What are AI agents and what use are they to a business?
There are two of us, and two AI agents running alongside, one of which is actually running and the other still only exists on paper. What they do, what they're explicitly not allowed to do, and where the headline above doesn't hold up.
Lees dit in het Nederlands →$ ps aux | grep agent
What are AI agents and what use are they to a business?
An AI agent is more than a chat window: it remembers what happened before, uses tools itself, and starts on its own at set moments. What you get out of it: it replaces the work you'd outsource, not the work you'd hire someone for. Below is what that means for us, including the part that doesn't work.
"AI agent" is a term that sounds bigger than it is. Below is what's actually running for us, including the part that doesn't work.
01 / the definitionWhat is an AI agent and what use is it?
An AI agent is more than a chat window. The difference lies in three things: it remembers what happened before, it's allowed to use tools itself (a CRM, a calendar), and it starts on its own at set moments. A chat window waits for you to type something and forgets the conversation afterwards.
What you get out of it is limited. This piece previously ran under the headline "two people, running as twenty", and that headline is a feeling, not a measurement. What is true: the two of us do work you'd normally hire someone for: preparing content, keeping files up to date, monitoring systems, drawing up reports.
What isn't true: that this replaces twenty people. Twenty people do things no system does. They call back a client who's angry. They see a deal slipping away before it shows up in the figures. They make a decision with incomplete information.
An agent replaces the work you'd outsource. Not the work you'd hire someone for.
If you confuse the two, you're buying disappointment.
02 / who's runningWhat does the agent that actually exists do?
That's called Sigi, and it's been running since 24 June 2026.
Sigi sits on Bram's computer, is reachable through a messaging app, and works on the commercial side: preparing texts, recording client observations, doing research, setting up assignments. There's one task that starts on its own, every Monday morning.
How it's set up is more interesting than what it does:
That third one is the most important, and also the dullest. An agent that can access everything is easier to build and much harder to trust.
03 / what isn't running yetWhat's design and what's production?
The second agent is design. None of it is running.
That one's called Saga and is meant to handle the technical side: monitoring servers, flagging outages, reporting each morning on what happened. The design is there, with six connections to systems it would read from.
Why it isn't there: four of those six systems aren't running yet themselves. The CRM, the client inbox and the management environment have been decided on but not yet installed. Building a monitoring agent for something that isn't running means doing the work twice.
We're writing this down because it's exactly the kind of claim businesses get away with. "Our AI agents run 24/7": for us that covers one of the two, and not even 24 hours a day.
What Saga does do, today: helping write the build and the knowledge base, in conversation, with a human checking. That's an assistant, not an autonomous service. This post, incidentally, is one such piece.
04 / the hard lineWhy doesn't either of them publish on its own?
Because it isn't a restriction we added around it afterwards. It's the design.
Everything that goes out (a blog post, an email, a proposal) arrives with a human as a draft. They read it, adjust it, and send it. That takes time, and that's the point.
Three reasons, in order of weight:
- Errors aren't rare, they're rarely visible. A model that invents an amount, a name or a date does so in a text that otherwise reads perfectly. That's exactly why you can't catch it with spot checks afterwards.
- Responsibility doesn't shift along with it. If it's under your name, it's yours. "The system did it" isn't an answer you can give a client.
- The corrections are the fuel. Every time a human adjusts something, that's the information that makes the next version better. Publishing automatically throws that away.
Two things follow from that: neither of our agents talks to clients, and neither is allowed to change anything in systems holding real client data without a human saying yes.
05 / what this means for youWhat can you copy from this?
Four steps, and you don't need an agency for it.
- Step 1: start with repetitive work that follows a fixed pattern. Call notes in your CRM, quotes from a template, a weekly overview. Not the things you enjoy.
- Step 2: build in the approval step from the start, not later. Adding a human into a process that was designed without one almost never works.
- Step 3: give it as few permissions as possible. Read access where reading is enough.
- Step 4: write down what it's not allowed to do. That list is more useful than the list of what it does, and it's shorter.
And the honest expectation: in the first few weeks you'll correct a lot. That's not a sign it's failing, that's the work. Anyone who tells you it sounds right straight away has never done it.
What is an AI agent, as opposed to a chatbot?+
Do you really run as twenty people with just two?+
Do your agents publish anything themselves?+
Do your agents talk to clients?+
How do I start with this myself?+
Want to know which work in your business could take on a role like this?
Book a call
We look at where your time leaks away and say honestly whether we can do anything about it. Often the answer is: you can do this yourself. Then we say so, and you'll have spent thirty minutes getting a clear answer.
This text was produced with AI assistance and checked and approved for publication by a human.