What AI agents do for a benefits agency

I’m Chris Cordon, a 17 year benefits producer. I run my own work on a team of AI agents built on Claude. I haven’t built an agent team for a benefits client yet, and I’ll say so on every page. Everything below runs for me today.

What does an AI agent team cover?

An AI agent team at DoBetter covers seven services:

AI executive assistant.

Starts each session with what’s due, what’s overdue, and what the other agents need from you, then hands the work out.

AI agent teams.

Separate agents with separate jobs that pass work to each other.

Research agents.

A meeting brief on the person and the business before you walk in.

Inbox agents.

Sorts incoming email and drafts replies for a person to approve.

Shared memory.

One business brain every agent reads, so nobody explains yesterday twice.

Custom agents.

An agent built for one job at your agency, with its own written rules.

Human approval and guardrails.

The rules about what an agent can draft, what it can send, and what it can never touch.

What does a day look like before and after?

Before, my mornings depended on whether I’d left myself notes. I built a “close to open” button into the first version of my CRM. Close to open is a restaurant term. If you closed on Saturday night and Sunday was brunch, you set out the salt and pepper, the coffee cups, and the Tabasco so the morning crew could just get going. “The only problem is I’m forgetful, so a lot of times I would open up, and no one had closed.”

After, the closing happens without me. My executive assistant agent opens every session with one call to the business brain: tasks due today, overdue items, and messages other agents left for me. A weekday digest lands at 7:00 every morning. Anything that looks already done shows up in a numbered list, and I clear it by replying with the numbers.

Meeting prep changed the same way. I used to Google a business about 15 minutes before the meeting and wing it. Now a research agent builds the brief in about 5 minutes, with six questions to ask before I leave the room.

What won’t AI agents do?

AI agents won’t replace your account managers or producers. They won’t give legal, tax, or medical advice. And at my agency, they don’t send email on their own.

“If it’s not one of a few set issues that we have a pretty clear template on, my judgment is the bar that needs to be met.” My inbox agent can read and write drafts. It can’t send. The send button stays with a person.

AI agents also won’t fix a process nobody has written down. An agent follows rules. If nobody at your agency can say how a renewal should go, the agent can’t either.

Who are AI agents a fit for?

AI agents fit a benefits agency where the owner or a producer is the bottleneck: every meeting needs prep, every reply needs their judgment, and every decision lives in their head. They’re a poor fit for an agency that wants software to act without anyone checking, or one that has no time to review drafts.

I’ve built custom agents for one job before. Finn, a benefits Q&A chatbot, answers from about 300 emails’ worth of real questions and answers plus all of an agency’s plans. Finn ran on every landing page I built.

Quick answers

What do you build on?

Claude, the AI model from Anthropic. It’s the same one my own agents run on.

Have you built this for a client?

Not an agent team yet. I run one for my own work every day, and I’d build yours the same way.

Will an agent email my clients without asking?

No. Drafts wait for a person unless you decide a narrow, templated reply can go.

Where would you start?

Usually with shared memory and a meeting brief. Everything else reads from the memory.

What does it cost?

Every build is a fixed fee, scoped after a $1,500 discovery.