Skip to content
DT Nova
AI consulting

AI agent development: agents that work inside your systems, under your rules

AI agent development is the step from an assistant that answers questions to software that does the work: it reads the incoming request, looks up what it needs in your systems, takes the next steps, and hands the case to a person when a decision or an approval is needed. We build AI agents for business processes such as order intake, supplier queries, customer service, finance and HR administration, with narrow permissions, an approval step before any action that sends, pays or changes records, and a log of everything the agent did. Hosted in the EU, and your company owns the code.

  • You see the result before it goes live
  • Fixed price for the agreed scope, in writing before work starts
  • Source code, domain and accounts in your company's name

Request a quote

What an AI agent does, before and after

Take a typical agentic workflow: a supplier emails to ask why an invoice has not been paid. Today someone opens the email, finds the supplier in the ERP, looks up the invoice, checks whether it is waiting for approval or blocked by a missing goods receipt, asks the purchaser, and writes back. Five systems and a quarter of an hour, many times a day.

With an agent, the email is read when it arrives. The agent identifies the supplier and invoice, reads the status from the ERP, sees that the goods receipt is missing, sends the purchaser a request to confirm delivery and drafts a reply to the supplier. A person in accounts payable sees the case with the facts already collected and the draft ready, and approves or edits it. When the purchaser confirms, the agent posts the receipt for approval and the invoice moves on.

The same pattern fits many processes: read, look up, decide the next step, prepare the action, ask a person where it matters. It is the agentic part of our AI consulting and build work, and it sits on top of ordinary business process automation rather than replacing it. For the basics, read what is an AI agent.

Where AI agents for business pay back

Agents are the wrong tool where the process is fully fixed and the data already structured. There, a plain integration is faster, costs less to run and is easier to check, and we will say so.

What we build

Keeping agents safe: permissions, autonomy and approval

The OWASP Gen AI Security Project lists Excessive Agency as one of its top ten risks for large language model applications in 2025 (LLM06:2025). It describes it as the vulnerability that enables damaging actions in response to unexpected, ambiguous or manipulated outputs from a model, and names three root causes: excessive functionality, excessive permissions and excessive autonomy (OWASP, read on 12 November 2026).

We design against each of them. The agent only gets the tools the task needs, not a general connector to a whole system. Each tool runs under its own account with the smallest rights possible, so a tool that reads invoices cannot change bank details. And autonomy is decided per action with you: what the agent may do alone, what needs approval, and what it never does. Text that arrives in emails and documents is treated as data, never as instructions to the agent, and the approval step means a manipulated message cannot make the agent pay or send anything on its own.

The rules that apply to AI agents

The GDPR gives people the right not to be subject to a decision based solely on automated processing which produces legal effects concerning them or similarly significantly affects them, with exceptions that still require safeguards such as the right to obtain human intervention and to contest the decision (Regulation (EU) 2016/679, Article 22). An agent that refuses a customer, a supplier or an employee is therefore built to prepare that decision for a person, not to take it.

Under the EU AI Act, providers must ensure that AI systems intended to interact directly with natural persons inform them that they are interacting with an AI system, unless this is obvious from the context (Regulation (EU) 2024/1689, Article 50(1)), and providers and deployers must take measures to support the development of AI literacy of their staff and others who operate and use AI systems on their behalf (Article 4, as replaced by Regulation (EU) 2026/1744, read on 12 November 2026). The European Commission's AI Act page states that the Act became applicable on 2 August 2026, with the transparency rules in effect from August 2026, and that after the AI Omnibus, which entered into force on 27 July 2026, the rules for high-risk use cases listed in Annex III, such as employment and access to essential services, apply from 2 December 2027, and the rules for AI in regulated products such as machinery and lifts from 2 August 2028. (All read on 12 November 2026.) An agent that screens job applicants or assesses creditworthiness falls into those high-risk areas; our AI governance guide covers what that means for the deploying company.

Accuracy, measured on your cases

No one can tell you in advance how well an agent will handle your process, so we measure it. Before the pilot we collect a set of real past cases with the outcome a person chose, and run the agent against them in a test version. You see, per case type, where it took the right steps, where it handed over, and where it was wrong and why. The threshold for what the agent does alone is then set by you from those results, and the test set runs again after every change to the model, prompts or tools.

Agent platform or custom AI agent development?

If answering questions would be enough, a chatbot may do the job; our chatbot vs AI agent guide explains the difference.

Ready-made agent builders, such as Microsoft Copilot Studio, Salesforce Agentforce, ServiceNow AI Agents and workflow tools like n8n or Zapier with AI steps, are a good start when the process lives in one vendor's system and the standard actions fit. Custom work pays off when the agent must work across several systems, follow your own approval rules, or run with data in your own EU account. Our guide to AI agent development cost explains what sets the budget and the cost per task.

SituationAgent platformCustom agent
Process inside one vendor's suiteUsually the quickest routeRarely worth building
Steps across ERP, CRM, email and your own appsConnectors of varying depthTools built for each system
Approval rules by amount, role or case typeGeneric approval stepsYour rules, versioned
Model choice and EU data regionThe vendor's choiceYour choice, changeable later
Audit log and test setPlatform logsFull case log and your own test set

If you are not yet sure which processes deserve an agent, start with our AI strategy guide or an AI consultant review of your use cases.

How we build it

  1. Walk the process. We follow real cases with the people who handle them today and write down the steps, systems, decisions and exceptions for your approval.
  2. Scope and fixed price in writing. Which case types, which tools, which actions need approval, and what the first phase leaves out, with a fixed price for that agreed scope.
  3. A test version on real data. The agent runs against your past cases in a test environment, and you see the results per case type.
  4. Pilot and acceptance. The agent works on live cases with approval on every action, and the phase is accepted against the written scope.
  5. Roll out and measure. Autonomy is widened only where the results support it; handling time and hand-over rates are compared with the numbers recorded at the start.
  6. Maintenance. Monitoring, model updates, test set reruns and changes to tools or rules, on a monthly plan you can cancel with one month's notice.

As a standard term, the source code, prompts and test sets sit in a repository in your company's own account, and the hosting is in your name in an EU data region. See prices.

Questions we get

What is an AI agent?

Software built around a language model that does more than answer: it reads a task, decides which steps to take, uses tools such as your CRM, ERP, email or document library to take them, checks the result and either finishes or hands the case to a person. A chatbot talks; an agent works through a process.

What is the difference between an AI agent and RPA?

A software robot in RPA repeats fixed clicks and fails when the screen or the input changes. An agent reads unstructured input, such as an email or a PDF, and chooses between steps. In practice we often combine them: fixed rules and interfaces where the process is fixed, and an agent only for the parts that need reading and judgement.

Can an AI agent make decisions on its own?

Only the ones you allow. Each action has a permission level: some the agent may take directly, such as tagging a ticket or drafting a reply; others, such as sending, paying, deleting or changing a customer record, wait for a person to approve. Decisions with legal or similarly significant effects on people stay with a person.

Which systems can an agent work with?

Any system with an interface the agent can use under its own service account: Microsoft 365 and Google Workspace, CRM and ERP systems, ticketing tools, databases and your own applications. Each tool is wrapped in a function with the narrowest permissions the task needs.

How do you know the agent does the job well?

We build a test set from your real past cases before the pilot and run the agent against it. You see, per case type, how often it took the right steps, how often it handed over and why. The same test set runs again after every change to prompts, tools or model.

Which AI models do you use?

The one that fits the task and your data rules, run in an EU region where personal data is involved, for example through Azure OpenAI, AWS Bedrock or Google Vertex AI, or an open-weight model on your own infrastructure. The agent is built so the model can be changed later without rebuilding the workflow.

Who owns the agent?

Your company. The source code, prompts, tool definitions and test sets sit in a repository in your own account, and the hosting is in your name in an EU data region.

Get a written quote

Tell us in a few lines what you need. We reply within one working day with a written quote: scope, timeline and a fixed price or an hour estimate. Asking is free and commits you to nothing.

See every price · Contact details

Request a quote