Every agent we build is classified, documented, and handed over with full transparency.
The EU AI Act requires AI providers to understand what they are building, classify the risk, and give the client everything they need to use it responsibly. This is how Delancy does it on every engagement.
Before We Build
Every engagement starts with a risk classification.
Before any agent is built, we classify the use case against the EU AI Act's four-tier risk framework. The classification is based on what the agent does and who it affects, not the technology behind it.
We check against the eight prohibited practices in Article 5. If a use case falls into a prohibited category, we decline the engagement.
We then assess whether the use case falls into any of the eight high-risk categories in Annex III: employment, education, credit, insurance, law enforcement, migration, critical infrastructure, or justice. The classification determines what documentation and safeguards are required.
Minimal Risk
No mandatory obligations beyond AI literacy. Most internal automation falls here.
Limited Risk
Transparency obligations. Chatbots and content-generating agents must disclose they are AI.
High Risk
Full documentation required: risk management, data governance, human oversight, logging, conformity assessment.
Unacceptable
Prohibited. Social scoring, manipulative techniques, untargeted biometric scraping. We decline these engagements.
During the Build
What we document on every engagement.
01
System description
What the agent does, what data it processes, what decisions it influences, and what its intended purpose is. Written in plain language, not technical jargon.
02
Capabilities and limitations
What the agent can handle, what it cannot, known edge cases, and failure modes. No overselling. No vague capability claims. Specific, documented, testable.
03
Human oversight design
How and when a human reviews the agent's output. How to override or stop the agent. Who is responsible for oversight. Built into the workflow, not added as an afterthought.
Transparency
If it talks to people, it tells them it is AI.
Every chatbot, email responder, or customer-facing agent Delancy builds includes a clear disclosure. The person interacting with the agent knows it is AI. This is not optional and it is not left to the client to implement. It is built into the agent before handover.
At Handover
What you receive when the agent is deployed.
Agent documentation
System description, intended purpose, capabilities, limitations, data handling, and performance characteristics. Everything you need to understand what the agent does.
Human oversight instructions
How to review the agent's output, when to intervene, how to override or stop the agent, and who should be responsible for oversight in your organisation.
Risk classification record
Which risk tier the agent falls into, why, and what obligations that creates for you as the deployer. No ambiguity about where you stand.
Deployer obligations summary
Your responsibilities under the EU AI Act as the deployer of this agent. What you need to do, in plain language, with article references for anything you want to verify.
Transparency disclosure
If the agent interacts with people, the disclosure text and mechanism are built in and documented. You do not need to add anything.
Operational cost transparency
Where the agent uses third-party services to run, those costs are identified, documented, and communicated before the build begins. No surprises after handover.
High-Risk Engagements
When the use case is high-risk, the documentation goes further.
If an agent is classified as high-risk under Annex III, Delancy provides additional documentation covering risk management (Art. 9), data governance (Art. 10), technical specification (Art. 11), automatic logging (Art. 12), and accuracy, robustness, and cybersecurity assessment (Art. 15).
The cost and scope of these requirements are factored into the engagement from the discovery call. They are not an add-on or an afterthought. If the use case requires it, the documentation is part of the build.
Most Delancy engagements are minimal or limited risk. High-risk classification is rare but when it applies, it is handled properly.
Frequently Asked Questions
Common Questions.
Workflow systems are not AI systems. They define how work moves through stages, assign ownership, and provide visibility. The EU AI Act does not apply to workflow systems. This compliance process applies to vertical AI agents and task-specific AI agents only.
We build it with the additional documentation and safeguards required. Risk management, data governance, technical documentation, logging, human oversight design, and accuracy testing are all completed before the agent is deployed. The cost and scope of these requirements are factored into the engagement from the start.
Yes. Once the agent is deployed in your environment, you become the deployer under the EU AI Act. Your obligations include using the agent in accordance with the instructions we provide, assigning human oversight, monitoring the agent during operation, and suspending use if it presents a risk. We document all of this in the handover and walk you through it.
Article 50 requires that people interacting with an AI system are informed they are dealing with AI. We build the disclosure into the agent itself. It is not an optional step and it is not left to you to implement after handover.
Your data stays within your infrastructure. We document what data the agent processes, where it comes from, and how quality is maintained. For high-risk use cases, this documentation is part of the formal data governance record required under Article 10.
Every agent includes human oversight and override capability. If a serious incident occurs, we investigate immediately, notify you, and if the agent is high-risk, report to the relevant market surveillance authority as required under Article 73. Corrective action is taken and documented.
Ready to discuss your project.
Book a discovery call. We will classify your use case and walk you through exactly what the engagement includes.
Book a Discovery Call