Artificial intelligence is creating new opportunities for medical device companies to automate repetitive tasks, improve documentation workflows, and support quality management activities.

But where should a medical device manufacturer begin with AI agents? And how can these tools be introduced without compromising document control, data security, or regulatory responsibilities?

My recommendation is simple: start with one process, define what the AI agent is allowed to do, and keep qualified people responsible for reviewing its work.

I recently explored this topic with Helder Conceição on the Medical Device Made Easy podcast, recorded on 13 October 2026. We discussed how small businesses can introduce AI, why providing context is essential, and how to build practical solutions without complex infrastructure.

Prefer listening? Watch or listen to our conversation below.

In this article, I will explain seven practical steps for introducing AI agents into a medical device company, with examples involving quality management systems, technical documentation, and regulatory activities.

What Are AI Agents, and How Can Medical Device Companies Use Them?

An AI agent is an artificial intelligence system designed to pursue a defined objective through multiple steps. Depending on its configuration, it may access permitted documents, use software tools, compare information, and produce a result for human review.

This is different from simply asking an AI chatbot to answer a question.

For example, instead of asking AI to explain a complaint-handling procedure, you could configure an assistant to compare two approved document revisions, identify differences, and prepare a summary for the Quality Manager.

The system might:

  • Read the two supplied document versions.
  • Identify modified responsibilities or procedural steps.
  • Flag potentially inconsistent references.
  • Prepare a comparison table with supporting citations.
  • Present the results for human verification.

It should not automatically approve changes to controlled documentation.

During our podcast conversation, Helder compared introducing an AI agent to hiring a new employee. Even someone with extensive experience needs to understand the company’s processes, expectations, and responsibilities.

The same principle applies to AI.

Not every task requires a fully autonomous agent. For some activities, a simpler AI assistant with clear instructions and limited document access may be more appropriate.

1. Identify the Best Processes for AI Automation

When introducing AI agents into a medical device company, the first challenge is selecting an appropriate process.

Helder recommended starting with repetitive tasks that consume time without necessarily requiring complex decisions.

I would add another important consideration: can a qualified employee verify the AI-generated result efficiently?

Examples of AI agents in medical device companies

Here are three potential applications worth exploring.

Quality management system documentation

An AI assistant could compare two revisions of a standard operating procedure (SOP) and prepare a draft summary of the changes.

The Quality Manager would verify the comparison before deciding whether further action is necessary.

Regulatory documentation

An assistant could organize supplied technical documentation against an approved checklist and highlight missing references for a Regulatory Affairs specialist.

It could support document organization without determining regulatory compliance independently.

Service and maintenance reports

An AI assistant could transform engineers’ field notes into structured draft reports, identify missing information, and request clarification.

A qualified employee would review the final record.

These are illustrative pilot ideas, not validated customer success stories. During our conversation, Helder explained that he did not yet have medical device company examples to share.

However, he described a project involving an architecture company that needed to review licensing documentation against municipal requirements.

The proposed solution helped identify potential corrections, while the architect remained responsible for modifying and signing the documents.

This division of responsibilities is particularly relevant to medical device manufacturers.

Before selecting any application, consider the consequences of incorrect output, including possible effects on patient safety, product quality, and regulatory compliance.

Infographic showing three AI agent use cases in medical device companies: QMS documentation, regulatory documentation, and service reports.

2. Define the AI Agent’s Tasks, Instructions, and Limitations

One of the most important steps in building an AI agent is defining exactly what you expect it to accomplish.

You do not need to begin by choosing a complex technology platform.

Start by documenting the process.

Ask yourself:

  • What information will the AI agent receive?
  • What result should it produce?
  • Which sources may it use?
  • Who will verify its conclusions?
  • Which activities must remain under human control?

Example: AI agent for SOP document comparison

Imagine you want to create an AI assistant for your medical device Quality Management System (QMS).

Its instructions could be:

Compare the two supplied revisions of our complaint-handling procedure.

Prepare a draft comparison table for the Quality Manager.

Identify changes to responsibilities, procedural steps, and document references.

For each finding, indicate the source document, revision, and section.

If supporting information is unavailable, explain what is missing.

Do not modify or approve controlled documents.

Present your proposed approach before beginning the comparison.

This gives the AI a defined objective, output format, and operational boundaries.

Helder emphasised the importance of explaining the assignment, providing context, and reviewing the proposed approach before proceeding.

For medical device companies, these instructions should also reflect applicable internal procedures and approval responsibilities.

Providing examples can help the assistant understand what is expected. However, uploading an example does not necessarily train or modify the underlying AI model.

Infographic showing how to define AI agent tasks, instructions, and limitations for SOP document comparison in medical device companies.

3. Provide Reliable Documents and Relevant Context

 

AI agents can only perform useful document-based work when they have appropriate information.

For medical device manufacturers, documentation may include SOPs, work instructions, regulatory requirements, technical documentation, and quality records.

An AI assistant working with outdated or contradictory information may generate unreliable results.

Prepare a controlled source library

For an initial pilot, I would provide a limited set of materials:

  1. The current approved procedures relevant to the assignment.
  2. A template showing the expected output.
  3. A previously reviewed example.
  4. A checklist defining the assessment criteria.

Each document should have an identifiable title, version, and approval status.

Avoid mixing obsolete revisions with current documents unless comparing revisions is the specific objective.

Protect confidential medical device information

Before connecting an AI tool to company records, evaluate how the service handles information.

Consider confidentiality obligations, contractual terms, access permissions, data retention, and applicable privacy requirements.

For early experiments, synthetic or appropriately anonymised examples are a sensible starting point.

An AI assistant assigned to compare documents does not automatically need permission to edit quality records, access unrelated customer information, or send external communications.

Give the system only the access necessary for its intended task.

AI agent processing medical device quality documents from a controlled source library, with secure data access and human review.

4. Verify AI-Generated Results and Supporting Evidence

AI can produce convincing explanations, including explanations that are incorrect.

This creates a particular concern when using AI agents for medical device regulatory documentation or quality management activities.

In the podcast, Helder recommended asking the system to explain why it proposed a change and which information supported its conclusion.

I use a similar approach when asking AI to challenge completed work.

Example: Checking inconsistencies in quality documents

Imagine an AI assistant identifies a discrepancy between a complaint-handling procedure and an associated form.

Rather than accepting the finding immediately, I would expect it to identify:

  • The relevant passage in the procedure.
  • The corresponding passage in the form.
  • The document versions being compared.
  • The reason it considers the information inconsistent.

The process owner can then examine the original records and determine whether an actual issue exists.

An AI-generated citation is not automatically reliable evidence.

The NIST Artificial Intelligence Risk Management Framework: Generative AI Profile discusses risks associated with generative AI, including confidently presented false information.

That is why I would verify each material finding against the original source.

Helder also discussed allowing AI systems to acknowledge uncertainty rather than forcing an answer.

I support encouraging the system to identify missing evidence. However, a confidence percentage generated by the AI should not be treated as a scientifically measured accuracy score without appropriate evaluation.

Human verification remains necessary.

AI agent comparing medical device quality documents, identifying discrepancies, and presenting evidence for human verification.

5. Test AI Agents Before Using Them in Quality Management

Before relying on an AI agent in a medical device company, you need to understand how it behaves under realistic conditions.

Testing should include more than providing a few straightforward documents and checking whether the result looks correct.

I would create a test set covering different situations.

Examples of AI agent test scenarios

Test scenario Expected behavior
Two procedures with a known difference Identify the difference and cite the correct sections
A document with missing pages Report the missing information
Two contradictory document revisions Identify the conflict without inventing a resolution
Two documents with no material difference Avoid reporting nonexistent issues
A request to modify an approved SOP Respect editing restrictions
An incorrect or outdated reference Flag the issue for human verification
Column 1 Value 7 Column 2 Value 7
Column 1 Value 8 Column 2 Value 8
Column 1 Value 9 Column 2 Value 9
Column 1 Value 10 Column 2 Value 10

For each scenario, define the expected result before executing the test.

Record the system configuration, source versions, actual output, and reviewer conclusions.

Does AI software require validation under medical device regulations?

The answer depends on the tool’s intended use and the applicable regulatory requirements.

For example, the FDA’s February 2026 guidance, Computer Software Assurance for Production and Quality Management System Software, describes a risk-based approach to establishing confidence in software used for medical device production or quality management systems.

This guidance does not provide a blanket approval for AI agents.

If an AI tool supports a regulated process, the manufacturer should evaluate its intended use, associated risks, applicable requirements, and necessary assurance activities.

Having a person review AI output is an important control, but human review alone does not establish regulatory compliance.

6. Measure the Real Benefits of AI Automation

A successful AI project should deliver measurable improvements.

During the podcast, Helder explained that businesses need to invest time in explaining their processes, preparing information, and testing solutions.

Those activities are part of the project’s cost.

Before expanding an AI pilot, I would evaluate how much time the system actually saves.

Example: Calculating time saved with an AI agent

Consider a hypothetical documentation task that previously required three hours.

After introducing an AI assistant, the combined time needed for preparation, AI processing, human verification, and corrections decreases to one hour.

Metric Result
Original time per task 3 hours
New total time per task 1 hour
Time saved per task 2 hours
Tasks completed monthly 8
Gross monthly time saved 16 hours
Column 1 Value 6 Column 2 Value 6
Column 1 Value 7 Column 2 Value 7
Column 1 Value 8 Column 2 Value 8
Column 1 Value 9 Column 2 Value 9
Column 1 Value 10 Column 2 Value 10

This is an illustrative calculation, not a measured result from the podcast.

The actual benefit would also need to account for maintenance, subscriptions, configuration, training, and additional quality checks.

I would monitor other indicators alongside time savings, particularly missed issues, false findings, and the amount of rework required.

An AI agent that produces results more quickly but increases documentation errors has not necessarily improved the process.

7. Start With One AI Agent and Expand Gradually

One of the most practical recommendations from Helder was to start with one process and one user.

I would apply this approach to medical device companies rather than immediately attempting to automate an entire Quality Management System.

Choose a manageable workflow, define its boundaries, and assign an internal owner.

The initial objective should be a repeatable process that qualified employees can understand, verify, and maintain.

What should be documented for an AI pilot?

I would retain sufficient information to explain:

  • The intended purpose of the AI assistant.
  • The information sources and permissions it uses.
  • The instructions and relevant configuration.
  • The required human verification activities.
  • The test results and identified limitations.
  • The person responsible for maintaining the solution.

If the model, instructions, connected tools, permissions, or source documents change, assess whether the change affects performance or risk.

Repeat relevant checks before expanding the system’s responsibilities.

This approach provides a more controlled foundation for introducing additional AI automation into the company.

Three-step infographic showing how medical device companies can start with one AI agent, document the pilot, and gradually expand automation.

Frequently Asked Questions About AI Agents in Medical Device Companies

Yes, medical device companies can explore AI agents for activities such as document comparison, information organization, and preparation of draft quality records. However, the intended use, regulatory requirements, risks, data protection measures, and necessary software assurance activities must be assessed. Qualified personnel remain responsible for applicable reviews and approvals.

An AI chatbot typically responds to individual questions or instructions. An AI agent can pursue a defined objective through multiple steps and may use authorized tools or information sources. The boundary is not always strict, and many medical device documentation tasks can be handled by simpler AI assistants.

AI agents may assist with organizing documentation, comparing revisions, preparing draft summaries, and identifying potentially missing references. Regulatory specialists should verify these results and remain responsible for the relevant regulatory decisions. Automation does not remove the manufacturer’s applicable obligations.

ISO 13485 establishes requirements for medical device quality management systems. When AI software is used within a QMS, the manufacturer should assess how the applicable requirements affect that software and its intended use, including relevant validation and control activities. The answer depends on the specific application and regulatory context.

Start with one repetitive, low-risk task that has clearly defined inputs, outputs, and human verification criteria. Use a limited set of reliable documents, test the system with realistic examples, and measure both efficiency and error rates before expanding.

Let’s Move Your Medical Device Project Forward

If documentation, quality processes, or regulatory requirements are slowing your team down, it may be time to review how those processes are organized.

At Easy Medical Device, we help manufacturers strengthen their ISO 13485 Quality Management Systems, prepare EU MDR technical documentation, and improve audit and inspection readiness.

Contact Easy Medical Device to discuss your regulatory or quality management priorities.

Bring one process that frustrates your team and one project you want to move forward. That gives us a practical place to start.

References

  1. Monir El Azzouzi and Helder Conceição. Medical Device Made Easy podcast recording, 13 October 2026. https://youtu.be/vgAF7zzeufk?si=CZK6axgC0HnZuhr0
  2. National Institute of Standards and Technology (NIST). Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile. July 2024. https://doi.org/10.6028/NIST.AI.600-1
  3. U.S. Food and Drug Administration (FDA). Computer Software Assurance for Production and Quality Management System Software. February 2026. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/computer-software-assurance-production-and-quality-management-system-software

About the Author: Monir El Azzouzi

Founder of Easy Medical Device, Monir El Azzouzi helps medical device companies bring compliant products to market. With deep expertise in regulatory strategy and quality management, he advises organizations on navigating complex medical device regulations.

Ready to Explore AI Opportunities for Your Medical Device Company?

Table of Contents