Shaken, not stirred—how Agentic AI is shaking up existing business models
A DMEXCO column by Thilo Kölzer, COO of DocCheck AG, on Agentic AI, MCP, and MPC—and how autonomous, self-learning AI agents are remixing healthcare and marketing while transforming existing business models.
Agentic AI: The Next Big Step After ChatGPT & More
AI has gone mainstream, and just as we were getting used to working with ChatGPT, Perplexity & Co., achieving initial successes and implementing improvements in everyday work life with the help of AI tools, the next game changer is already on the horizon: Agentic AI!
“Agentic AI” refers to AI systems that not only respond to commands and prompts, but also independently pursue goals, plan subtasks, and continuously learn from their actions—comparable to a digital assistant with decision-making and implementation skills.
For example, in the healthcare sector an AI agent could independently review patient data, generate diagnostic hypotheses, consult appropriate specialists, and even adapt therapy recommendations based on the latest study results, while taking all regulatory requirements into account. This opens up enormous potential for more precise and faster care and reducing the workload of medical staff.
MPC and MCP—the revolution lies in these two abbreviations
What is the Model Context Protocol (MCP)?
The Model Context Protocol is a framework or protocol used to define and exchange context information between AI agents. It also regulates the responsibilities and interactions of an agent within a larger agent system. MCP is kind of a “horizon of understanding” for an AI agent. It sounds abstract, but you can think of it as a set of rules or a structure in which an AI agent acts and makes decisions.
Why is such a set of rules important?
To prevent AI agents from acting chaotically or contradicting, a standardized framework is needed—and that is exactly what MCP provides. Such a protocol is comparable to “http”, the “Hypertext Transfer Protocol”—everyone is familiar with “http://www.websiteadress.com”. It provides a standardized framework for displaying web pages in a browser. If this framework would not exist, websites would be displayed differently in each browser or sometimes not at all. The same challenge, namely that of standardization, now also exists in the application of artificial intelligence.
MCP example: Agent network for patient discharge management
The initial situation: A hospital uses a system of autonomous AI agents for discharge management. One AI agent checks the medication prescription, another coordinates nursing services, and a third communicates with the patient’s doctor’s office. These agents must act in a coordinated manner to avoid contradictions and confusion among those involved. For this reason, each agent receives a defined role via the MCP, as well as information about the patient’s condition and treatment plan. For example, if the nursing agent learns that the discharged patient requires insulin but also has mild dementia, it automatically organizes a local nursing service. The AI agents involved exchange relevant information via the protocol without duplicating tasks or conflicting with each other. The potential for reducing the workload of medical staff is enormous—and this is just a very simple example.
What is “multi-party computation” (MPC)?
MPC is a cryptographic method that enables multiple parties to jointly perform a calculation on their respective (confidential) data without disclosing this data to each other. At the end, each party only receives the result—but not the data of the other parties. In relation to multiple AI agents—AI systems that can make decisions and perform actions autonomously—MPC is used to ensure data protection for collaborative AI agents, to make distributed decision-making processes secure, or to protect confidential information between AI agents while they work together to process and solve tasks.
A specific example: Several hospitals want to train an AI model that can automatically evaluate mammograms and thus enable better diagnoses. Each hospital has a lot of image data but is not allowed to share its patient data for data privacy reasons. With MPC, the data can be processed jointly without it being disclosed. The AI model is trained in a decentralized manner by performing (encrypted) partial calculations for each hospital and then “merging” them into a jointly trained model. In the end, all participating hospitals receive the result (e.g. a trained model)—but not the data itself. The larger database nevertheless allows for a significant improvement in diagnostic accuracy.
H2: AI agents in marketing communication and sales
There is also potential for marketing and sales activities in the pharmaceutical and medical technology sectors: experience shows that compiling sales content to be presented to healthcare professionals in sales talks to channel the course of the conversation can be a very complex matter—because a lot of different, sometimes scientific information has to be condensed and summarized.
With a suitable and specially trained AI agent, this task could be partially automated and significantly accelerated. The main advantage, however, would be to ensure quality improvements—both in terms of content preparation and situational adaptation of argumentation aids—depending on who you are speaking with. Based on clinical studies, competitive information, and latest findings, AI could create personalized presentations that adapt in real time during customer meetings to questions, objections to a particular therapy, budget restrictions, or other specifics of the conversation partners, ensuring that the meeting proceeds smoothly.
The preparation and follow-up of sales measures, e.g. regarding to HCP segmentation, content orchestration, and omnichanneling, will also change significantly: AI can analyze prescription data, CRM data, analytics, conference attendance, publications, and perhaps even patient reviews of medical practices, then forms micro-segments and delivers tailored marketing emails, landing pages, CME modules, eDetailer charts, or webinar invitations. For example, if a neurologist opens a whitepaper more often than videos, the AI agent will independently adjust its content mix.
Agentic AI: Turbocharging the Marketing of the Future
My prediction is that over the next 24 months, the entire marketing process from product development to communication, will undergo a complete transformation—with the help of “regular” AI, but above all due to Agentic AI, which act independently as “marketing assistants” and will raise marketing as a whole to a previously unknown level. Whether this is good or bad will ultimately be decided by the target groups and consumers. At the very least, it opens up enormous potential for a hyper-personalized and hyper-relevant world of communication.
