The DigiKat Learning Center

The Five Levels of AI Agent: L1 to L5 Explained | DigiKat

Written by Mick Goman | Nov 16, 2026, 8:30:00 PM

Every week someone asks me what "an AI agent" costs. It's a fair question with a frustrating answer: it depends on what the agent is allowed to do. A website helper that answers questions from your FAQ is an agent. So is a phone receptionist that books appointments, sends SMS confirmations and updates three systems. They are very different builds.

So we stopped quoting agents as one thing. Every DigiKat agent is now scored on six factors, and the score puts it at one of five levels, L1 to L5. Each level has a published build price and a published monthly fee on our pricing page. This post explains how the scoring works, what each level actually does, and how to work out which one you need.

All prices are in Australian dollars, ex GST.

The six things we score

Before we talk levels, here's what moves an agent up or down.

  1. Channels. One chat box on your website is simple. Web, email, SMS and phone together is not. Each channel has its own quirks, and voice is the hardest of all.
  2. What the agent is allowed to do. Answering a question changes nothing. Writing to a contact record, booking a slot or sending an email changes something. Taking a payment changes something that matters a lot.
  3. How many systems it connects to. An agent that only works inside HubSpot has one place to break. Add a booking system, a payment provider and a pathology feed and you have four.
  4. Its logic. "Answer from this knowledge base" is one step. "Check eligibility, pick the right product, quote it and chase the paperwork" is many steps, with branches.
  5. The risk. What happens if it gets something wrong? A slightly clumsy FAQ answer is a small problem. A wrong quote, or anything near a clinical decision, is a big one, and it needs approval gates built in.
  6. The volume. Ten conversations a week and ten thousand need different testing, monitoring and tuning.

The total sets the level. Our calculator uses the same scoring, so the estimate you see there and the level we talk about in a proposal come from the same place.

L1 Answer: $500–$1,000 build, $50 a month

An L1 agent answers from your knowledge on one channel and changes nothing. Think of a website helper that knows your services, opening hours, pricing and policies, or an FAQ agent on a members' portal.

It can't book, it can't update a record, and it can't send anything on your behalf. That's the point. It's the lowest-risk agent you can run, and it tells you quickly whether your knowledge is actually written down anywhere. Often it isn't, and fixing that is half the value.

L2 Act: $1,500–$3,000 build, $100 a month

An L2 agent does things. It captures details, writes them to HubSpot, books appointments and sends emails. Typical examples are intake, booking, a document chaser that follows up missing paperwork, and a renewals agent that notices when something is due and starts the conversation.

This is where most businesses should start. The work is repetitive, it's easy to measure, and everything the agent does lands on the HubSpot record where you can see it.

L3 Transact: $3,000–$6,000 build, $175 a month

L3 agents handle multi-step work, quote or take payment, and connect to one or two systems outside HubSpot. Examples are a quote agent, an order agent, and a results agent that takes incoming results, structures them and routes them to the right person.

The jump in price comes from the risk and the integrations. Once money or a second system is involved, there's more to test and more to watch. Our rule is simple: if it touches money, a person signs off.

L4 Orchestrate: $6,000–$9,000 build, $300 a month

L4 agents work over voice or SMS, connect several systems, and run other agents. A phone receptionist that answers calls at any hour, books appointments and logs every call is L4. So is a portal concierge that can hand work to other agents behind the scenes, such as intake, booking and document chasing.

Voice is harder than it looks. People interrupt, mumble, change their minds and call from noisy cars. Coordinating several agents adds its own failure points. That's why L4 costs what it does.

L5 Custom: quoted, from $500 a month

L5 is regulated work, custom software and approval gates. Clinical-adjacent workflows in telehealth live here, along with custom engines that don't fit a standard pattern. We quote these individually after the audit, because the scope, the controls and the testing are specific to each one.

If someone offers you a fixed-price agent for regulated work without asking about your approval process, be careful.

What the monthly fee sits alongside

Each agent's monthly fee sits alongside a managed agents retainer of $50 a month for each website or portal your agents run on. So a portal with one L1 agent is $100 a month. Two L2 agents on one website would be $250 a month.

The retainer covers hosting, monitoring, tuning, knowledge updates and a monthly report. AI usage is charged separately, and voice minutes and SMS are passed through at cost.

Every build starts with the AI Readiness Audit, from $2,500. It reviews your HubSpot record, your data and your customer journeys, and fixes the build price in writing before we start.

How to tell which level you need

Most people overestimate the level they need. These are the questions I ask in a first conversation.

Does it need to change anything?

If the job is answering questions, you need L1. If it has to write to a record, book something or send something, you're at L2 or above.

Does money or a second system come into it?

Quotes, payments, orders, or a system outside HubSpot such as a booking or pathology platform push you to L3.

Is it on the phone or SMS, or is it running other agents?

That's L4. If you want a phone receptionist, there's no shortcut around it.

Would a mistake hurt someone or breach a rule?

If the honest answer is yes, you're looking at L5 with approval gates. Or you redesign the job so the agent prepares the work and a person makes the call.

Can you split it?

This is the question that saves the most money. A "do everything" L4 concierge is often an L1 helper plus an L2 intake agent, built months apart. You get value from the first one while you learn what the second one really needs to do. It's the same thinking as choosing between an agent and a workflow: use the simplest pattern that does the job.

One more test. If you can't describe the queue the agent will work from, you're not ready for any level yet. Build the queue before the agent.

Why we publish the prices

A lot of agent pricing is "contact us". I understand why. Scope really does vary. But a range you can see beats a number you have to chase, and the six factors give us a shared language. When a client asks why their agent is L3 instead of L2, we can point to the factor that moved it. Usually it's money, or a second system.

The levels also make trade-offs obvious. If dropping SMS takes an agent from L4 to L3, you can decide whether SMS is worth the difference. That's a much better conversation than arguing over a single number. And whatever the level, the same rule applies: a person approves anything that matters until the agent has earned more freedom.

Where to start

Have a look at the full table and the calculator on our pricing page. Score your idea against the six factors, see where it lands, then book an audit when you want the price fixed in writing.