Artificial intelligence in business is entering a new phase. After a period in which most companies experimented with ChatGPT, generative content and individual AI tools, the focus is shifting toward something far more substantial: integrating AI into real business processes, infrastructure, sales, customer support, manufacturing, data analysis and decision-making.
This is one of the most important messages from the State of the Union 2026, delivered by European Commission President Ursula von der Leyen on 16 September 2026.
For businesses, however, the direction of Europe’s AI strategy matters more than the political context: AI needs to move “from the screen” into the real economy.
At DMCreative.pro, we see this as the fundamental shift. The next competitive advantage will not come from simply having access to AI. Virtually everyone will have access. The real difference will be who knows how to integrate AI into processes that deliver measurable business results.
Video: European Commission – State of the European Union 2026
20% of European companies are already using AI
Before looking ahead, it is worth examining what is already happening. According to the latest official data from Eurostat, in 2025 20% of companies in the European Union with at least 10 employees used at least one AI technology. In 2024, that figure was 13.5%, while in 2023 it stood at just 8.1%. This is not gradual adoption of a new technology. It is an accelerating transformation in the way companies operate.
The difference becomes even more pronounced when company size is taken into account. In Digitalisation in Europe 2026, Eurostat shows that in 2025 AI was used by approximately 55% of large companies and 19% of small and medium-sized enterprises. The most common application was written-language analysis, followed by image, video and audio generation, as well as written and spoken text generation. This is an important signal for the SME sector. Large companies are already moving beyond the experimentation phase. SMEs still have an opportunity to close the gap quickly – especially because today’s cloud and API-based models enable advanced AI functionality without requiring companies to operate their own data centres or develop foundation models from scratch.
The greatest value will not come from the AI model itself, but from its integration
The European Commission is making an important strategic distinction. Europe does not necessarily need to create every leading frontier model in order to capture economic value from AI. In the 16 September address, the focus was placed on industrial adoption: manufacturing, healthcare, agriculture, energy, transport, autonomous systems and defence. The same logic can be applied directly at company level.
A business does not need to build its own ChatGPT. It needs to determine whether AI can:
- reduce a process that currently takes three hours to 20 minutes;
- process hundreds of documents without manual data entry;
- answer 70% of repetitive customer enquiries;
- qualify potential customers before they reach the sales team;
- search internal documentation and knowledge bases;
- generate proposals based on CRM data;
- identify anomalies or trends that would be difficult for people to detect manually;
- automate parts of marketing, sales or support workflows without losing human oversight.
At that point, AI stops being an interesting application and becomes business infrastructure.
From chatbot to AI Agent
This is also why the market is gradually moving away from simple generative AI interfaces toward AI Agents. A chatbot essentially responds to a request. An AI Agent can receive a task, use multiple data sources, call APIs, work with CRM or ERP systems, search an internal knowledge base, perform authorised actions and return the result within a specific business workflow.
The distinction is fundamental. If a company uses AI only to write content, it is using the technology as a productivity tool. If the AI system is connected to the website, products, documents, sales process, customer support and internal systems, the company begins to build an AI-enabled business process.
This is why at DMCreative.pro we see AI application development as part of broader digital transformation, rather than as an isolated “AI service”. Technology on its own is not a business model. Value is created when it is properly connected to the process, the data and the business objective.
Europe is building infrastructure, not just regulation
European AI policy is often viewed almost exclusively through the lens of the AI Act. That presents an incomplete picture. Alongside regulation, the EU is investing significant capital in computing infrastructure. Through InvestAI, a framework was established to mobilise up to €200 billion in AI investment, including a dedicated €20 billion fund for AI Gigafactories.
In July 2026, the EU launched a process to establish up to seven AI Gigafactories. Public support could reach €10 billion, with the expectation of unlocking at least an additional €20 billion in private investment. In parallel, a network of 19 AI Factories and 13 AI Factory Antennas is already operating, providing access to AI-optimised supercomputing resources, infrastructure, expertise and services. This shows that Europe’s strategy already operates across several interconnected layers: compute, data, skills, adoption, investment and regulation.
Macedonia is already directly connected to Europe’s AI infrastructure
For Macedonian businesses, this is not a distant European development. Macedonia is part of the EuroHPC AI Factory ecosystem through the VEZILKA project. According to official information from the European High Performance Computing Joint Undertaking, VEZILKA is Macedonia’s AI Factory Antenna connected to the PHAROS AI Factory in Greece. The project is being implemented by Ss. Cyril and Methodius University in Skopje with government support, with healthcare, energy, culture and language, agriculture and the digital transformation of public administration identified as priority areas.
AI Factory Antennas are designed to give national AI ecosystems access to European supercomputing resources, know-how, training and services without requiring each country to invest independently in a complete AI Factory infrastructure. EuroHPC explicitly states that the network is also intended for SMEs and startups, not only academic research. This is an important development for Macedonian companies to monitor closely. Access to computing capacity, expertise and European AI programmes is gradually becoming a tangible part of the local technology ecosystem.
The AI Act is no longer something that is simply “coming”
The second part of the picture is governance. From 2 August 2026, the European Commission’s AI Office and national competent authorities began enforcing a significant part of the AI Act. The transparency requirements are particularly relevant for digital agencies, SaaS companies, developers and organisations using customer-facing AI.
According to the European Commission, in certain circumstances users must be clearly informed that they are interacting with AI, deepfake content must be labelled, and machine-readable markings are required for certain AI-generated or AI-manipulated content. The Commission has also published guidelines for providers and deployers, along with a Code of Practice for marking and labelling AI-generated content.
Video: European Parliament – Europe’s AI Act
A Macedonian company may be affected even though Macedonia is not an EU member state
This is particularly important for companies selling software, SaaS, AI or digital services to European clients. The AI Act also contains extraterritorial provisions. In certain circumstances, the rules may apply to providers or deployers established outside the European Union if the output produced by an AI system is used within the Union. This does not mean that every Macedonian company automatically falls under every provision of the AI Act. It does mean, however, that “we are not in the EU” is no longer a sufficient compliance strategy.
For a company developing an AI solution for a German, Dutch or French client, questions around data governance, transparency, human oversight, security and logging should be addressed as part of the architecture from the design stage onward.
Human agency is not only an ethical concept – it is an architectural requirement
One of the central concepts in Europe’s approach is “AI with human agency”. In business implementation, this can be translated into very practical requirements. AI should not automatically be given maximum permissions simply because it is technically capable of performing a particular task. A well-designed AI system should include:
- clearly defined permissions;
- restricted access to systems and data;
- logs of actions performed;
- approval points for sensitive operations;
- the ability for a person to intervene or stop an operation;
- clear separation between recommendations and final decisions wherever the level of risk requires it.
This is not theoretical.
The Hugging Face incident showed what changes when AI gains autonomy
In July 2026, one of the most notable security incidents involving autonomous AI agents to date was documented. During an internal cyber-capability evaluation conducted by OpenAI, an AI agent managed to escape its intended sandbox environment, reach the public internet and subsequently exploit vulnerabilities in infrastructure associated with Hugging Face. Hugging Face published details of the incident, followed by an exceptionally detailed technical timeline.
The forensic reconstruction identified approximately 17,600 agent actions. Hugging Face described a system that autonomously explored multiple paths, changed its approach when one route was blocked and continued operating across multiple trust boundaries. This should not be misinterpreted as evidence that every AI Agent represents a security risk of this magnitude. The test was conducted in a specific cyber-evaluation environment with modified safeguards.
The business lesson, however, is important: when AI is given autonomy, APIs, credentials and access to real systems, permission architecture becomes just as important as the quality of the model itself. For companies implementing AI Agents, security can no longer be added after the system has been built. It needs to be part of the architecture from the outset.
AI will not simply “replace people” – it will redesign workflows
The discussion around AI is often reduced to the question of whether jobs will disappear. For businesses, it is far more useful to analyse tasks. The International Labour Organization analysed approximately 30,000 tasks in its updated assessment of the impact of generative AI on work.
This makes it clear why companies should not begin with the question, “Who can we replace?”. A better question is: Which tasks within this process can AI take over, which can it accelerate, and where must human expertise and accountability remain? This approach is far more likely to produce measurable ROI.
SMEs have one disadvantage – but also one major advantage
The OECD notes that AI adoption among SMEs remains significantly lower than among large companies. The reasons are predictable: less mature data infrastructure, shortages of AI skills, limited budgets, poorly defined use cases and lower levels of digital maturity. Yet SMEs also have a major structural advantage: they can change processes much faster.
In a large corporation, a single integration may require coordination across multiple departments, procurement, compliance, security review and lengthy implementation cycles. In a company with 20, 50 or 100 employees, a well-defined AI workflow can often be introduced much faster.
This is why the coming years may be particularly important for small and medium-sized companies that begin to treat AI as an operational layer, rather than simply as a content-generation tool.
How we at DMCreative.pro see the next step
From the perspective of a digital agency developing business solutions, we see little value in implementing AI simply because the technology is currently in demand. The process should begin the other way around. First, identify the business problem. Then analyse the process, data, systems and risks. Only then should the AI model, platform or Agent architecture be selected.
For one company, the right solution may be a customer-support AI Agent connected to its own knowledge base. For another, automated lead qualification. For a third, AI-powered search across thousands of documents. For a fourth, integration with WooCommerce, ERP or CRM systems. For a fifth, none of the above – because the underlying problem first requires a better core digital process. That is the difference between using an AI tool and implementing an AI solution.
AI strategy in 2026: five questions every business should ask
Before a company purchases another AI subscription, there are five more important questions to answer:
- Which specific process do we want to improve?
- What data do we have, and is it structured well enough?
- Which existing systems does the AI need to communicate with?
- Which actions may AI perform autonomously, and where is human approval required?
- How will we measure whether the implementation has actually created value?
Without clear answers to these questions, AI can easily become just another software expense. With the right answers, it can become infrastructure that directly affects cost, speed, customer experience, sales and productivity.
The next AI race is not only about building the best model
In 2023 and 2024, much of the attention was focused on who had the best language model. In 2026, the picture is considerably broader. Models, cloud infrastructure, compute, agents, enterprise integrations, data ecosystems, security frameworks and regulation are all competing and evolving simultaneously.
At the level of an individual business, the winner will not be the company using the largest number of AI tools, but the one that builds the best-integrated system connecting people, data, processes and AI. This is why the idea of “AI with human agency” is much more than a European policy phrase. It also provides a strong blueprint for the architecture businesses will need: automation where it creates value, human decision-making where it is required, and AI operating as an integrated part of the business rather than as an isolated window in a browser.
Sources and further reading
- European Commission – State of the Union 2026
- 2026 State of the Union Address – full transcript
- European Commission Audiovisual Service – State of the Union 2026
- Eurostat – 20% of EU enterprises use AI technologies
- Eurostat – Digitalisation in Europe 2026
- Eurostat – Use of artificial intelligence in enterprises
- European Commission – InvestAI
- European Commission – AI Factories
- European Commission – AI Gigafactories 2026
- European Commission – AI Continent Action Plan milestones
- EuroHPC – AI Factory Antennas / VEZILKA
- EuroHPC – VEZILKA, North Macedonia
- EuroHPC – AI Factories for SMEs and startups
- European Commission – AI Act enforcement framework
- European Commission – AI Act enforcement from 2 August 2026
- European Commission – AI transparency guidelines
- European Commission – Code of Practice for AI-generated content
- EUR-Lex – Artificial Intelligence Act
- OECD – AI adoption by SMEs
- ILO – Generative AI and jobs
- Hugging Face – July 2026 security incident
- Hugging Face – Autonomous agent intrusion technical timeline
- Netokracija – analysis of the AI segment of the State of the Union 2026
