The definition that matters

An AI agent is software that monitors a situation, applies logic, and takes action — without waiting for a human to tell it to start.

That's different from an AI assistant (which waits for a question) and different from a basic automation (which fires a fixed sequence when a trigger occurs). An agent has judgment. It reads data, decides what matters, and acts accordingly.

For an advisory firm, the practical meaning is this: an AI agent can watch 500 client portfolios overnight, identify the three that need rebalancing attention before market open, prepare a brief for each, and route them to the right advisor — before anyone in the office turned on their computer.

That's not a chatbot. That's an analyst who doesn't sleep.

What agents do vs. what chatbots do

The distinction matters because advisory firms are being sold both as interchangeable. They're not.

A chatbot:

• Responds when someone asks it something

• Works with the information provided in the conversation

• Produces text output (an answer, a draft, a summary)

• Requires a person to initiate every interaction

An AI agent:

• Acts on a schedule or when a trigger event occurs

• Accesses live data from your firm's systems (portfolio, CRM, custodian feeds)

• Produces actions, not just answers (alerts routed, reports drafted, tasks created)

• Operates without a person starting every workflow

The test: when the markets close on Friday, does your AI tool have any idea what happened in your clients' portfolios? If it only knows what you told it, that's a chatbot. If it already pulled the data, ran the analysis, and queued the Monday morning briefings — that's an agent.

Specific tasks Navigator AI agents handle today

These are not hypotheticals. These are the workflows advisory firms run on Milemarker's Navigator AI.

Portfolio monitoring and drift detection
Navigator scans every account against target allocations daily. When an account drifts past the firm's defined threshold — or a model changes and existing accounts need to catch up — an alert fires to the responsible advisor with the specific account, the current position, the target, and a proposed action. The advisor makes the call. The agent did the watching.

Pre-meeting briefings
Fifteen minutes before a scheduled client meeting, Navigator pulls 12 months of account activity, calculates performance relative to benchmark, surfaces recent household events (new accounts, RMDs taken, beneficiary changes, service requests resolved), and generates a one-page brief. The advisor walks into the meeting informed, not scrambling.

New relationship analysis
When a new client comes on board, Navigator analyzes the transferred assets: maps positions to Milemarker's normalized data schema, identifies concentration risk, flags tax lots with embedded gains, and flags positions that may conflict with the firm's model. The advisor gets a structured analysis before the first planning conversation, not after.

Compliance event logging
When a trade, transfer, or model change crosses a threshold defined in the firm's compliance ruleset, Navigator creates a log entry, routes it to the compliance officer for review, and prepares supporting documentation. The compliance team sees every relevant event without manually reviewing every event.

Report generation
Navigator pulls structured data from the Milemarker Data Engine, applies the firm's calculation methodology, and generates draft client reports. The advisor reviews and approves the report — they don't build it from a spreadsheet.

Anomaly detection
Unusual activity — a large withdrawal request, a position that doesn't match the household's risk profile, a transaction that doesn't match past patterns — triggers an alert. Navigator doesn't decide what to do about the anomaly. It makes sure the right person sees it before anyone else does.

Where the hype stops

AI agents are genuinely useful. They're also genuinely overhyped in three specific ways:

1. They need good data to produce good output.
An agent that pulls portfolio data from three different systems with three different schemas and no reconciliation will produce three different analyses of the same client. The data infrastructure has to be right. In wealth management, that means normalized, structured data from every custodian and platform — not raw feeds duct-taped together.

2. They can't replace judgment.
An agent surfaces a portfolio that needs rebalancing. The advisor decides whether to rebalance now, wait for a tax event, or discuss it with the client first. The agent handles the observation and the presentation. The decision is still human.

3. Generic AI doesn't know what an RIA looks like.
ChatGPT doesn't know what your firm's model allocations are. It doesn't have your clients' positions. It doesn't know what a Schwab 1099 looks like or how your billing methodology works. Advisory firm AI agents need to run on real advisory firm data — normalized, structured, and inside your firm's environment.

The infrastructure requirement

AI agents in wealth management only work when they're connected to the right data. That means:

• Custodian feed data normalized to a common schema (not raw CSV files from three different portals)

• Client household data structured and maintained in a CRM the agent can query

• Portfolio performance data calculated consistently across accounts and custodians

• A compliance event log the agent can write to and read from

This is why the Data Engine has to come before the agents. Agents are a layer on top of infrastructure. They can't manufacture good data from bad inputs.

The practical starting point

Advisory firms that get the most out of AI agents start with one workflow. Not 20. One.

The most common first deployment: pre-meeting briefings. It requires the least compliance configuration, delivers visible value to advisors in the first week, and builds team confidence in agent-generated output before expanding to higher-stakes workflows like compliance monitoring or trade alerts.

Start where the time cost is obvious. Build trust in the output. Then expand.