AI agent development
AI agents that work inside your CRM, ERP and documents: they read, check and prepare. A person makes the call on anything important.
An AI agent is a program that uses a language model to understand a task, chooses its own steps and carries out actions in a company's systems: it reads a document, finds data in the CRM or ERP, fills in a record or prepares a reply. It differs from a chatbot because it acts rather than only answers, and from conventional automation because it works with unstructured data.
Where AI agents deliver value for business
- Document processing - invoices, contracts and applications: data is extracted and checked against what is already in the system.
- Email and enquiry handling - classification, routing to the right person and a draft reply with data from the CRM.
- Internal AI assistant - answers drawn from company policies, contracts and the knowledge base, with a link to the source.
- Data queries in plain language - a manager asks how many orders are running late this week, and the agent builds the query and shows the result.
- Summaries and reports - digests of long email threads, project status and meetings.
AI integration with your existing systems
An agent is only as useful as the data it can reach. We connect it to your CRM, ERP, accounting, document management and email through the same API connections we build in our system integration work. Each agent gets defined permissions: what it can read, what it can change and what requires human approval.
If a process can be described with clear rules, it does not need AI: ordinary code will do the same job more cheaply and more reliably. We propose an agent only where the data is unstructured or every case is slightly different.
An internal assistant built on company knowledge
This kind of assistant answers from your documents, not from general knowledge: before answering, it finds the relevant passages and shows where they came from. The method is called RAG (retrieval-augmented generation). Access rights still apply: an employee never gets an answer from a document they are not allowed to see.
Human oversight and data security
- Approval step - actions that change data or go out to a customer wait for a human decision.
- Action log - every step the agent takes is visible with its sources, so any error can be traced.
- Data location - the model provider and processing region are chosen according to data sensitivity; where needed, the model runs on your own server.
- EU AI Act compliance - the risk category is assessed before development, not after it.
This matters especially for public bodies, so we cover the requirements in more detail under solutions for the public sector.
How the work is done
We start with one process and collect a set of real examples: documents, emails or enquiries together with the correct answers. We use it to measure the agent's accuracy during the pilot, and the agent goes live only once results reach the agreed threshold. After launch, the same set tests every change. More about our process.
Chatbot, rule-based automation and AI agent
| Criterion | Chatbot | Rule-based automation | AI agent |
|---|---|---|---|
| What it does | Answers questions | Follows predefined steps | Chooses steps and takes actions |
| Data | Scripts or documents | Structured fields | Emails, documents, free text |
| Access to systems | Usually none | Through integrations | Through integrations with permissions |
| When a case is unusual | Hands over to a person | Stops with an error | Tries to resolve it or hands over to a person |
| Predictability | High | Complete | Measured with a test set |
| Best suited for | Frequent, identical questions | Clear rules | Every case slightly different |
Frequently asked questions
What is the difference between an AI agent and a chatbot?
A chatbot answers questions from a script or a set of documents. An AI agent can also act: it reads an email, finds the order in your CRM, checks the data and prepares a reply or a record. The difference comes from access to your systems and clearly defined permissions for each action.
Will our data be used to train the language model?
No. We use the business APIs of model providers, whose terms state that requests are not used for model training. Where the nature of the data requires it, we choose processing in the EU region or an open model running on your own infrastructure.
Can an AI agent understand documents written in Lithuanian?
Yes. Modern language models read and write Lithuanian well. Even so, we test the terms and wording specific to your field against real documents during the pilot, because that is where errors most often hide.
What happens when an AI agent makes a mistake?
Actions that change data or go out to a customer wait for human approval by default. Every decision the agent makes is logged together with the sources it was based on. We measure accuracy against a test set before launch and after every change.
How does the EU AI Act affect AI agents used in business?
Most internal agents, such as document processing or an assistant for staff, fall into lower risk categories where transparency matters most: people must know they are interacting with AI. Areas that affect people's rights, such as recruitment or access to public services, carry stricter requirements. We assess the risk category during analysis, before development starts.
Is EU funding available for implementing AI solutions?
The Lithuanian Innovation Agency periodically opens calls for implementing and developing artificial intelligence solutions. Eligibility conditions differ from call to call, so we check the requirements of the specific call. For the application we prepare a technical description of the solution and a scope that matches the real work.
Where should a company start with AI automation?
With a single process where the same work repeats every day and there is a clear criterion for a correct result. A pilot project with real data shows within a few weeks whether accuracy is good enough, before you invest in a full rollout.
Which task in your company repeats every day?
Tell us about the process, and we will assess whether an AI agent is the right fit or a simpler solution will do.
Or directly: info@desamedia.lt · +370 686 59999