execution. See The Agent Loop.
That definition puts agency on a spectrum. A single call with structured output has zero agency. A workflow with one LLM-driven branch has a little. A loop where the model picks among 20 tools and decides when to stop has a lot. Each step up buys flexibility and pays for it in unpredictability, latency, cost, and harder testing, which is why Choosing the Right Abstraction matters more than the word "agent".
In practice, agentic systems therefore need things a chatbot does not: termination conditions, budgets, tool permissions, and traces. A senior engineer describes an agent by its loop, its tools, and its stopping rules, not by its framework.
Green flags · Red flags
- Defines agency as the model controlling the control flow / choosing actions at runtime
- Describes the loop: observe, reason, act, evaluate, repeat, with a stopping rule
- Treats agency as a spectrum from single call to multi-agent
- Mentions that agents need budgets, limits, and traces that a chat completion does not
- Says most problems do not need an agent
- Distinguishes tool use from agency (a workflow can call tools without being agentic)
- Defines agentic as "uses tools" or "uses a framework"
- Treats agent and chatbot as synonyms
- No mention of termination, cost, or unpredictability
- Presents more autonomy as inherently better