Reinforcement learning
Reinforcement learning, also called learning through reinforcement, is a field of machine learning in computer science concerned with how agents act in a dynamic environment to maximize the cumulative reward resulting from their actions. Reinforcement learning is considered one of three machine learning paradigms, alongside supervised learning and unsupervised learning.
On this page
Source content checked . English translation published 2026-10-27.
In business automation
Reinforcement learning trains an agent using rewards from its environment, rather than only labeled examples. It is relatively uncommon in business automation; most agents operate using rules, tools, and a language model, not ongoing reward-based training.
Examples
These examples illustrate a possible workflow. Adapt the connections and rules to your own systems.
Testing recommendations, not collections
A team evaluates a model that suggests a sequence of actions for customer service. The reward is measured by whether a request is resolved correctly in a simulated environment. The model is not connected to collections or price changes without human approval, even if it receives a high score.
Relevant tools
- LangGraph
- You can build an agent with steps and human approval; that is process control, not reinforcement learning training in itself.
What to check
- A faulty reward teaches the agent to shorten conversations or approve discounts in ways the business did not intend.