Glossary
Agentic Workflows
Multi-step AI workflows where the model selects tools, makes decisions, and chains actions — rather than producing a single response from a single prompt.
An agentic workflow is an AI system where the model is given access to tools (function calls, API endpoints, retrieval over a knowledge base) and is allowed to choose which to invoke, in what order, to accomplish a task. The model produces structured tool calls; an orchestrator executes them; the results are returned to the model; the loop continues until the task is complete.
Agentic patterns range from simple — a single tool call to look up a fact before answering — to elaborate, with planning steps, multiple specialized agents, and recovery loops. The main appeal is that they handle tasks that a single prompt cannot: multi-step reasoning, fresh data lookup, document-grounded calculations, structured output generation.
In regulated industries, agentic workflows raise the stakes on auditability. Every tool call is a security boundary. The audit log must capture not just the final answer, but each tool the agent invoked, each parameter it passed, each piece of data it received, and the full reasoning chain. Without that, you cannot reconstruct what happened — and "we cannot reconstruct what happened" is not a defensible posture in front of an auditor.
The design principles we apply: minimal tools, narrow scopes, deterministic where possible, explicit retry and abort conditions, and a structured trace stored alongside the conversation log. Human-in-the-loop checkpoints at key decision points. Hard caps on agent depth and runtime to prevent runaway behavior.
See also
Related terms
RAG (Retrieval-Augmented Generation)
An LLM pattern that retrieves relevant documents at query time and feeds them to the model as context, instead of relying on the model's training data alone.
Audit Logging for AI
Recording every model invocation — the requesting user, the input, the retrieved context, the model's output, and the tools it called — in a tamper-evident log.
Human-in-the-Loop
An AI system design where humans review or approve specific model outputs before they take effect — the standard pattern for any AI decision with regulatory or safety consequences.