Pitfalls for Agentic AI

As we look at the first artificial intelligence applications in healthcare, we have seen unprecedented advancements in our ability to collect and interrogate data. This has given rise to companies like Empatica, who are deploying inference at the edge and Optellum, whose radiomics solution is enabling earlier lung cancer diagnosis. As we move towards Agentic AI, we are seeing solutions that take direct and proactive actions that have the potential to further revolutionize the patient journey.  

As we sit at the precipice of the next generation of the AI revolution, we must ask ourselves of the pitfalls to deployment and sustained implementation, i.e. the “P.I.T.S.. Here are a few with implications to the near and long term:  

• People: Who is going to design, develop, and deploy these agentic systems? Will the workforce be able to work with these systems effectively? What happens if the system goes down? Do we have human backup? 

  
  

• Infrastructure: Do I have enough compute infrastructure and money to implement agentic AI? At scale, how do I ensure continuous uptime at the lowest cost and highest resilience? How and when do I have to upgrade my compute infrastructure?

  
  

• Trust: Which agents do I pick? How do I know what is inside the system? Who do I trust to orchestrate and manage additions and updates to approved agents? How do I ensure that these agents do not put me out of business? 

  
  

• Security: How do I avoid agentic ai being the cause of more ransomware attacks? How do I allow differential private across agents and ensure cybersecurity at scale? How do I safeguard against post-quantum encryption?  

  
  

Black Opal has taken a strategic approach to investing in companies addressing these areas. Blaze.Tech has built a platform that allows non-technical staff to build customized agentic systems that are HIPAA compliant, solving some of the problems with the human bottleneck. Hyro has built responsible AI agents for healthcare that is built around the core pillars of clarity, control, and compliance, enabling organizations to deploy trustworthy conversational AI. Parasail’s first-of-its-kind AI Deployment Network provides fast, scalable, and cost-efficient AI compute for teams tackling large-scale AI workloads. Lastly, AuthMind is offering protection for both human and nonhuman identities enabling security at scale.  

Agentic AI holds immense promise for healthcare, from automating administrative tasks to clinical decision making. Realizing this potential requires navigating the complex web of technical, regulatory, ethical, and organizational bottlenecks. Companies leading the way are those that build solutions to address these pitfalls.