An AI agent needs a clear job. A shopping assistant should help someone find the right product. A service desk agent should help diagnose a problem. An access request assistant should understand what needs approval, not assume it can grant access.
That is the idea behind this collection in Autom Mate AI Agent Composer: twelve distinct roles, managed in one place.
We created these examples in our presales environment to make the possibilities easier to see. They are deliberately minimal demo configurations, not twelve production deployments. Each has a name, a model selection and a short role prompt. We have not connected knowledge sources, action skills or channels, and we have not run them.
Real product screenshot. The catalog shows configuration examples, not live customer workloads.
Start with the job
It is tempting to start an AI project by choosing a model. I would start with a more practical question: what should this agent be responsible for, and where should its responsibility end?
For our L1 Ticket Resolution example, the intended job is to guide a user through approved troubleshooting and recognise when a human needs to take over. The name alone does not make it a working resolution agent. It still needs the right knowledge, permitted actions and testing before anyone should rely on it.
The same principle applies across the collection:
| Demo role | Intended use case once configured and validated |
|---|---|
| Shopping Assistant | Help shoppers navigate an approved product catalog |
| Website Concierge | Answer visitor questions and route enquiries |
| L1 Ticket Resolution | Guide first-line troubleshooting and escalation |
| Ticket Triage | Recommend a category, priority and destination queue |
| Employee Onboarding | Guide a new-starter checklist and its approvals |
| Employee Offboarding | Prepare a departure checklist for authorised review |
| Access Request Assistant | Clarify an access request and required approvals |
| CMDB Data Steward | Identify incomplete or inconsistent asset information |
| Invoice Intake | Extract invoice information for human review |
| Order Status Assistant | Answer questions using an approved order source |
| Renewal Follow-up | Prepare reminders for review before customer contact |
| Knowledge Assistant | Answer questions grounded in approved content |
These are starting points for implementation, not a claim that every integration is already configured or that these agents collaborate autonomously.
Give the agent the right context
Composer separates the agent’s role and model configuration from its knowledge sources. In the Knowledge section, the interface offers files, public websites, applications and text.
That separation matters. A well-written prompt cannot substitute for accurate product information, a maintained support knowledge base or an authorised source of order data. Before adding content, we should decide what the agent may use, who owns that content and how it will stay current.
The demo has no knowledge sources attached. The screenshot shows the available configuration categories.
Decide what it may do
Composer also provides a Skills section for configuring actions. We left it empty for this showcase.
In a real implementation, this is where the conversation must move from an attractive answer to a carefully scoped action. Reading an order status is different from changing an order. Suggesting an access request is different from granting permission. The workflows, credentials and approvals behind those actions need to reflect that difference.
A useful first release might only answer questions and prepare a handoff. More actions can follow when the team has tested the behaviour and agreed the boundaries. A role prompt is useful guidance; it is not a substitute for enforceable access controls.
Choose where people will use it
The Channels section presents WhatsApp, Microsoft Teams, Mate Chat and Webhook options. Those options require their own setup; none is enabled in these examples.
The intended audience should drive that choice. A website assistant and an internal service desk agent may need different channels even when they follow the same configuration approach.
Channel options shown in the product. No channel was connected for this demo.
Take one role from example to production
The goal is not to collect the largest number of agent cards. It is to make each role useful and accountable.
For a first pilot, I would choose one recurring request, connect the minimum approved knowledge, add only the actions it genuinely needs and test both successful and unsuccessful cases. Define when a person takes over. Decide what success means before launch: a correct answer, a useful handoff or a verified completed task.
Composer brings the configuration areas together, including Playground and Logs tabs. Moving a demo into production still requires implementation, testing and ownership. The screenshots here show the starting point, not the end of that work.
For me, that is the interesting part of an AI agent team: different responsibilities, made explicit, rather than one assistant expected to do everything.
Which role would you put to work first in your organisation?
Get in touch and let us explore it in Autom Mate AI Agent Composer.
