Client support
Reads the enquiry, retrieves order and client details, answers routine questions and passes the rest to the right person with a summary.
AI agents that do the work – not just answer
A chatbot answers questions. An agent takes the task further: reads the email, retrieves the details, makes the change and lets you know when it’s done.
The customer wants to move a deliveryExample
Copilot Studio, Agent Builder or custom code
The choice comes down to your systems and contracts, not which platform we enjoy using most. If you already have Microsoft 365, it’s often a good starting point, but the agent can just as easily be built independently. We prefer to connect your systems using MCP, so those connections work regardless of the model.
Copilot Studio
Often the right starting point if you have Microsoft 365. The agent works where your team already works – in Teams, Outlook and SharePoint – using the permissions you’ve already set up.
Agent Builder
OpenAI’s tool for agents that need to work across multiple systems or with your own data, powered by ChatGPT models.
n8n
For when the agent is one step in a larger process: something comes in, the agent assesses it and the workflow handles the rest. Can run on your own server.
Custom code
Connect directly to Claude, ChatGPT or Gemini when the agent needs to be built into your own product, or when you need full control over logic, data and hosting.
Do you need an agent, or will a workflow do?
An agent is the right choice when a task requires judgement across several steps and the information sits in different systems. If the task follows the same path every time, a standard automated workflow is cheaper and more reliable. We’ll tell you which fits and build what you need.
- Respond to complaints that require checking orders, stock and terms/ AIMultiple systems and a different case each time. This is where an agent adds value.
- Send an order confirmation when an order is placed/ AutomationThe same steps every time. A standard workflow is enough.
- Qualify new leads and suggest next steps/ AIRequires judgement, but within clear rules.
- Approve a new supplier/ NothingA decision for a person to make. The agent can gather the supporting information.
How an AI agent works, step by step
The same four steps, whichever platform the agent is built on.
- 01
Receives a goal
The agent is given a clear task, such as "respond to incoming complaints", and rules for what it can and cannot do. We write those rules with you: which cases it handles, what amounts it can deal with and when it must always ask first. The clearer the goal, the more predictable the agent.
- 02
Plans
It breaks the task into steps: what do I need to know, where can I find it and what should happen next? The plan isn’t hard-coded. If information is missing, the agent looks elsewhere or asks a question, rather than following the same path every time. That’s what sets an agent apart from a standard automated workflow.
- 03
Uses your systems
It reads and writes in the systems it has permission to access – email, CRM, business systems or documents – through connections we build, using MCP or the systems’ own APIs, for example. The agent only gets the permissions its task requires, and sensitive actions can require human approval first.
- 04
Checks the result
It checks the result against the goal, asks a person when it’s unsure and logs everything it has done. The log shows you what the agent read, what it decided and why. That makes it easy to correct mistakes, adjust the rules and give the agent more responsibility once it has proved it can handle the task.
Tasks an AI agent can take on
Sales
Qualifies new leads against your criteria, drafts follow-ups and keeps your CRM updated after every interaction.
Inbox
Reviews incoming emails, sorts them by what the sender needs and prepares a reply or action for the person responsible.
Reports
Pulls figures from several sources, compares them with the previous period and writes a summary you can read in two minutes.
Quotes and supporting information
Gathers the client’s requirements, price lists and previous quotes, then writes a first draft for a salesperson to review.
Monitoring developments
Tracks public procurement opportunities, competitors or regulations and alerts you when something relevant comes up.
You decide how far the agent can go
We always start with an agent that makes suggestions for a person to approve. It only gets more responsibility once its results have proved reliable, and you can always take that responsibility back. You get the benefits early without giving up control.
01The agent suggests, you approve
The agent does the groundwork: reads the case, retrieves the information and drafts a proposal. A person approves it before anything is sent or changed. Every agent we build starts here, giving you a chance to see how it reasons.
- Nothing is sent without approval
- You can see how the agent reasons
- Corrected suggestions lead to better instructions
02The agent acts within limits
Once its suggestions are reliable, the agent can handle straightforward cases on its own, within limits you set: which systems it can access, what amounts it can deal with and which clients it can contact. Anything outside those limits still goes to a person.
- Limits on systems, amounts and clients
- Exceptions go to the right person
- Permissions can be extended or withdrawn
03Everything is logged and can be reviewed
Every step the agent takes is recorded: what it read, what it decided and why. That makes it easy to correct mistakes and show how you use AI if clients or auditors ask.
- Every decision is traceable
- Supporting records for clients or auditors
- Alerts when something is out of the ordinary
The difference between generative AI and agentic AI
Generative AI creates text, images or code when you ask it to. Agentic AI uses the same models but takes action: it plans the steps, uses tools and follows up. The first saves the person writing a few minutes; the second can handle the whole task from start to finish.
For an agent, the model itself is therefore only one part of the picture. Just as important are the systems it can access, the rules it follows and how it hands over to a person. We explain how we connect AI to the systems you already have in AI integration.
What is an AI agent?
An AI agent is a language model given a goal, access to tools and permission to take several steps on its own. It can search your systems, make simple decisions and act on them – handing over to a person when something is unclear. This is what’s known as agentic AI.
Unlike an AI chatbot, an agent doesn’t stop at the answer. It makes the change in your CRM, sends the draft for approval and follows up. Unlike a standard automated workflow, an agent chooses its own next step based on what it finds, rather than following the same path every time.
Frequently asked questions about AI agents
Can’t find the answer? Book an AI review or call us on 070-953 53 99.
AI review
30 minutes, one of your workflows
We review a task that takes up time today and give you an honest recommendation: AI, standard automation or no change at all.
- A brief written proposal for next steps afterwards
- Meet the people who will build it
- Free of charge
