AI agents for business: vendor, platform or your own AI employee?
If you look into AI agents for business, you'll come across three kinds of offers: ready-made products for one specific task, platforms for building your own, and support with building your own solution. Technically, the differences are big, and there's plenty to read about them. For you as the owner, a different question matters more: who maintains the rules, where does the data sit, and who checks the work once the system is running? That's what this article is about. We don't name or rate any vendors. We describe the three ways so that you can decide which one fits your business. For the rest of the article, we talk about AI employees. Why, you'll find further down.
The real question is a procurement question
When you need a new staff member, you also have three ways. You can hire a service provider that does the task with its own people. You can buy tools and organize the work yourself. Or you hire someone and train them. Each way has its place, and none of them is right for every task.
With AI employees, it's similar. The difference lies less in the technology than in who ends up carrying the responsibility for the running process. Before you compare offers, it's worth answering three questions for yourself:
- Who maintains the rules? Almost every process has variants. At some point, a new one comes along. Who teaches it to the system?
- Where does the data sit? In your filing system, with the vendor, or with both?
- Who checks? Who notices when the system makes a mistake, and who decides the cases where it's unsure?
With these three questions, we look at the three ways.
Way 1: a vendor's ready-made product
A vendor has built an AI system for one specific task and offers it to many businesses. Typical examples are tasks that run in a similar way in many businesses. You set up the product, connect it to your systems and use it.
Who maintains the rules? Largely the vendor. They keep developing the product, and you benefit from that. In return, the rules are usually kept more general, because they're meant to fit many businesses. How far you can adapt them to your variants depends on the product.
Where does the data sit? That depends on the product. Ask about it before real client data starts to flow.
Who checks? Some products come with their own reporting. Who on your team gets the exceptions is still something you decide yourself.
A fit if your task runs the same way in many businesses, you have few variants of your own and you want to get started quickly.
One question is worth asking about any ready-made product: what happens to your settings and rules if you want to switch later? Can you take them with you, or do you start from scratch? The answer tells you how tightly you're tying yourself in.
Way 2: a platform for building it yourself
A platform gives you building blocks: connections to email, filing and software, workflow blocks, AI steps. From those, you put together what you need yourself. Some of these tools come from the world of fixed if-then workflows and have had AI steps added.
Who maintains the rules? You or someone on your team. That's freedom and effort at the same time. Every new variant and every changed interface lands with you.
Where does the data sit? That depends on where the platform runs and how you set it up.
Who checks? Whoever builds it also has to decide who checks. If that decision is missing, the person who built it often ends up doing the checking.
A fit if someone on your team enjoys working with tools like these, has time for it and wants to take on the maintenance.
How fixed rules, click robots and AI employees differ is covered in more detail in Intelligent process automation: workflow tool, RPA or AI employee, and which fits when.
Way 3: your own AI employee, built with your team
This is the way we take. An AI employee gets its own narrow position in your business: one process or a small family of processes. Our experience: the narrower the position, the sooner it runs on its own.
It's hired like a new staff member: with a job description, a trial task and a permanent position. The job description sets out its role, tasks, permissions and limits, along with a contact person on your team. First, we watch your team at work. Then we train the AI employee, every variant once, with someone sitting right next to it. It's only signed off after several clean runs and a period in which it's monitored side by side.
Who maintains the rules? Your team, with our help at the start. If the AI employee is unsure, it sends the case to a person on your team. In a feedback session, the AI employee is trained on the new variant, and it's saved as a rule. That way, the rules grow with everyday work.
Where does the data sit? Your documents stay in your own filing system. We don't run servers for you; we help you set up the AI employees in your business. Where they run, which models they use and how far that can be set up in line with data protection rules, we decide together in the first sessions.
Who checks? At the start, your team, on every run. After sign-off, a report comes after every run, and anything uncertain goes to a person. Later, a reviewing role can bundle the reports. In Kevin's own business, that role is still on trial.
A fit if your process has many variants of its own, you want to keep it in your own hands for the long term, and one person on your team can help look after it. Templates and code stay with you.
The price of this way is time at the start. Watching, collecting variants, training and signing off can't be done in an afternoon. In return, at the end, every one of your variants has been trained in individually.
The three ways side by side
The following comparison is our own classification. Individual products and platforms can differ, and mixed forms exist.
Who maintains the rules?
- Ready-made product: mostly the vendor.
- Platform: you or your team.
- Your own AI employee: your team, through the escalation path and the feedback session.
How well does it fit your variants?
- Ready-made product: as far as the product allows.
- Platform: as far as you build it.
- Your own AI employee: every variant is trained in individually.
Where does the data sit?
- Ready-made product: depends on the product.
- Platform: depends on the setup.
- Your own AI employee: in your filing system; you decide where it runs.
Who checks?
- Ready-made product: the product's reporting plus your team.
- Platform: whoever built it, or whoever you designate.
- Your own AI employee: your team, then the report, later a reviewing role.
What do you need for it?
- Ready-made product: setup and connection.
- Platform: technical know-how and time on the team.
- Your own AI employee: one person on the team and time at the start.
If you don't want to commit to one way
Combinations are normal: a ready-made product for a standard task, your own AI employee for the process with all the variants. That's sensible. What matters is that the three questions are answered for every single process. A process where nobody knows who maintains the rules starts to drift, whatever tool it was built with.
When you talk to vendors, the questions in Hiring an AI consultant for a small business: ten questions to ask every provider, including us will help. And what control looks like once several processes run on their own is described in AI governance in SMEs: who checks the AI when you no longer do?.
Why we say AI employee
Many people say AI agent. What they mean is an AI system with a fixed task that does work in your business. We say AI employee, because the word describes how we handle it: it's hired, gets a job description, is trained and is led. The word is a reminder that it needs a contact person, rules, an escalation path and a report. That doesn't make it a colleague with a will of its own. The name is a handle for an agreement: what it does, where it starts, where it stops and what it doesn't do under any circumstances.
If you want to understand the technology behind it, you'll find it on Kevin's blog: What Is an AI Agent? Explained Simply.
What this looks like for us
At the tax advisory firm in our example, there were two processes: filing the digital assessment notices and filing the notices that arrive on paper. For them, we built two AI employees with two narrow positions. One files. A second one checks the inbox every 30 minutes, spots the emails with paper notices and hands them over to the first. Two narrow positions instead of one broad one.
The rules are the firm's own, not general ones. The assessment notice process distinguishes around 45 rules: many types of assessment, different types of tax and types of letter, and mail from the city with different layouts. Anything the AI employee can't match with certainty goes to the office team by email, followed by a feedback session. Filing happens in the document management system the firm already had. Both processes are at Autonomy Level 2, Autopilot, as of September 2026.
In Kevin's own business, 14 AI employees are at work, each with a personnel file, a fixed role and clear limits. An AI manager reviews their work; that role is still on trial.
Setting up and looking after a vendor's ready-made product for a client: we haven't done that yet. So we can only say as much about it as the comparison above does.
Runs without me
Every two weeks, one process that runs without the owner: what came in, what the AI did, where a person decided. Two minutes to read.
Worksheet
AI Rulebook on One Page
What your AI employee may do, what it must never do, who keeps an eye on it and what is checked before release: eight fields to fill in and an acceptance checklist. Comes as a PDF by email once you confirm.
Strategy session
In a 20-minute strategy session, you tell us where things get stuck. We'll tell you honestly whether an AI employee is a good fit for you.
Book a strategy sessionFrequently asked questions
Is your own AI employee more expensive than a ready-made product?
That can't be said in general. A ready-made product often takes less time at the start. Your own AI employee needs time for training, but in return it fits your variants. What it costs with us, we go over on the call, once it's clear which process would come first.
Do I need a programmer for my own AI employee?
You don't need to know how to program. It's built together in the 1:1, with ready-made AI employees as building blocks, and one person on your team looks after them with you. Where they run in your business and who sets up the access, we settle in the first sessions.
Which permissions does an AI employee get in my systems?
As few as possible. Our rule of thumb: read-only where that's enough, write access only where the process needs it. With us, a person sets up the access. The AI employee doesn't create any access for itself.
How do I notice that I've picked the wrong way?
By three signs. The exceptions land back with you, because nobody else is responsible. New variants pile up, because nobody maintains the rules. And you find yourself asking whether something has been done, because no report comes. Then it's rarely the tool. Usually, one of the three answers from the start is missing: who maintains the rules, where the data sits, who checks. That's our experience, not a rule for every case.
Can I switch to a different way later?
That depends on where your rules live. If they're written down on your side, in the process map, in the process profile and in the list of variants, you have them in your own hands. What you can take with you from a ready-made product or a platform depends on the vendor. Ask them before you start, as described above for the ready-made product. With us, templates and code stay with you, and we start with the process map and the process profile before anything gets built.
Your next step
Whether a ready-made product, a platform or your own AI employee is the right fit is decided by the process: by its variants, by your data and by who on your team can look after it. Bring your candidate for process no. 1 to the call. We'll go through the three questions together and tell you honestly whether our way is the right one for it. That's also where we go over how it works and what it costs.