• Written by Azhar Aliyar, Director

A clinical operations manager stares at a screen displaying the hospital's scheduling system, trying to optimise the utilisation of operating theatres while ensuring that the strict privacy obligations of patient data are met. This is a daily challenge for healthcare providers, and one that can be alleviated by the use of agentic AI workflows.

Healthcare provider challenges

Healthcare providers face unique challenges that make the implementation of agentic AI workflows more complex. Scheduling and utilisation of resources, such as operating theatres and medical equipment, are critical to ensuring that patients receive timely and effective care. However, the variability in supply consumption, such as the use of surgical supplies and medications, can make it difficult to predict and plan for future needs. Additionally, the documentation burden on healthcare professionals can be overwhelming, taking away from the time they can spend on patient care.

The use of agentic AI workflows can help to alleviate these challenges by automating routine tasks, such as scheduling and supply chain management, and providing healthcare professionals with real-time data and insights to inform their decision-making. For example, a shift supervisor can use AsscherAi to check the maintenance log of medical equipment, ensuring that it is in good working order and reducing the risk of equipment failure during procedures.

Agent characteristics

So, what separates an agent from a chatbot? The key difference lies in the agent's ability to use tools, maintain state, and work towards a goal. An agent can interact with multiple systems and tools, such as electronic health records and laboratory information systems, to gather information and take action. It can also maintain state, remembering previous interactions and using that information to inform future decisions. Finally, an agent is designed to work towards a specific goal, such as optimising patient flow or reducing hospital-acquired infections.

In contrast, a chatbot is typically designed to answer a single question or provide a simple response to a user's input. It does not have the ability to use tools, maintain state, or work towards a goal, and is therefore limited in its ability to provide meaningful support to healthcare professionals.

Human approval

One of the key considerations when implementing agentic AI workflows is the role of human approval. While it may be tempting to remove the human approval step in order to streamline processes and reduce delays, this is usually premature. Human approval is necessary to ensure that the agent's actions are safe, effective, and aligned with clinical guidelines and policies.

For example, an agent may be designed to automatically order medications or laboratory tests based on a patient's diagnosis and medical history. However, a human approval step is necessary to ensure that the agent's actions are appropriate and do not pose a risk to the patient. By including a human approval step, healthcare providers can ensure that the benefits of agentic AI workflows are realised while minimising the risks.

Failure modes

An agent fails differently from a model, and it is therefore important to log and monitor the agent's actions and outcomes. An agent's failure may not be immediately apparent, as it may be able to continue to function and take action even if it is not operating correctly. However, the consequences of an agent's failure can be significant, and may include patient harm, financial losses, or reputational damage.

By logging and monitoring the agent's actions and outcomes, healthcare providers can quickly identify and respond to any issues that arise. This may involve reviewing the agent's decision-making processes, updating its training data, or adjusting its parameters to ensure that it is operating safely and effectively. For more information on how to implement agentic AI workflows, please visit our website.

Limitations

Agentic AI workflows are not a panacea for all the challenges facing healthcare providers. There are certain situations in which they may not be the best choice, such as when the decision-making process is highly complex or nuanced, or when the stakes are extremely high. In these situations, human judgement and decision-making are still essential, and agentic AI workflows may not be able to provide the necessary support.

Additionally, agentic AI workflows may not be suitable for situations in which the data is incomplete, inaccurate, or biased. In these situations, the agent's actions and decisions may be compromised, and may not reflect the best interests of the patient. If you have any questions about whether agentic AI workflows are right for your organisation, please contact us to discuss your specific needs and challenges.