Ariath glossary: AI Agent
A software component that uses an AI model to carry out a specific task within a defined context, such as drafting a document or retrieving information from connected systems. It works within the tools, data and permissions it has been given.
In brief
- What it does
- Takes in information, decides on a next step, and acts toward a defined goal.
- How it differs from automation
- It uses context to decide what to do, instead of only following fixed rules.
- What it needs in a wealth management firm
- Defined permissions, approved data connections, human review for sensitive actions, and records of what it did.
How it works
How AI agents work
An AI agent generally operates through a repeating process, where each loop forms the next step:
Receive context
The agent is given a goal or request and the information relevant to it, such as client records, a document or a task description.
Determine the next step
The agent decides what to do next based on the goal, the context and the rules it has been given.
Take action
The agent carries out the step, for example retrieving data, drafting content or passing a request to another system.
Evaluate the result
The agent checks the outcome against the goal, then continues, stops, or hands the work to a person for review.
An agent's performance depends on the quality of its instructions, the information available to it, and the controls governing its actions.
Reliable agents require defined permissions, validation rules, monitoring and escalation paths, not just access to an AI model.
Comparison
AI agents vs. traditional automation
Both act on a firm's behalf. They differ in how each decides what to do next.
Traditional automation
Follows predetermined steps or rules. It does the same thing each time the same condition is met, which makes it predictable but limited to the cases it was written for.
For example
Sends a reminder email whenever a task reaches a certain status.
AI agent
Uses context, such as the contents of a record or a document, to decide on a next step and adjust as the situation changes. It can handle more variation, which also means its actions need defined limits and review.
For example
Reads a new client's documents, notes which required items are missing, and drafts a request for a person to approve.
The two can work together. Automation can handle fixed, repeatable steps, while an AI agent can handle the context-dependent parts, with a person reviewing the output wherever the firm requires it.
In practice
AI agent examples in wealth management
Within a wealth management firm, an AI agent could help coordinate tasks that depend on information from multiple systems. These examples are illustrative.
Meeting preparation
Gathering relevant client details from the CRM, planning and custodial systems ahead of a meeting.
Client onboarding
Checking whether required information has been provided and flagging missing items for review.
Advisor follow-up
Drafting a post-meeting summary and next-step notes for the advisor to review.
Data review
Identifying mismatches, missing fields or duplicate records for staff to review.
Service requests
Classifying incoming requests and routing them to the appropriate team member.
In each case, the agent works within defined permissions. The firm decides what information it can use, which actions it can take, and where a person reviews the result before anything is sent or changed.
What it needs
What AI agents need to work with firm data
An agent that touches client data needs limits around it. Firms typically define what an agent can access, change and share. Safeguards often include:
- Role- or attribute-based access controls
- Approved systems and data connections
- Human review for sensitive actions
- Validation before information is changed
- Monitoring and audit records
- Defined escalation and recovery processes
Together, these controls define what an agent can see and do. They are firm decisions, set by the firm's own policies, and they sit between the firm's data and the AI model or tool.
In Ariath
Ariath documents role-based access checked for the specific user before connected data reaches a model, audit and retention settings configured around each workflow, and governance settings set per workflow.
Common questions
Frequently asked questions about AI agents
Is an AI agent the same as a chatbot?
Not necessarily. A chatbot mainly responds to questions in a conversation. An AI agent can also take actions, such as retrieving data from connected systems or preparing a draft, within the permissions it has been given. Some products combine both.
Does an AI agent have to act without human review?
No. Firms decide where an agent acts on its own and where a person reviews or approves the output. Sensitive actions, such as sending a client communication or changing a record, are commonly placed behind human review.
How can a firm control the data an AI agent uses?
Through access controls tied to the user and workflow, approved data connections, and records of what the agent accessed and did. Which controls a firm sets depends on its own policies, and it should confirm requirements with its compliance lead or counsel.
This entry is educational. It is not legal or compliance advice and does not guarantee regulatory compliance. Confirm the requirements that apply to your firm with your compliance lead or counsel.
Give AI the context it needs to act
Ariath connects your CRM, custodian and planning tools into one governed context, with your firm in control of how data is accessed and used.