In the current craze about AI, it’s sometimes easy to forget that AI can be used for more than creating chatbots or generating images. Agentic AI or ‘autonomous AI’ brings a different dimension in a sense that agents can make decisions, carry out tasks and learn from interactions. Whereas generative AI relies on human input in the form of prompts or set rules to create specific output, agentic AI is designed to make decisions independently and take proactive actions. Generative AI is focused on creative outputs like images or video. Agentic AI makes decisions or takes actions that are aligned with achieving a specific goal.
“Chaining” an essential element of agentic AI, which enables the AI to perform a sequence of actions in response to a single request. Complex tasks can thus be broken down into smaller steps. Healthcare, for example, is an area where AI agents can introduce significant patient benefits and efficiencies by predicting patient health issues. This prediction is a complex task which the AI agent would tackle autonomously through specific steps:
- Collect patient data — Gathering patient data from electronic health records and combining this with data from other sources (e.g. wearables or genetic info).
- Risk factor analysis — Identifying known risk factors for certain diseases and analysing patient data to see if these risk factors are present.
- Pattern recognition — Using machine learning to detect patterns in patient data, and comparing individual patient data to large data sets of patients with similar patterns.
- Predictive modelling — Developing predictive models for specific diseases and applying these predictions to the individual patient.
- Longitudinal analysis — Tracking changes in patient health metrics over time and using this tracking to identify trends that might indicate decline or that require immediate attention.
- Medication and treatment analysis — Reviewing a patient’s current prescriptions and treatments. Suggesting alternative treatments based on the understanding of the patient medical history, risk factors and other factors.
Examples of Agentic AI in the medical space are Biofourmis’ Biovitals solution which uses AI to analyse data from wearable sensors to detect signs of patient deterioration or Aidoc’s radiology AI platform which analyses medical images such as CT scans and MRI scans to automatically flag critical findings.
Thus far a lot of innovations in the AI space have been based on a single AI Agent interacting with a task or a human. What if you have multiple agents working together to solve complex tasks?! The first step is to enable AI agents to behave more human-like; connecting different pieces of information and applying this information to specific context in which the AI agent operates. Researchers are working on an architectural environment called the Memory Stream which stores all the events that happen in an AI environment. By storing all events and making them easily accessible, the AI agent can use its previous interactions to inform its current actions.
Especially when you have multiple agents interacting with each other it’s important to ensure they’ve got an understanding of previous events, reason and understand current context. There are already a number of platforms that provide a framework for a multi-agent conversation:
- Adept — Adept is an enterprise tool that enables agentic AI, utilising Adept’s in-house models, agentic data and web interaction software.
- LangGraph — LangGraph is an open source orchestration framework for agentic systems (see example from Kamal Dhungana below). It’s an open source framework built on top of LangChain, which provides a standard interface to interact with models and other components, useful for straight-forward chains and retrieval flows. CrewAI is another open source agent collaboration framework built on top of LangChain.
- AutoGen — This Microsoft framework enables the development of LLM applications that use multiple agents to solve tasks together.
Main learning point: I’m genuinely excited about the promise of agentic AI and the promise that it holds for the automation of complex problem solving; agents autonomously working through complex tasks and working with other agents to connect different pieces of information to solve a complex problem.
Related links for further learning:
- https://www.bombaysoftwares.com/blog/agentic-ai-the-future-of-ai-beyond-automation
- https://www.techtarget.com/searchenterpriseai/definition/agentic-AI
- https://www.linkedin.com/pulse/agentic-ai-vs-generative-whats-difference-how-shape-future-pharr-iou2c/
- https://techcrunch.com/2024/10/09/health-insurtech-startup-qantev-raises-e30-million-to-outperform-llms-with-small-ai-models/
- https://medium.com/@carlosrl/how-does-agentic-ai-differ-from-traditional-ai-0e255bb7246c
- https://towardsdatascience.com/how-to-choose-the-architecture-for-your-genai-application-6053e862c457
- https://www.forbes.com/sites/bernardmarr/2024/09/06/agentic-ai-the-next-big-breakthrough-thats-transforming-business-and-technology/
- https://openloophealth.com/blog/real-world-examples-and-applications-of-ai-in-healthcare
- https://www.netguru.com/blog/ai-agents
- https://medium.com/@kbdhunga/langgraph-multi-agent-collaboration-explained-c0500b0f2e61
