Glossary: software, AI and automation

The AI Act is an EU regulation that sorts AI systems by risk. Most business agents are low risk: you only have to tell people they are dealing with AI. Strict rules apply to areas such as hiring, credit and biometrics.

An AI agent is software that uses a language model to complete a task end to end on its own. Unlike a chatbot, it does not wait for a question; it does the work.

How we do it: Our support agent answers the customers of an online retail client within 5 minutes. Customer support now handles only the complex cases.

An AI developer for a project is an external developer who writes code with AI agents and answers for the result. They join your repository for a specific project or backlog, not as an employee. It is development outsourcing, not IT administration.

How we do it: Today AI writes all of our code, and Filip answers for production.

The SDLC (software development life cycle) is the whole development cycle: brief, design, code, tests, deployment and operations. AI-native SDLC means AI agents do the work at every step while people set the rules and check the result. Speed comes from automation, quality from judgment.

How we do it: Today AI writes all of our code, and Filip answers for production.

An API is an agreed interface through which two programs exchange data without a person in between. With an API, an online shop can connect to accounting or the warehouse on its own. Without one, data gets retyped by hand.

CI/CD is an automated line from a code change to deployment. CI (continuous integration) runs the tests on every change, CD (continuous delivery) ships the verified version. Bugs get caught before a customer sees them.

The cloud is servers and services you rent from a provider such as AWS, Azure or Google Cloud. You pay for what you use and never touch hardware. What matters is whose account it runs on and who has access.

How we do it: The infrastructure we build belongs to the client from day one.

Code review is another person checking a code change before it reaches production. They look for bugs, security holes and code the next developer will not understand. With AI-written code, it is the most important quality safeguard.

A CRM is a system for managing customers: contacts, deals, quotes and the history of every conversation in one place. Small companies often use Excel instead, until deals start slipping through it.

Custom software is an application built around one company’s processes, not a box for everyone. You pay for development, not per user. What decides the value is who owns the code, data and infrastructure.

How we do it: Since 2023 we have shipped 15+ production projects. Code, data and infrastructure belong to the client from day one.

Data migration moves records from an old system to a new one with nothing missing and every relation intact. The hard part is not copying but cleaning duplicates and aligning formats. It is rehearsed before the old system is switched off.

Article 20 of the GDPR gives people the right to get their personal data in a machine-readable format and move it to another provider. It covers personal data, not all company data. For company data, your contract with the vendor decides.

An e-invoice is an invoice in a structured format a computer reads without retyping, not a plain PDF. ISDOC is the Czech format for it; the EU standard is EN 16931. In Slovakia, e-invoicing between businesses becomes mandatory in 2027.

An ERP is a business system that joins finance, inventory, production, purchasing and sales in one database. Instead of five spreadsheets, everyone sees the same numbers. An off-the-shelf ERP pays off when your processes fit its logic.

A harness is everything around the AI model that the agent works in: rules, project context, tools, checks and the moment it stops to ask a person. You rent the model; you build the harness yourself. The harness decides whether you can trust the agent's work.

How we do it: Today AI writes all of our code, and Filip answers for production.

An LLM is an AI model trained on a huge amount of text, such as GPT, Claude or Gemini. It writes, summarizes and pulls data out of text. On its own it does nothing, so an agent with rules and tools is built around it.

Low-code and no-code are tools where you assemble an app or automation by clicking instead of writing code. They are enough for simple processes and prototypes. With complex logic, large data volumes or the need to own the code, you hit the platform ceiling.

MCP is an open standard through which an AI model connects to tools and data: a database, a CRM or a code repository. One MCP server works with many AI apps. A company gives AI access to its systems without a custom integration per model.

An MVP (minimum viable product) is the smallest version of a product that real users actually use. It tests whether anyone wants the solution before you spend the budget on every feature. It is not a slide deck or a clickable design.

How we do it: On one project we shipped the first version in under 10 working days.

OCR (optical character recognition) turns an image or scan into text a computer can process. For invoices we add AI on top that pulls out the supplier, amounts and due date. That is called invoice data extraction.

How we do it: In our own accounting, 70 % of invoices go through this way with no person involved.

Process automation means software does repetitive work, such as retyping orders or invoices, instead of a person. People only handle the exceptions. The goal is saved time and fewer errors, not the technology itself.

How we do it: In our own company we cut processing an invoice from 10-15 minutes to under 1 minute.

A prompt is the instruction or question you give an AI model. The quality of the result depends on the context, examples and rules it contains. In business software, prompts are part of the code, versioned and tested.

RAG (retrieval-augmented generation) means the AI first searches your own documents for what is relevant, then answers. The model answers from them, not only from what it learned. It makes things up less and can cite its source.

Refactoring changes the internal structure of code without changing what the software does. The aim is code that is easier to read and cheaper to change. It usually comes before a bigger extension of an app taken over from someone else.

RPA (robotic process automation) is a software robot that mimics a person clicking through applications. It fits where a system has no API. It is fragile: when a screen changes, the robot stops working.

SaaS (software as a service) is software you rent over the internet for a monthly fee, often per user. You start fast and the vendor runs it. But the vendor sets the rules and the data format, not you.

How we do it: Off-the-shelf software cannot do what makes your company different, and you pay for features you never use. That part is what we build.

Source code is the text of a program that the software is built from. Whoever owns it and can access it can fix the software, extend it or hand it to another vendor. Without it, you only have a licence to use it.

How we do it: The software we build is yours from day one, code included.

Technical debt is the shortcuts in code that save time today and slow down every change tomorrow. It grows under deadline pressure and in prototypes nobody cleaned up. You pay it back with refactoring and tests.

Vendor lock-in is when you cannot leave a vendor without high costs or losing data. Typically you do not own the code, the data sits in their format or the contract ignores exit. It is cheapest to solve before you sign.

How we do it: If you leave us, we still fix the bugs from the last month. Code, data and infrastructure are yours from day one.

Vibe coding is building an app by prompting AI without reading the code it produces. It gets you a working prototype fast. For production it lacks security, tests and an architecture that survives growth.

How we do it: Vibe coding gets you a prototype. We build software that runs a business.

A website informs: it presents a company, its services and contact details. A web application does work: you sign in and handle jobs, orders or documents in the browser. An internal system, a client portal or a customer app are web applications, even when they run on a phone.

How we do it: We build only web applications, including mobile apps for iOS and Android from one codebase. We do not build websites.

A crawler is a program that automatically walks through websites and collects data from them, such as prices or listings. The collecting itself is called web scraping. Websites change, so a crawler needs monitoring or it silently stops working.

How we do it: When the crawler of a client with an online marketplace died, traffic fell from 250,000 to 66,000 pageviews a month.

A webhook is an automatic message one system sends another the moment something happens. For example, a shop reports a new order and the warehouse processes it at once. Unlike calling an API, nobody has to keep asking whether anything is new.