AI agent questions, answered
Short answers to the questions we hear most often about AI agents and the Model Context Protocol.
Can agents run without any human involvement?
For low-risk, reversible tasks, yes. For anything that changes access, money or customer data, a person should approve. Autonomy should be earned in stages.
What is MCP, in one sentence?
An open standard that lets AI applications use tools and data from your systems through small, well-defined servers, instead of custom code for every integration.
Are agents safe?
As safe as their permissions. An agent with narrow tools, approvals for risky actions and a full audit log is manageable. An agent with broad admin access is not, however good the model.
Which model do agents need?
Agents need models that are good at following instructions and using tools. Larger models usually do better at planning; smaller ones can handle simple, well-defined steps more cheaply.
Will agents replace our service desk?
They take over the routine part: gathering details, diagnosis and standard fixes. People handle exceptions, judgement and anything that needs approval.
How do we start?
Pick one frequent, rule-based process, run the agent in shadow mode for a few weeks, and measure. See running agents.
More agent guides: MCP architecture · Guardrails and approvals · Running agents · Agent FAQ · Use case: IT service desk agent
Thinking about AI agents?
Tell us the process you would like to automate and the systems it touches. We will come back with a plain view of what an agent could safely do, what should stay with people, and how to start small.