AI and agents
AI agent
An AI agent is software that uses a language model to decide which steps to take toward a goal, calls tools such as APIs to take them, and checks the result before carrying on.
How it works in practice
A fixed automation follows the same path every time. An agent reads the situation, chooses an action from the tools it has been given, looks at what came back and decides what to do next. That makes it useful for work with variation in it: emails that ask for different things, documents in different layouts, requests that need a lookup before anyone can answer.
The tools are what make an agent safe or unsafe. An agent that can only read your order system can give a wrong answer; an agent that can issue refunds can give away money. We scope each tool to the least it needs, log every call, and send anything costly or irreversible to a person for approval. Most agents we put into production run inside an ordinary workflow, with the agent handling the one step that needs judgement.
A worked example
A 30-person wholesaler gets order changes by email. The agent reads each one, finds the order in the ERP, drafts the amendment and posts it to the sales team’s Slack channel. A person approves it with one click, and the agent writes the change and replies to the customer.
Examples in this glossary describe typical builds for companies of 1 to 1,000 people. They are not named client results.















