Artificial intelligence is gradually taking hold in customer service, marketing, sales, human resources and internal processes. Conversational agents, chatbots, AI assistants and generative AI tools are now used by a growing number of companies.
To accompany that shift, the European Union has adopted the AI Act, the first regulation dedicated to artificial intelligence. 2 August 2026 marks an important milestone in its application: several new provisions become applicable and the supervision mechanisms are reinforced.
For companies using AI agents, this is not about questioning their projects, but about making sure those projects are built and used in a way that is transparent, responsible and documented.
Not every automation is AI
One of the most useful contributions of the AI Act is that it finally clarifies the definition of an artificial intelligence system. Contrary to a widely held idea, not every automation counts as AI under the regulation.
Under the AI Act, an AI system is notably one capable of inferring results from the data it receives: understanding a request, generating an answer, making a recommendation or producing content with a degree of autonomy. Conversely, software that only applies predefined rules is generally not considered an AI system.
These uses generally do not count as AI:
- automatically sending a WhatsApp message after a purchase
- triggering a marketing follow-up after a few days
- running a Make, n8n or Zapier workflow based only on rules
- launching an automated sequence in a CRM
These uses, on the other hand, fall within the scope of the AI Act:
- understanding a request written in natural language
- automatically qualifying a need
- rephrasing an order or an agreement
- searching a document base
- generating a personalised answer
- preparing a quote, a contract or a document from a conversation
The distinction matters. Artificial intelligence brings real value when it makes it possible to understand, reason or generate content. Where a classic automation is enough, it usually stays simpler, more reliable and cheaper.
What the AI Act means for conversational uses
Most conversational agents built with Sandra use language models and therefore fall within the scope of the AI Act. Four measures deserve particular attention.
1. Tell users they are interacting with an AI
The AI Act sets out a transparency obligation where people interact directly with an AI system, except where that is already obvious. A simple line at the start of the conversation is usually enough:
Hello, I'm [Company]'s AI assistant. I'll help you with your request. You'll always be able to review and confirm the information before it is sent.
That transparency meets the regulatory requirement while strengthening user trust.
2. Provide human oversight where it is relevant
AI agents can qualify a prospect, prepare an order, generate a quote, answer questions or assist an employee. Where a process involves a contractual commitment, a sensitive decision or a complex situation, however, it is advisable to provide for human validation suited to the context.
3. Document the AI agents
The GDPR led companies to set up a register of personal data processing. The AI Act is gradually pushing towards a comparable approach for AI systems. Every company should be able to identify:
- the AI agents deployed
- their purpose
- the models used
- the data handled
- the tools they are connected to
- the business owner of each agent
Beyond the regulatory aspects, this documentation is a genuine governance tool and makes maintenance, audits and future changes easier.
4. Train the teams
The AI Act also introduces an AI literacy obligation. Employees who use or supervise AI systems must have a level of knowledge suited to their responsibilities: understanding what AI can do, its limits, the main risks and the good practices to adopt.
Towards an AI Control Center
The AI Act confirms a deeper shift: tomorrow, companies will not only need to create AI agents, but also to inventory, document and govern them.
That is the thinking behind the next developments of Sandra, built around an AI Control Center designed to give a central view of the AI agents configured in the platform, of their documentation and of their governance. Our ambition is to bring that approach close to the principles of the GDPR, so as to offer, in time, a unified register of data processing and AI agents.
In conclusion
The AI Act is an important step in the maturity of artificial intelligence in business. For the vast majority of organisations, it is not a brake on innovation but a framework for building uses that are more transparent, better documented and better governed.
Going further
Read next
Conversations that convert
A 30-minute demo is enough to see what your WhatsApp channel can produce. In Geneva or by video call.