Strategic call An agent in n8n is a workflow with a brain. It combines an AI model with memory and tools, such as your CRM, your inbox or other workflows, and decides for itself which tool to use for the task in front of it. You can start one from a chat, a form, a schedule, a webhook or an AI assistant such as Claude through n8n’s MCP Server.
That flexibility is exactly why identity matters. Today, most agents borrow someone’s login. Everything the agent does looks like that person did it, the agent inherits all of that person’s permissions, and when the person leaves, the agent either breaks or keeps access it should not have. n8n is fixing that with three ways an automation can act.
1. Agents become identities
An agent gets an identity of its own, just like a new colleague gets their own account on day one. Its actions are attributed to the agent, and its permissions and credentials are managed separately from those of people.
Think of an invoice agent that runs every morning and sends payment reminders. Nobody triggers it, so it should not run as anyone. With its own identity, it gets exactly the access it needs for that job, every reminder in the log says “invoice agent”, and you can switch it off without touching anyone’s account.
2. Agents act on behalf of a person
Sometimes an agent works for someone. A sales rep asks an agent to update a deal, write a follow-up and book a meeting. The agent does the work, but it does it for that person.
In that case n8n records both: who acted, and for whom. The log does not just say “Sales agent” or “Anna”. It says the sales agent acted on behalf of Anna. That is the difference between an audit trail that raises questions and one that answers them.
3. Automations run with your identity
When a person calls a workflow or an agent, it runs with that person’s own accounts. Their permissions, tools and access are respected, so nobody can do more through an automation than they could do themselves.
This part is already here: end-user credentials are available in preview on n8n’s Enterprise plan, on the MCP Server, Form and hosted Chat triggers.
Which one fits when
- Agent identity: an agent sends invoice reminders every morning, or sorts and answers support tickets around the clock.
- On behalf of a person: a sales agent updates a deal and books a meeting when a rep asks, or an assistant prepares a report for a manager from the team’s data.
- Your identity: a colleague asks Claude, through n8n, to search their own Google Drive, or a form request is sent from the requester’s own mailbox.
A simple rule: if nobody starts it, the agent acts as itself. If someone starts it and the agent does work for them, it acts on their behalf. If the person should simply use their own access, the automation runs with their identity.
Treat agents like colleagues
The shift behind all of this is that agents are becoming part of your team. That means they deserve the same care you give a new employee:
- Onboarding. Every agent gets its own identity and only the permissions its job needs.
- An owner. Every agent has a named person who is responsible for what it does.
- Limits. Spending and usage caps, so an agent that loops cannot run up costs unnoticed.
- Approval. A person signs off before irreversible actions such as paying, deleting or sending to customers.
- A record. Every action is logged with who acted and, where relevant, for whom.
- Offboarding. When an agent is retired, its access goes with it.
This is the access and guardrails part of our governance framework, applied to agents. It also makes agents easier to explain to your clients and auditors, and to keep in line with the GDPR when personal data is involved.
What you can do today
Native agent identity is on its way, but you do not have to wait to work this way:
- Stop using personal logins for agents. Give every agent its own credentials, ideally a dedicated service account per agent or per job.
- Use end-user credentials where a person triggers the run and should act as themselves.
- Keep tools narrow. An agent only gets the tools it needs, not everything you have connected.
- Name an owner and add alerts, so a failing agent reaches a person before it reaches a customer.
- Log who asked. Until n8n records it for you, store the requesting person with every agent action.
How we help
Viralistic builds AI agents in n8n for real business processes, from sales and support to finance and operations. As an official n8n Ambassador, Marrallisa follows n8n’s identity and governance roadmap closely, so the agents we build today are ready for what is coming. We decide with you which agents act as themselves, which act on behalf of your people and which run with each person’s own access, and we set up the credentials, limits and logging to match. See our automation service, read how we govern automations, or catch up on n8n’s new engine and governance roadmap.
Frequently asked questions
What is an AI agent in n8n?
A workflow that combines an AI model with memory and tools, and decides for itself which tool to use for a task. Tools can be apps such as your CRM or inbox, other n8n workflows or services connected through MCP.
What does agent identity mean?
The agent has an identity of its own, separate from people. Its actions are attributed to the agent, and its permissions and credentials are managed on their own, so you always know what the agent did and what it was allowed to do.
What is the difference between acting on behalf of a person and end-user credentials?
With end-user credentials, the automation runs with the person’s own accounts, as if the person did it. When an agent acts on behalf of a person, the agent keeps its own identity, and n8n records both who acted and for whom.
Can I use agent identity in n8n today?
Agent identity and acting on behalf of a person are on n8n’s roadmap. End-user credentials, where automations run with the identity of the person who calls them, are available in preview on the Enterprise plan. Until the rest arrives, give each agent its own credentials instead of a personal login.