AI agents
AI agents with one job, a full log and human approval
AI agents are language models given access to tools so they can solve a task in several steps. I build them bounded, with a log of every step and a human involved where it matters. The experience comes from 400+ projects since 2010.
- I usually reply the same day
- Written estimate before work starts
AI agents
What agents are good at
Agents work best on tasks where the steps are known but the order depends on what turns up along the way. A person would check three places in whatever order makes sense given what appears, and an agent can imitate that.
They work badly on tasks with no right answer, or where a mistake only surfaces much later. Judgement calls, prioritising between customers and anything requiring discretion should stay with people, and I say so from the start.
Typical tasks for an agent
- Customer service: investigating a case across the order system, carrier and email history, and preparing a draft reply for approval.
- Document handling: filing, naming and routing incoming files based on their content.
- Internal tools: maintaining data by finding duplicates, filling gaps and flagging what looks wrong.
- Integration with existing systems: working through APIs in your CRM, accounting system or webshop.
Approval and control
The most useful setup is rarely the most autonomous one. Often it is the one where the agent finishes the work and queues it for approval, so a person spends two minutes instead of forty.
That model is also easier to start with, because nobody has to let anything loose on day one. Trust grows as the log shows what the agent does and why.
What you get
Concrete deliverables, not a report of recommendations someone else has to find time for.
- One bounded agent
An agent with one clear remit, a step limit and a way to stop it in the middle of a task.
- A tool layer
Access to your systems, split between reading and writing, where write access only arrives once the read side has proven itself.
- A full log
Every step, every tool call and the reasoning behind it, so you can debug and always answer what happened.
- An approval flow
Where the consequences are large, the agent queues its work for approval instead of carrying it out itself.
How it works
Four steps. You know what happens when, and you can stop after any of them.
- 01
Pick the process
Together we find the piece of work that repeats often enough for automation to pay off. The more contained the task, the faster the first version reaches production.
- 02
Set a baseline
We record how long the task takes today, how often it occurs, and how many errors slip through. Without those numbers, nobody can tell afterwards whether the solution was worth the effort.
- 03
Build against a test set
I collect real examples from your day-to-day and test every prompt and model change against them. That keeps quality stable when the solution is tuned later on.
- 04
Production with logs and a kill switch
The solution is wired into your existing systems with full logging and a switch that can turn it off. I document the setup and train someone on your side to maintain it.
Brands I've worked with
Frequently asked questions
- What is an AI agent, briefly?
- A language model with access to tools, allowed to use them across several steps until the task is done. It can look things up in systems, gather information and propose or carry out actions. The difference from a chatbot is that it does work instead of only answering.
- What does an AI agent cost?
- The price depends on what the agent needs to do. What matters most is how many systems it works in, whether those systems have an API, and how much approval flow needs to be built around it. Send a description of the task and I will come back the same day with clarifying questions and then a fixed quote.
- What if the agent does something wrong?
- Then the mistake has to be detectable and reversible. That is exactly why I start with read access and add write access afterwards, one action at a time, with logging on everything.
- Which systems can an agent work in?
- Anything with an API: Shopify, Google Workspace, HubSpot, e-conomic, Jira or your own system. If the API is missing, a layer can often be built in between, but that belongs in the estimate because it is frequently the largest part of the job.
- What about GDPR when an agent has access to customer data?
- The agent only gets access to the fields it needs, and everything it does is logged. The log in itself helps with documentation requirements. Personal data can often be masked before it reaches the model, and I walk through the setup with you before anything goes live.
- Do you build agents for companies outside the Copenhagen area?
- Yes, most of the work happens remotely anyway. I have clients in Aarhus, Aalborg and Odense whom I mostly meet over video. If you are near Copenhagen, I am glad to come out for the kickoff meeting.
- How long does it take to build an agent?
- A bounded agent on one system is typically two to four weeks including testing. Working across several systems, it is the integrations rather than the AI part that set the timeline.
- Can the agent work outside office hours?
- Yes, and that is often the point. Cases that arrive in the evening can be investigated and prepared by the time someone starts in the morning. The decision itself can still sit with a person.