AI without the buzzwords: what is really behind it.
How a language model computes under the hood, where it takes work off a mid-sized business, where things are heading over the next two years, and how we use AI at MeRa ourselves. No jargon: the terms are on the cheat sheet.
Under the hood: how a language model works
Tools like ChatGPT or Claude are built on a language model. It does not think like a person, it computes. The principle fits into four steps.
Learning from a lot of text
During training the model reads books, websites and program code on an enormous scale. It does not store the texts. Instead it adjusts billions of numbers until it recognises the patterns of language: which words follow each other in which context.
Breaking text into pieces
Your question is split into tokens, small pieces of text made of words or parts of words. Each token becomes a row of numbers the model can compute with.
Predicting the next piece
From everything in front of it, the model calculates how likely each possible next token is. It picks one, appends it and starts the calculation again. Word by word, a whole answer emerges.
Working only with what is there
Whatever is in the context window – your question, attached documents, the conversation so far – the model can use. It only knows your company data if you give it to it.
Example, simplified
Input, split into tokens
Pleasesendtheinvoiceby
Which token comes next?
What AI brings a small or mid-sized business
AI pays off wherever people today read, retype, sort or summarise text. Five examples from everyday office work:
Reading documents
Invoices, delivery notes or forms are read and the details go straight into the right fields.
instead of
retyping paper and PDFsSorting emails
Requests are recognised, routed to the right person and come with a draft reply.
instead of
one inbox that one person works through aloneMaking knowledge findable
Staff ask in plain language and get the answer from your own manuals, contracts and notes, with a link to the source.
instead of
digging through folders and asking colleaguesWriting faster
Quotes, descriptions, minutes and translations are ready as a draft in minutes. A person checks and approves them.
instead of
a blank page every timeBuilding software faster
AI also takes routine work off programming. Custom software is ready for use sooner.
instead of
waiting months for simple features
AI in two years: what changes for SMEs
Nobody can predict the future with certainty. But some developments are clearly taking shape. Our assessment for 2028:
2028Assessment as of October 2026
AI agents take over whole workflows
Instead of answering single questions, AI agents work through several steps in a row: read the request, look up the data, prepare the quote. A person still gives the approval.
AI is built into the software you already use
Less of a separate chat window, more built-in help: in your inventory system, in the customer portal, in your own application.
Asking instead of clicking
You get a report with one sentence (“Which orders have been open for two weeks?”) instead of through nested filters.
More choice on data protection
Capable models increasingly run in European data centres or on your own servers as well. That makes it easier to use them with sensitive data.
Clean data becomes an advantage
AI can only work with data it can find and understand. Whoever moves their workflows from scattered spreadsheets into a proper application today can use them with AI tomorrow.
What stays: people make the decisions.
How we use AI at MeRa
We work with AI every day, as a tool and not as a replacement for responsibility. Four examples from our work:
Programming with Claude Code
We develop with Claude Code, an AI assistant for programming. It writes code suggestions, tests and refactorings. A developer reviews every change before it goes live.
How a change reaches your software with us:
- AI suggests
- tests run
- developer reviews
- live
A knowledge base with sources
We keep agreements, decisions and research in an internal knowledge base. An AI maintains it and backs every statement with its source. That way the context is there for every project, and nothing gets lost between two meetings.
Introducing AI at clients
For an existing client we are guiding the introduction of AI tools in the development team: choosing the tools, costs, data protection and an AI policy everyone can follow.
Design, copy and videos
This website was also made with AI support: design, copy and videos. The ideas, the choices and the final say stay with us.
A person always has the final say.
Where could AI take work off your hands?
Tell us in a few sentences where things get stuck today. Mesut Barut replies personally and tells you honestly whether AI helps there or a simple software feature.
Prefer chat? Message us on WhatsApp