Agentic systems break a goal into steps, select and use tools such as search or document retrieval, evaluate intermediate results, and adjust their approach, rather than producing a single response to a single prompt.
Autonomy exists on a spectrum, from systems that propose a plan for approval to those that execute end to end.
Alternative Names:
AI Agents, Autonomous AI
Why it Matters?
Agentic approaches suit multi-step litigation work such as gathering material across sources, cross-referencing records, and assembling a draft analysis. They also compound risk, since an error early in a chain propagates through every subsequent step and the final output may look coherent while resting on a flawed intermediate conclusion. Reviewable intermediate steps and approval gates before consequential actions are what keep supervision meaningful.
Frequently Confused with
Related terms
Frequently asked questions
What makes a system agentic rather than just automated?
What is the main risk with agentic workflows?
