Published on September 23, 2025.

Last week I wrote about the concept of NetSuite AI Agents — how they can watch, decide, act, and even learn. Then I followed that up with a case study of an Accounts Receivable AI Agent that I've been building for a client who was struggling with cash flow.

Since then, I've had a lot of people reach out. Some with curiosity. Some with skepticism. Some who are NetSuite developers that are anxious to build agents of their own.

So I thought I'd take a step back and answer some of the most common questions I've been getting.

What exactly is an AI Agent? How is it different from regular automation?

This is the question I've been getting asked most often.

A standard NetSuite automation is rules-based. You set a condition, define the action, and that's it. If the condition is met, the action fires. It's very consistent, but fixed.

An AI Agent is different. It doesn't just wait for a trigger and run a script. It observes what's happening, weighs the context, and decides whether to act. And when it does act, it can adjust based on what happens next. Over time, it learns patterns, refines its approach, and becomes more effective.

Where does the “AI” actually come in?

The AI is used in the decision making and learning aspects of an agent.

For example, an Accounts Receivable (AR) Agent can:

• Adjust thresholds for late payments if it notices customers consistently paying a few days slower.

• Learn from AR staff overrides so it stops flagging things that turn out to be harmless.

• Refine its scoring of which accounts to chase first, based on what's actually worked in the past.

That's the “AI” part. It's the difference between a rigid workflow and something that adapts.

Are AI Agents safe? What if they make mistakes?

Mistakes happen. That's why I add guardrails and Human-in-the-Middle (HITM) review to the agents.

In the early stages, no invoice goes out, no collection email gets sent, without a person approving it. Every action is logged. When the agent gets it wrong, it's not really a failure — it's data that can be used for training and improvements.

I like to think of it like onboarding a junior employee. You don't let them send emails on day one. You watch them, correct them, and gradually give them more trust as they prove themselves.

Will AI Agents replace people?

No - at least not the way I'm building them.

The primary goal of an AI Agent is to make people's lives easier. Make them more efficient. To take the noise away. To enforce consistency. To give clearer visibility.

That frees up AR teams to focus on higher-value tasks: building relationships, resolving disputes, planning ahead. The goal isn't fewer people — it's better use of the people you already have.

What other kinds of agents could we build?

There are a lot of possibilities...

Accounts Payable Agent: Catch duplicate vendor invoices, flag odd payment runs, make sure early-payment discounts aren't missed.

Inventory Agent: Watch for backorders, detect anomalies in fulfillment, alert when shrinkage patterns appear.

Revenue Recognition Agent: Monitor compliance, surface odd recognition events before they snowball.

Operational Agents: Keep an eye on sales orders, supply chains, and more.

Once you start thinking about AI Agents and what they're good at, you realize: every messy, expensive problem in NetSuite could have a digital teammate watching over it.

How are these agents deployed?

I build them using Suite.js as the foundation, using SuiteQL for data queries and SuiteAPI for actions. They run externally — usually on inexpensive AWS LightSail servers.

By building and deploying agents in this manner, they're independent of NetSuite's scheduling limitations. They don't wait for a script deployment. They don't need someone logged in. They're always on, and always running.

How do AI Agents learn?

Two ways:

Patterns in data: The agent sees when behavior shifts (say, DSO creeping up week over week).

Human feedback: Every time staff approve, reject, or edit what the agent suggests, that gets logged. Over time, the agent adapts.

Learning is slow, deliberate, and based on actual work. No magic, just discipline.

What's the ROI?

It depends on the client, but the most common wins are:
• Lower DSO.
• Earlier detection of problems.
• Less stress at month-end.
• AR teams spending time where it matters, not buried in reports.

The cost of running an agent is a fraction of hiring an extra AR staffer. And it scales as your business grows.

How does the AR Agent improve invoicing?

It ties invoice creation directly to fulfillment events. No more invoices that slip through the cracks or go out too early. And because of HITM, nothing gets posted without human sign-off (at least for now).

How does it help with collections?

Each morning, the AR team starts with a ranked list of at-risk accounts. Instead of wasting hours pulling saved searches, they know exactly where to focus first.

How does it track DSO differently from NetSuite?

Most of the companies I've worked with calculate DSO only once a quarter. The AR Agent calculates it every day. That gives finance leadership a rolling view — and early warnings if things are trending the wrong way.

What happens when the agent gets something wrong?

It gets logged. Staff correct it. And the correction becomes training data.

That's how the agent evolves. Slowly. Safely. And always with accountability.

Wrapping Up

Those are the kinds of questions I've been geeting since I started writing about AI Agents. And I think they're very good questions.