EU AI Act Headache
As the EU AI Act moves toward enforcement, practitioners warn that applying its rules to ‘agentic’ AI—tools that act across processes—is messy and unclear. That uncertainty means teams must decide which AI uses are internal productivity aids and which are product features that trigger documentation and governance. For engineering managers, this shifts AI adoption from an engineering decision to a cross-functional compliance one. (artificialintelligence-news.com)
A lot of companies thought the European Union’s Artificial Intelligence Act was mainly about chatbots and deepfakes. In 2026, the harder question is whether an artificial intelligence “agent” that books, routes, approves, or updates things across a business is just an internal tool or a regulated system that needs formal controls. (digital-strategy.ec.europa.eu) The law is already arriving in pieces, not all at once. The Artificial Intelligence Act entered into force on August 1, 2024, bans on prohibited practices started applying on February 2, 2025, rules for general-purpose artificial intelligence models started on August 2, 2025, and the big deadline for many high-risk systems is August 2, 2026. (commission.europa.eu) (digital-strategy.ec.europa.eu 1) (digital-strategy.ec.europa.eu 2) That staggered rollout sounds neat on paper. Inside companies, it creates a sorting problem: the same model can start as a coding helper for employees and end up making decisions inside a customer-facing product, which can move it into a different compliance bucket. (digital-strategy.ec.europa.eu) (dlapiper.com) “Agentic” artificial intelligence is the label people use for systems that do more than answer a prompt once. A simple chatbot is like a calculator that waits for instructions, while an agent is closer to a junior employee that can plan steps, call tools, and keep going until it finishes a task. (dlapiper.com) That is where the paperwork gets messy. The Artificial Intelligence Act regulates uses and roles such as provider, deployer, importer, and distributor, but an agent can blur those lines when one company builds the model, another wraps it in workflow software, and a third deploys it inside human resources, customer support, or finance. (digital-strategy.ec.europa.eu 1) (digital-strategy.ec.europa.eu 2) The European Commission has been issuing guidance because companies kept asking where those lines sit in practice. It published guidelines on prohibited practices on February 4, 2025, and said in December 2025 that the Artificial Intelligence Office was preparing more practical instructions to help businesses apply the law. (digital-strategy.ec.europa.eu 1) (digital-strategy.ec.europa.eu 2) For engineering teams, that changes who gets to say yes. A product manager may want an agent to auto-resolve support tickets, but legal needs to check whether it affects people’s rights, security needs logs and access controls, and compliance needs to know whether the feature falls under transparency or high-risk obligations. (digital-strategy.ec.europa.eu) (dlapiper.com) The internal-versus-product split is becoming the key argument in many companies. A tool that helps an employee draft code or summarize meetings may stay an internal productivity aid, but the moment the same system starts screening job applicants, scoring creditworthiness, or steering access to essential services, the law treats the stakes very differently. (digital-strategy.ec.europa.eu) (digital-strategy.ec.europa.eu) The reason people are nervous now is the calendar. The European Union has already set up the Artificial Intelligence Office and national authorities for supervision, and the full enforcement date for many high-risk systems is less than four months away on August 2, 2026. (digital-strategy.ec.europa.eu) (commission.europa.eu) So the practical shift in 2026 is simple: adding an artificial intelligence agent is no longer just a software decision. In Europe, it is becoming a classification decision first, because the answer determines whether a team is shipping a handy assistant or a regulated system with documentation, oversight, and enforcement risk attached. (digital-strategy.ec.europa.eu) (digital-strategy.ec.europa.eu)
