DEEN
Initial call

AI agents that take on tasks. Not the responsibility.

I design, develop and integrate AI agents for small and mid-sized businesses: software that completes multi-step tasks, draws on your company knowledge and uses your systems via interfaces. First we clarify whether an agent is needed at all. Often a simple workflow is enough.

Arrange an initial call

Reply within 24 hours. Free of charge and without obligation.

Service
Designing, developing and integrating AI agents for clearly defined tasks
Tasks
Handling enquiries, researching, updating the CRM, checking documents, preparing tasks
Access
Your company knowledge and your systems, via interfaces and with defined permissions
Control
Approval by your team wherever decisions or risks call for it
Principle
An agent only where a fixed workflow is not enough

01 Difference

Automation, assistant, agent.

Three terms that are often mixed up. The difference lies in who decides the next step.

Rule-based automationAI assistantAI agent
What it is A defined process that is carried out step by step according to rules. A language model that responds to a question or instruction: it writes, summarises, explains. Software that uses a language model to break a task down into steps and carries them out itself with approved tools.
Next step Decided by the rule that was defined beforehand. Decided by a person, with every new instruction. Decided by the agent, within the limits you define.
Access to systems Yes, via interfaces set up in advance. Usually none. The result is transferred by hand. Yes, via defined actions: look up, enter, trigger.
Example enquiry Form enquiry → create contact in the CRM → send acknowledgement. An employee pastes the email in and has a reply drafted. Reads the email, checks customer data and delivery date, prepares reply and CRM entry for approval.
Strength Predictable, cheap to run, easy to check. Ready to use, helps with individual texts and questions. Copes with freely written input and many variants.
Limit Fails with freely written or unforeseen input. Does not complete a process, but delivers individual answers. Results are very likely to be right, but not guaranteed. Needs checks and clear limits.

The more a system decides on its own, the more limits and checks matter.

02Applications

What AI agents are used for in a business.

Always for a clearly defined task. Never for “everything”.

The examples show what an agent prepares and completes, and where your team takes over. Which task fits in your business is something we work out on a specific process.

  1. Handling customer enquiries

    The agent reads incoming emails, recognises the request, adds customer data from the CRM and prepares a reply. It asks specifically for any missing details.

    Stays with people: Commitments on price, dates or goodwill.

  2. Researching information

    Before a meeting or a quote, it gathers what is known internally and in approved sources about a customer, product or case, each with its source.

    Stays with people: The assessment and what follows from it.

  3. Updating the CRM

    From emails, call notes and forms, it enters contacts, activities and next steps into the CRM, instead of them staying in the inbox.

    Stays with people: Changes to terms and sales opportunities.

  4. Checking documents

    It compares purchase orders, delivery notes or contracts with the order, price list or internal rules. It flags discrepancies and explains them.

    Stays with people: The decision on every discrepancy.

  5. Preparing tasks

    An enquiry becomes a prepared case: a draft quote, a proposed appointment or a task with all documents for the person responsible.

    Stays with people: Costing, approval and sending.

  6. Using company knowledge

    It answers questions from product data, manuals, policies and earlier cases. Every answer states what it is based on.

    Stays with people: Maintaining the knowledge and questions without a basis.

  7. Triggering follow-up processes

    Once a step is done, it starts the next one: create the order, inform colleagues, set a follow-up.

    Stays with people: Steps with external impact or financial consequences, which only run after approval.

These are examples, not packages. An agent often starts with one of these tasks and only takes on a second one later.

03 How it works

How an AI agent works on an enquiry.

The agent takes over the steps in between. Anything with consequences is approved by a person.

Example / AI agent in customer service

What an AI agent consists of

Goal
A clearly described task, for example: answering enquiries that can be answered from existing data.
Tools
Defined actions in your systems: look up, enter, prepare a message. It cannot do more than that.
Limits
Rules for when it stops and hands over to a person: when it is uncertain, data is missing or a step has consequences.

04 Knowledge and systems

An agent is only as good as its access.

It works with your company knowledge and your systems. Both are clarified and limited beforehand.

Company knowledge

what it relies on

  • Product data, price lists and terms
  • Manuals, instructions and internal policies
  • Earlier quotes, cases and correspondence
  • Frequently asked questions and their answers

So that answers can be traced, the agent relies on a maintained knowledge base, not on the general knowledge of a language model. An AI knowledge system makes this content searchable and keeps it up to date.

More on AI knowledge systems

Systems

what it may operate

  • Email, calendar and shared inboxes
  • CRM and ERP
  • Document filing
  • Industry software with an interface

The agent uses your systems via interfaces (APIs), with its own permissions, kept as narrow as possible: reading where it needs to read, writing only where agreed. Your software stays, it gets connected.

More on system integration

Permissions are granted as for a new employee, only more narrowly. What the agent does not need, it does not see.

05 Control

People decide where it matters.

Human-in-the-loop means a person is involved at defined points, checks and approves. Where these points are is something we define before the agent is built.

  1. Runs on its own

    For example

    Reading, classifying, looking up, summarising, creating drafts and internal notes.

    Why

    Steps without external impact that can be corrected at any time.

  2. With approval

    For example

    Replies to customers, quotes, purchase orders, changes to master data.

    Why

    The agent prepares, a person checks and approves or corrects.

  3. Stays with people

    For example

    Prices and terms, goodwill, complaints, legal and personnel matters.

    Why

    Decisions involving responsibility or risk. At most, the agent provides the documents.

What every agent gets

Log
Every step can be traced: what the agent did and what it relied on.
Handover
If it is uncertain or data is missing, it stops and hands over instead of guessing.
Narrow permissions
Only the access its task requires, and nothing beyond that.
Tests
Before going live with real cases from your everyday work, and spot checks afterwards.

06 Fit

Agent or workflow?

Most processes don’t need an AI agent. What matters is which solution makes the process reliable in the simplest way.

A fixed workflow is enough

and is then the better choice

  • The input is structured: form fields, tables, data from a system.
  • The process can be written down as if-then rules.
  • Every step has to run exactly the same way and be traceable.
  • There are few, known exceptions.

An agent makes sense

when several things come together

  • The input is written freely and inconsistent: emails, PDFs, notes.
  • The next step depends on the content and cannot be set in advance.
  • A result requires information from several sources.
  • Mistakes can be caught by checks and approvals.

Not the biggest solution. The most sensible one.

Often it is a combination: a workflow carries the process, AI takes on individual steps, such as reading an email. An agent is only added if that is not enough. If a fixed workflow is enough, I will tell you.

More on process automation

07 Approach

From the task to an agent that works alongside your team.

Design, development and integration from a single source, with a defined result and an end.

  1. 01

    Initial call

    30 minutes by phone or video, free of charge. We talk about one specific task and work out whether an agent is the right approach for it at all.

  2. 02

    Define the task and its limits

    What should the agent do, and what explicitly not? Where is approval needed, and how will we measure success? If it is still open whether a change is worth it, a Process Check will clarify that. You can prepare with the Process Profile.

  3. 03

    Clarify access

    Which knowledge and which systems does the agent need, via which interfaces and with which permissions? This includes data protection and the choice of AI service.

  4. 04

    Build and test

    I develop the agent in stages and test it with real cases from your everyday work. It only goes live once the results are right.

  5. 05

    Introduce

    At first, your team checks more than will be necessary later. Responsibilities, approvals and documentation are clear.

  6. 06

    Measure and hand over

    We compare against the starting value, which you can record with the “Measure First” tracking sheet, for example, and close with a hand-over. Approvals are only relaxed if the results justify it.

Arrange an initial call

08 Questions

Frequently asked questions.

On definitions, autonomy, company knowledge, costs and what happens when something goes wrong.

What is an AI agent?

An AI agent is software that uses a language model to break a task down into individual steps and carries them out itself with defined tools: searching for information, operating programs via interfaces, entering results. It works within limits that are defined beforehand.

Agentic AI is the umbrella term for AI systems that do not just answer, but act.

What is the difference between an AI agent and a chatbot?

A chatbot holds a conversation and answers questions. An AI agent completes a task: it looks things up in systems, enters data and triggers follow-up steps. A chatbot can be part of an agent, for example as the interface through which customers submit their enquiry.

Which tasks are AI agents suitable for in small and mid-sized businesses?

For recurring tasks with freely written input, where the next step depends on the content: handling customer enquiries, researching information, maintaining the CRM, checking documents and preparing cases.

Less suitable are tasks that can be described entirely as fixed rules. For those, rule-based automation is simpler, cheaper and more predictable.

Are there autonomous AI agents that handle entire processes on their own?

Technically, AI agents can carry out many steps without intervention. In day-to-day operations, full autonomy rarely makes sense, because the results of a language model are very likely to be right, but not guaranteed.

The agents I develop therefore only work on their own for steps without external impact. Anything with consequences for customers, money or contracts is approved by a person first.

Can an AI agent access our company knowledge?

Yes, if the knowledge is available digitally and made accessible, such as product data, manuals, policies or earlier cases. The agent then bases its answers on these sources and names them. Which content it may see is defined through permissions.

Does an AI agent work with our existing systems?

Usually yes, provided there is reliable access: an interface (API), an import and export, or another stable route. I check whether it exists before anything is built. Existing systems are used and connected, not replaced.

What happens if the AI agent makes a mistake?

Every agent has checks and clear limits. If it is uncertain or data is missing, it hands over to a person instead of guessing. Steps with external impact go through an approval, and every step is logged, so it remains traceable what the agent did and what it relied on.

What does it cost to have an AI agent developed?

That depends on the task, the systems involved and the approvals needed, which is why there is no price list. Before implementation, we define the scope, the goal and a fixed end to the project, and you receive a quote for it.

Running costs include use of the AI service, depending on volume. They are quantified before implementation. Support after the project is optional and starts at 250 euros a month.

Do you build an AI agent even if a simple workflow would do?

No. If a fixed workflow solves the task reliably, that is what I recommend: it is more predictable, easier to check and cheaper to run. An agent is only used where fixed rules reach their limits.

How do you handle data protection?

Which data the agent processes, where it is stored and which AI service is involved is something we clarify before implementation. This includes the provider, where the data is stored and a data processing agreement. The agent only gets the access it needs for its task.

Get in touch

Which task should an AI agent take on in your business?

Describe it to me in the initial call. Then we will know whether an agent is the right way or whether there is a simpler one.

Arrange an initial call

Reply within 24 hours. Free of charge and without obligation.