Face ID on your phone, Netflix recommendations, a thermostat that learns when you are home: AI has been part of everyday devices for years, and hardly anyone notices it any more. The more interesting question is what changes in 2026 for business owners and professionals. AI has left the demo stage and increasingly runs inside ordinary work: in the inbox, on the phone, in the bookkeeping.
In this piece I go through what is shifting, where it hurts, and how a small or mid-sized business can start sensibly.
From experiment to daily work
Three things come together. The models from Anthropic, OpenAI and Google have become good enough for work where a mistake costs something. You rent computing power per use through an API, so there are no servers to buy. And there are tools to connect AI to the processes you already run, such as Make.com, n8n or a custom integration. Only with all three does AI become something that runs every day, instead of a trial that ends up in a drawer after a month.
The numbers point the same way. In McKinsey's State of AI survey, nearly nine in ten responding organisations say they use AI regularly in at least one business function. Gartner expects 40% of enterprise applications to include task-specific AI agents by the end of 2026, up from less than 5% in 2025. It is still a forecast, but the direction is clear.
What has changed technically
Your data does not have to go to the US
Next to the well-known closed models there are good open models, such as Meta's Llama and the models from the French company Mistral. You can run them on your own hardware or in a European cloud. For a Dutch organisation with sensitive data that makes the GDPR question much simpler: the data stays in-house, or at least in the EU.
Models read far more at once
The context window, the amount of text a model takes in at once, has grown a lot. Many of Google's Gemini models work with a million tokens or more. According to Google, that is enough for the text of eight average-length English novels. A complete case file or a year of correspondence fits into a single request, and that changes what you dare to ask a model.
Limits built into the model
Anthropic, the maker of Claude, trains its models with what it calls Constitutional AI: a fixed set of principles the model uses to check its own answers. For business use, what counts is the behaviour you see in practice. A model that says it does not know something is worth more in a business process than one that confidently makes something up. So always test that on your own examples before you build on it.
Where AI sits inside businesses
In small and mid-sized businesses, three uses keep coming back.
Customer contact. A voice agent that answers the phone outside office hours, a chatbot that takes its answers from your own documents, or a system that sorts incoming mail. Busy periods get absorbed without asking people to work overtime.
Content. Product descriptions, blog posts, social media posts and translations. AI writes a first version and a person finishes it. The dull groundwork goes faster and the writer is still needed.
Processes. Taking data out of one system, running an AI step on it and putting the result into another system, for example from the inbox into Exact or AFAS. Often through Make.com or n8n.
What the three have in common: AI works as a layer between systems you already have. You do not get yet another dashboard.
AI in science and healthcare
Outside business it sometimes moves even faster. Demis Hassabis and John Jumper of Google DeepMind received the 2024 Nobel Prize in Chemistry for AlphaFold2, an AI model that predicts the structure of virtually all 200 million proteins researchers have identified. Among other things, it helps researchers understand antibiotic resistance better. In everyday healthcare things move more slowly, and rightly so: the closer AI gets to a decision about a patient, the stricter the rules. More on that in AI in healthcare.
The other side
Privacy
The GDPR applies to every AI application that processes personal data, whatever a vendor promises. Only use tools that comply with the GDPR, sign a data processing agreement and mention your use of AI in your privacy statement.
Bias in the data
The Gender Shades study from the MIT Media Lab showed that commercial face analysis got the gender of darker-skinned women wrong far more often than that of lighter-skinned men. For IBM, the gap in error rate between those two groups was 34.4 percentage points. A model takes on the bias of its training data. For AI in recruitment this also matters legally: the AI Act classifies that use as high risk.
Who is responsible?
If an AI system makes a wrong decision about a loan, a job application or a diagnosis, who is responsible? The AI Act sets a framework, but the responsibility stays with the organisation that uses the system. For decisions that affect people, always have a person review and sign off. "The system said so" is not an explanation.
What changes next
Voice AI is getting cheaper and more reliable, which puts a phone agent in your own language within reach of a small business. Models handle text, images, audio and video in a single request. Agents plan several steps, use tools and report back on what they did. And you no longer need your own data team to work with it seriously.
What stays the same: AI does the preparatory and repetitive work, you decide. AI spots patterns in your data, you weigh them against what you know about your customers and your market.
What this means for small businesses
The news is about new models and big breakthroughs. For a small or mid-sized business the gain usually sits in something smaller: automation that saves hours every week. A few examples:
- A voice agent that answers the phone outside office hours, books simple appointments straight away and passes everything else on to you or your team.
- Product descriptions written for a whole range in one go instead of one by one.
- Invoices and mail sorted automatically and prepared in Exact or AFAS.
- Enquiries assessed and followed up while you work on something else.
What that looks like for you depends on your processes. A short analysis shows where most of the time goes.
The rules: AI Act and GDPR
Working with AI means working within a new legal framework. The AI Act sorts applications by risk. Article 4 (measures that support your staff's AI literacy, without guaranteeing a specific level) and Article 5 (prohibited practices) have applied since 2 February 2025; Article 4 was amended in July 2026 by Regulation (EU) 2026/1744. The transparency rules in Article 50 have applied since 2 August 2026, and they differ by situation. Providers must build a chatbot or voice agent so that people know they are talking to AI, unless that is already obvious. Providers of AI that generates text, images, audio or video must mark that output in a machine-readable way as AI-generated, unless the AI only assists with standard editing or does not substantially alter the input; providers whose system was already on the market before 2 August 2026 have until 2 December 2026. If you use AI to make a deepfake (images, audio or video that look real), you disclose that it was generated or manipulated by AI. The same applies to AI-generated text you publish to inform the public on matters of public interest, unless a person has reviewed the text and someone holds editorial responsibility for the publication. The heaviest requirements, for high-risk uses such as recruitment and credit scoring, start on 2 December 2027. That date moved later through Regulation (EU) 2026/1744. More in What is the AI Act?.
For most small-business uses it comes down to three things: be open with customers about your use of AI, record where you use it, and have a person check where it matters. DataDream works with GDPR-compliant tools and signs data processing agreements as standard, so you do not have to fix this afterwards.
Start small
In 2026, AI changes work mostly through many small improvements that add up. Whoever starts now learns sooner what works in their own business.
Start with one process, one tool and a one-month trial. Then look at what it delivered and decide: continue, adjust or stop.
Want to know where AI pays off most in your business? Take the free AI Scan for a first analysis.
Curious what AI can do for your business?
Take the free AI Scan and find out in 1 minute.
Frequently asked questions
- How is AI changing work in 2026?
- Mostly through many small improvements that add up. The models from Anthropic, OpenAI and Google are good enough for work where a mistake costs something, you rent computing power per use through an API, and tools such as Make.com and n8n connect AI to the processes you already run. That is why AI increasingly runs every day instead of in a one-off trial.
- What does AI concretely offer small businesses in 2026?
- Usually automation that saves hours every week: a voice agent that answers the phone outside office hours and books simple appointments, product descriptions written for a whole range in one go, mail and invoices prepared automatically in Exact or AFAS, and enquiries followed up while you do other work.
- Will AI replace my work?
- AI does the preparatory and repetitive work, you decide. AI spots patterns in your data, you weigh them against what you know about your customers and your market. For decisions that affect people, such as a job application or a loan, a person should always review and sign off.
- What do the AI Act and GDPR mean for my use of AI?
- Article 4 (measures that support AI literacy, without guaranteeing a specific level) and Article 5 (prohibited practices) of the AI Act have applied since 2 February 2025, the Article 50 transparency obligation since 2 August 2026. The heaviest requirements, for high-risk uses such as recruitment and credit scoring, start on 2 December 2027. For most small-business uses it means: be open with customers, record where you use AI and have a person check where it matters. For the GDPR, use compliant tools and data processing agreements.
