AI and artificial intelligence

A large language model (LLM) is an AI model trained on a huge amount of text that can understand and generate language. ChatGPT, Claude, Gemini and most of today’s AI assistants and agents are built on LLMs.

An AI agent is an artificial intelligence program that doesn’t just answer questions but takes actions itself to reach a goal: it reads an e-mail, finds information in a system, fills in data and sends a reply. It works within the limits you set and, when needed, asks a person for approval.

Artificial intelligence (AI) is technology that lets computer systems do tasks that normally require human thinking: understanding text, recognising images, making forecasts and simple decisions. The most common kind today is generative AI, such as ChatGPT, which writes text and answers questions.

ChatGPT is an AI assistant developed by OpenAI that understands and writes text in many languages, including Latvian. At work it is used for drafting e-mails and texts, summaries, generating ideas and analysis. For company data, use business plans or integrated solutions rather than a free account.

Generative artificial intelligence is AI that creates new content — text, images, code, audio or video — from your instructions. ChatGPT, Claude and Gemini are examples of generative AI for text, and image generators for images.

Machine learning is a field of artificial intelligence in which a computer isn’t programmed with rules but learns patterns from data by itself. For example, from thousands of past orders it learns to forecast demand or spot an unusual transaction.

RAG (retrieval-augmented generation) is an approach in which AI first finds the relevant information in your documents or database and then answers based on it. That way AI can answer questions about your products, rules and customers, and the answer can point to its source.

The EU Artificial Intelligence Act (Regulation 2024/1689) is the law that regulates the development and use of AI systems in the EU according to their level of risk. It entered into force on 1 August 2024, and its requirements apply in stages: since 2 February 2025 unacceptable-risk AI practices are banned and companies must ensure their staff’s AI literacy.

An AI agent works in a loop: it receives a task, uses a language model to plan the next step, uses a tool (for example, searches a database or sends an e-mail), evaluates the result and continues until the task is done. The tools, permissions and limits are defined in advance.

Start with one specific, repetitive task and a measurable goal. Then decide which data and systems the agent needs access to, build a prototype with real data, add checks and human approval — and only then put it into everyday use.

Gather the documents that hold the answers (instructions, procedures, product descriptions), sort out access rights and build a RAG solution that finds information in these documents before answering. The assistant can live in your intranet, Teams or your own system.

AI adds the most value in repetitive work with text and documents: processing invoices and requests, customer questions, summaries and moving data. Start with the one process where the most time is lost, measure the result and only then expand.

The price depends on how many systems the agent has to connect to, how complex the decisions are and how much data it has to process. A simple agent for one process costs considerably less than an agent that works with ERP, accounting and e-mail. On top of development there are running costs for using the language model and maintenance.

A chatbot talks — it answers questions from a script or with the help of AI. An AI agent acts — it carries out actions in your systems itself, such as creating an order, entering an invoice or booking a time slot. Often a modern website chatbot is already an agent with access to specific tools.

Yes. Modern large language models — ChatGPT, Claude, Gemini — understand and write Latvian well, including diacritics and grammatical cases. For important texts it is still worth checking terminology and style, because there is less training data in Latvian than in English.

According to Eurostat, in 2024 AI was used by 8.83% of Latvian companies with at least 10 employees — almost twice as many as in 2023 (4.53%). On average in the EU, around 20% of such companies used AI in 2025, so Latvia has a lot of room to grow.

Yes, if AI solves a specific, recurring problem. In a small business every hour saved is noticeable, and today AI solutions can be introduced step by step — starting with one process. It isn’t worth it if it’s unclear what work AI will make easier.

It can be safe if you use business-grade AI services or your own solution where data isn’t used to train models, there is a clear data processing agreement and access rights are limited. Entering personal data or trade secrets into free accounts is not recommended.

A language model produces the most plausible answer rather than checking facts, so it sometimes confidently makes up incorrect information — this is called a hallucination. The risk drops when AI answers based on your documents (RAG), cites sources in its answers and, where it matters, the result is checked by the system or a person.

More often AI replaces not people but individual tasks — repetitive data entry, sorting, standard replies. Employees gain time for work that needs judgement, customer relationships and responsibility. Companies that use it can get more done with the same team.