Custom AI Agent or Off-the-Shelf: How to Decide
When a ready-made AI agent is genuinely enough, the signs you need one built for your own systems and rules, and the real risks of each choice.
The real question
Almost every business software product now has an "AI agent" button somewhere. Your help desk has one. Your CRM (customer relationship management system) probably has one. There are website chat widgets that call themselves agents, and general assistants that promise to run your inbox.
So when an owner asks us whether they need a custom agent built, the honest first answer is: maybe not. Plenty of businesses are well served by switching on what they already pay for.
The real question is not custom versus ready-made in general. Our guide to custom versus off-the-shelf AI covers that wider decision for AI tools of all kinds. This article is narrower. An agent does not just answer; it takes actions in your systems. Where it acts, and under whose rules, is what decides which kind you need.
When a ready-made agent is enough
A ready-made agent is a good choice when the job sits entirely inside one product, uses information that is not sensitive, and does not need your own business rules.
Some honest examples:
- A website chat widget answering public questions. Opening hours, service areas, what to bring to an appointment, how returns work. The information is already on your website, nobody is harmed if an answer is a little clumsy, and the widget hands over to a person when it is stuck. You may not need anything built. Our website chatbots page covers when a built one is worth it.
- An agent feature inside your help desk. If it sorts and drafts replies to tickets that already live in that help desk, using articles that already live there, the vendor has done the hard work of connecting it.
- An agent feature inside your CRM that writes follow-up emails or summarises a customer's history from records in that same CRM.
- A general assistant for one person's own writing, research and meeting notes.
In each case the agent works inside the walls of one product. The vendor controls the data, the connections and the permissions, and that is fine because the stakes are low and the job is general.
If this describes the job you have in mind, start there. Try it for a month with real work. You lose very little if it does not suit.
Signs you need a custom agent
Things change when the agent has to act across your business rather than inside one product. These are the signs we look for.
It must act inside your own systems. The job touches Xero or MYOB, your job management system, your CRM and your inbox, often in one go. A quote request arrives by email, the customer is looked up in the CRM, the job is created in the job system, and a draft invoice is raised in Xero. A ready-made agent inside one of those products rarely reaches the others, or reaches them only to read, not to act.
It must follow your rules. Your pricing rules, your credit terms, which customers get which service level, which jobs need a licensed person. A ready-made product follows the vendor's idea of how a business works. If you find yourself writing long instructions to work around that, the product is the wrong shape.
Some steps need a person to approve. You want the agent to prepare a bill, a quote or a refund, and a named person to approve it before it goes anywhere. Ready-made products often offer all or nothing: the agent acts, or it only suggests in a chat window.
Where the data sits matters. If the agent will read client records, health information, payroll or pricing you guard closely, you may need it to run on Australian hosting under your own control. Our guide on public versus private AI explains the trade-offs.
You need an audit trail. When a customer or your accountant asks why something happened, you need a record of what the agent read, what it decided and what it did. A ready-made product may or may not give you that record in a form you can use.
If two or more of these apply, you are usually looking at custom AI agent development rather than a setting to switch on.
What custom involves
Custom does not mean building artificial intelligence from scratch. Most custom agents use a large language model from one of the major providers, the same kind of engine that sits behind the ready-made products. What is custom is everything around it:
- The brief. What the agent does, what it must never do, and when it hands the work to a person. Our guide on choosing the first job for an AI agent walks through writing one.
- The connections. Secure links into Xero, MYOB, your job system and your CRM, with access limited to what the job needs. If it only drafts bills, it cannot approve them. Our article on connecting AI to business systems safely goes into how that access is controlled.
- Your knowledge. Price lists, procedures and past examples the agent looks up before it acts. Our explainer on agentic RAG (retrieval augmented generation, where the agent searches your own documents before answering) covers how that works.
- The approval screen and the log. Somewhere for a person to check and approve, and a record of every action.
- Testing on your real examples before it touches live work, then a supervised month on live work.
The usual path is a short discovery to agree the brief and check the systems can be connected, then a fixed price after discovery for the build. The first version does one job. More jobs come later, once the first has proven itself.
The risks on each side
Neither choice is risk free. It helps to name the risks plainly.
Risks of a ready-made agent
- Shallow connections. It may read your other systems but not act in them, so a person still does the last step.
- The vendor changes it. Features, limits and pricing change on the vendor's schedule, not yours.
- Data location. Your information may be processed overseas, and you may not be able to see exactly where.
- Cost grows with headcount. Per-user pricing that looks small for three people looks different for forty.
- Generic behaviour. It does the job the way most businesses do it, which may not be the way you do it.
Risks of a custom agent
- It takes longer to start. Weeks of discovery and build, not an afternoon of setup.
- It needs an owner. Someone in the business, or a partner you keep, has to look after it when a connected system changes.
- Building too much. The biggest risk is building an agent for a job a ready-made product or plain automation would handle. Our explainer on automation versus AI agents helps spot that.
How to decide
Work through these questions in order. Stop at the first clear answer.
- Does the whole job happen inside one product you already use? If yes, try that product's agent feature first.
- Is the information public or low risk? If yes, a ready-made tool is probably enough.
- Does the agent need to act in two or more of your systems? If yes, check whether any product genuinely does that end to end, including the approvals. Often none does it well.
- Do your rules, approvals or data location requirements rule out the products you have looked at? If yes, custom is the realistic path.
- Who will own it? If nobody can, start with a ready-made product while you sort that out.
Many businesses end up with both: a ready-made chat widget on the website, and one custom agent doing the job that runs across Xero, the job system and the inbox. That is a sensible place to land.
If you are weighing it up, our page on AI agents for business shows the kinds of jobs we build agents for and how a project runs. For a sense of what a set of agents looks like in one real type of business, the dropshipping workflows article walks through five of them. And if you are still at the start, our plain explanation of what an AI agent is is a good first read.