FAQ
Frequently asked questions
Short answers to the questions we hear most often about implementing AI agents.
What is an AI agent?
An AI agent is AI-based software that performs a specific task: it reads an input, uses the tools and sources it is allowed to, and produces a result or takes the next step. An agent has a defined role, defined tools and rules for when it must ask a person for approval.
How is an AI agent different from a chatbot?
A chatbot answers questions in a conversation. An AI agent carries out the steps of a process — reading documents, producing outputs, using tools — and works as part of a larger workflow. That is why we also test whether an agent does the right work and leaves the wrong work alone.
What is the difference between a skill, an agent, a team and a workflow?
A skill is your team's know-how written so AI follows it, such as how to write a client email in your tone. An agent is a specialist with its own tools and skills. A team is 2–5 agents with a coordinator. A workflow is a complete recurring process, end to end, with your approvals and automated checks.
Which processes can be automated with AI agents?
Recurring work that can be described with rules and examples fits best: customer enquiries, quotes, document review, reports, CRM data, internal knowledge search and document classification. More detailed examples are on the use cases page.
How do you test AI agents?
Every version goes through two independent checks. A reviewer scores it against a weighted rubric (at least 8/10 and no blocking defects to be accepted). A tester runs it in a sandbox with a fresh AI session and measures whether it triggers on the right requests (≥85%) and whether the result meets expectations (≥80%).
How do you reduce the risk of AI hallucinations?
The solution relies on your sources, and every fact in the research phase is referenced. Behaviour tests check that the output meets expectations, and decisions that are yours stay with a person to approve. The risk can't be reduced to zero, but it can be measured and kept under control.
Can a person approve before an action is taken?
Yes. Approval steps can be added anywhere a decision must stay with a person. In our own process, too, you approve the plan before we start building.
Which platform do the solutions run on?
Solutions are packaged for the Claude environment: Claude Code, claude.ai and the API. Installed workflows can be followed on a live monitor.
How can I see what the solution does and what it costs?
The live monitor shows each run's stages, active agents, pending approvals and token usage. Every workflow has a budget; when it is reached, it asks before continuing.
How does working together start?
With a 30-minute intro meeting. If AI is a fit, we write the brief together, you receive a plan with tests and an effort estimate, and we only start building after you approve it.