Our Blog

Discover the limits of Copilot Studio in regulated industries. Learn when low-code bots fall short and how to build compliant workflows under GDPR and AI Act.

Preparing a patent application? Discover how specialized AI agents search worldwide databases, evaluate prior art, and structure claims accurately.

The Dutch Living Environment Decree (Bbl) enforces strict fire safety and energy standards in 2026. Discover how to automate design compliance.

Dutch law firm NautaDutilh is building its own AI applications with Bluefield. A good move. What sits underneath such an application, which part is the same for every firm, and how you will build on the Prudai Developer API, starting with free Dutch legal sources.

At the end of the year you need to clean up your WKR administration to prevent a final tax levy. Discover how AI helps you scan and check accounting rules quickly.

The Model Context Protocol (MCP) is an open standard launched in 2024 to securely and uniformly connect AI models with your corporate data. It prevents vendor lock-in and replaces vulnerable custom integrations.

AI helps a patent attorney as a first reader for novelty, opposition, FTO and EPO communications, citing the source of every statement. The attorney decides, as the April 2026 EPO Guidelines also require.

The Dutch Bar is opening the door to outside capital in law firms. What does that mean for AI, and who stays in control in an AI-first law firm?

Discover how legal AI agents transform due diligence. Learn how to screen data rooms according to current NOvA guidelines and the verification duty in 2026.

Building a Wlz dossier costs a lot of time. Discover how AI in healthcare helps analyze daily reports, demonstrate compliance, and reduce workloads.

A firm that bills by the hour can also sell services as products. Here is how to set a fixed fee per service from your own time tracking data.

With Sign in with ChatGPT, employees use their own ChatGPT subscription in other apps. What that means for the GDPR, ISO 27001, NEN 7510 and NIS2, and which agreements to make now.

Should you choose an AI agent or Microsoft Power Automate? Discover when to use deterministic low-code and when a flexible agentic workflow is required.

Since 1 January 2024, Dutch construction falls under the Omgevingswet: the Bbl sets the technical rules and consequence class 1 adds the Wkb, with a quality assurer, construction notice and completion notice. What architects, contractors and clients must arrange, what the 2026 suspensions mean and what is still to come.

NEN 7510 is mandatory for every Dutch care provider. Which version applies, what the IGJ checks, when the Cyberbeveiligingswet comes in and what AI changes, with a practical checklist.

Human in the loop in practice: where an AI agent fits in a process model, where a person decides, what BPMN, BPMM and DMAIC add and what the AI Act requires.

Dutch lawyers may use AI if they stay responsible. What the NOvA recommendations are and are not, what a disciplinary court ruled in July 2026, what the GDPR and AI Act require, and a checklist for your firm.

The Dutch Cyberbeveiligingswet (NIS2) has applied since 15 August 2026. Find out whether your organisation is in scope and what registration, the duty of care, incident reporting and board duties require.

Since May 29, 2026, the updated Dutch Bbl (stemming from EPBD IV) enforces stricter requirements for EV charging points in parking garages. Because fire safety risks are substantial, quality assurers are leveraging AI to automatically validate complex building designs.

Since August 2026, the Dutch Data Protection Authority (AP) has required stricter oversight of municipal decision-making algorithms. Discover how a well-executed IAMA ensures transparency and how the BEVER tool streamlines this complex administration.

Three Dutch quality assurers have been suspended and colleagues are taking over their projects. Anyone signing off on someone else's assurance needs the file to answer five questions, with the Bkl and Bbl articles to match.

Which half of a Dutch bailiff's work can AI take over, which half cannot, and what does that mean for your firm? A sober look at the KBvG figures.

BEVER put the 2027 draft budget next to the 2026 one and read the explanatory memorandum on the Municipalities Fund. The fund grows to € 49.96 billion, youth care and Wmo appear nowhere as separate lines, climate money for municipalities runs out after 2027 and the fund declines from 2028. With ten questions for the council debate in November, every figure traceable to a budget line or parliamentary paper.

Now that the Dutch Cyberbeveiligingswet (NIS2) has been in effect for a month, the focus around cybersecurity is definitively shifting to the boardroom. How do you prove you meet the duty of care and mitigate personal liability?

A Dutch trend monitor among 799 tax professionals shows AI has become a daily tool, while 88% name trust in its output as the largest barrier. Why verifiability matters more than speed.

In July 2026 a Dutch lawyer was reprimanded, partly because AI-generated references to judgments did not exist or pointed to other judgments. What the court decided, what the NOvA asks and how to check a reference.

In 2026, the AP and the EDPB strongly warn against 'puzzle identification' in Woo decisions. Municipalities must demonstrate they anonymize unique contextual traits, not just names. Traditional methods fall short in this regard.

In 2026, Dutch municipalities face massive backlogs processing Open Government Act (Woo) requests. By intelligently configuring automated anonymization, public organizations save thousands of hours while guaranteeing strict compliance with the Dutch Data Protection Authority (AP).

AI can sort medical records and list heads of damage in Dutch personal injury cases. The lawyer checks every source, every amount and every medical conclusion. How that fits the NOvA recommendations and the GDPR.

If your organisation falls under the Cyberbeveiligingswet, you register in the NCSC entity register via MijnNCSC. What you need, which details the register asks for and what to report within two weeks.

AI assistants do not fix outdated documentation; they present inaccuracies fluently as absolute truth. Why structural information alignment is crucial for responsible AI adoption.

Article 50 of the AI Act has applied since 2 August 2026. Providers mark AI output and disclose chatbots, deployers label deepfakes and certain public texts. Marking by existing systems has a deadline of 2 December 2026.

In 2026, the Dutch Health and Youth Care Inspectorate (IGJ) is strictly enforcing Wabvpz logging requirements. With manual log reviews becoming unsustainable, healthcare providers are turning to AI audits to effectively prevent unauthorized access to electronic health records.

With the Cybersecurity Act coming into effect on August 15, organizations are explicitly responsible for the security of their supply chain. What do the NIS2 supply chain requirements mean for your software vendors, and how do you handle these audits efficiently?

AI agents are fully operational in many organizations, but their underlying data remains fragmented. With the EU AI Act transparency rules and the Dutch Cybersecurity Act applying since August 2026, a centralized data foundation is no longer optional.

AI assistants often invent non-existent ECLIs. We built rechtspraak-mcp: an open-source connector that pulls real Dutch rulings from the official Rechtspraak Open Data into any AI assistant.

Prompt caching reuses the fixed start of a prompt. This is how the Prudai platform, which also runs ZIA, builds a prompt: what is fixed goes first, what changes goes last.

Vector search alone misses exact terms such as article numbers and case citations. Here is how we combine semantic search (Gemini embeddings in Weaviate) with keyword search in Meilisearch.

The requirements for high-risk AI under Annex III of the AI Act apply from 2 December 2027; Regulation (EU) 2026/1744 moved that date. Most obligations sit with the provider. An organisation that uses such a system has its own obligations under Article 26, such as human oversight and keeping logs.

An AI agent can never be liable itself: liability always rests with a person or organisation. There is no separate EU law on AI liability, because the proposal was withdrawn in 2025. For personal injury and private property damage caused by software placed on the market after 9 December 2026, the manufacturer is liable without fault; for business losses, your client will usually hold you, the deployer, to account through contract or tort.

Most organizations have seen enough pilots by now. The next question is more grounded and more important: where does AI create measurable value, how should that value be tracked, and how do you avoid confusing a strong demo with structural return?

The AI-enabled organization is not just about tools. It requires new work design, new team rhythms, and leadership that intentionally combines human expertise with agentic systems.

Public organizations want to use AI to make services faster, more consistent, and more accessible. The challenge is doing that without weakening explainability, human accountability, or data protection.

In manufacturing, engineering, and construction, AI is moving from isolated analysis to direct support for planning, quality control, and operational decisions. Digital twins are increasingly becoming the layer that ties that together.

Private AI is no longer only about security. For European organizations, the choice increasingly revolves around continuity, data control, explainability, and the ability to deploy AI without becoming dependent on one public environment.

Many organizations are experimenting with AI, but far fewer are turning that activity into scalable process improvement. That requires different design choices in workflows, governance, integration, and leadership.

AI literacy is not a soft side topic. Article 4 of the AI Act asks for measures that help employees, managers, and administrators understand how the system works, where the risks are, and how human oversight should operate.

At Prudai, we develop intelligent AI solutions that make organizations smarter and more effective. Together with Symbol, an authority in the field of Lean Six Sigma, we have developed a new product: the Lean Six Sigma AI Consultant.

As an AI coach and AI implementation partner, I see organizations working with AI along two tracks. On one hand, copilots help people within a task step through a prompt (writing, summarizing, classifying). On the other hand, agentic AI: software 'colleagues' that independently translate goals into plans and actions, with quality controls and integrations. The second category provides the greatest added value.

Large Language Models (LLMs) are revolutionizing how we interact with information, but they have inherent limitations. To address these challenges, two powerful methodologies have emerged: Retrieval-Augmented Generation (RAG) and the Model Context Protocol (MCP). While distinct, their true potential is unlocked when used in tandem, creating AI systems that are not only knowledgeable but also capable of taking meaningful action.

An AI agent is software with a language model that chooses the steps towards a goal by itself: it looks up information, uses tools and checks its intermediate results. Agents take over search and preparation work. A person approves the plan and assesses the result.

Generative AI is artificial intelligence that creates new content, such as text, images, audio or code; well-known examples are ChatGPT, Claude and Gemini. The model predicts what is likely to come next, not what is true. It only becomes reliable when you connect it to real sources and a person checks the result.

AI is safe for confidential data when you stay in control: you know where the data is stored, who can access it, whether it is used for training and how long it is kept. Put that in a data processing agreement. At Prudai, documents are stored in Hetzner's EU data centres (databases on our own servers, document files in Hetzner Object Storage) and chat requests and the processing of knowledge-base documents run through EU endpoints.