AgilePoint转发了
The World Will Never Be The Same When This Inevitable Happens. Agentic AI offers another opportunity for 10X business productivity gain after the introduction of the personal computer. In the thesis, AI use cases are brought closer to the core of business operations from peripheral tasks. This is the key to accelerating ROI for AI investments. However, Enterprise Agentic AI faces seismic challenges. Among them are: 1) Can we trust AI agents to make decisions for us? 2) Is it safe to have AI agents rely on GenAI and LLMs that use public data for training? 3) How can we ensure that AI agents practice safe and responsible AI? AI agency training requires understanding new perspectives: 1) Enterprise proprietary institutional knowledge is an excellent source of knowledge for training agents and enabling a higher level of trust. 2) The best way to train AI agents is to learn about enterprise exceptions and how human experts handle exceptions. The question: Are enterprise exceptions, and how are they handled by human experts captured for AI agency training? Unfortunately, as all IT systems and applications today are inherently deterministic, they are not able to handle exceptions at run-time, and system integrity falters when exceptions occur. As a result, exceptions are still commonly handled offline by human experts as manual workarounds. This resulted that most organizations captured massive amount of the “happy path” data, however, the crucial “exception” data was not captured as an integral part of the system audit trails needed for training AI agency. According to a Gartner 2023 IT Symposium survey, only 4% of CIOs and tech leaders reported they have AI-ready data for training. AgilePoint's abstracted, technology-agnostic composability enables human experts, or AgilePoint GenAI, to rapidly compose 'exception-resistant' workflows and end-to-end orchestrations. They can be dynamically adapted and mass-personalized at runtime to respond to exceptions and AI's non-deterministic nature while ensuring resilient system integrity. The exceptions and the exception handling captured are the most crucial institutional knowledge for AI agency training to accelerate safe enterprise Agentic solutions, paving the way for self-improving AI agency and closed-loop optimization. #Agentic #AgenticAI #AgenticWorkflow #Workflow #Automation #Orchestration #BusinessOrchestration #ProcessOrchestration #Hyperautomation #BOAT #RPA
Jesse, this is spot on. Great insights! Everything mentioned is highly relevant to the shift from task orientation to goal orientation. As you noted, this change pertains to decision intelligence, which today's GenAI and large language models (LLMs) are not designed to support. Below, I've included my complementary assessment: https://www.dhirubhai.net/posts/scotthebner_aiagents-aiassistants-agenticai-activity-7261038409526251520-uxcN?utm_source=share&utm_medium=member_ios
The reality is that on the left side there are a lot of unstructured data that causes friction and largely agents will be applied to make sense of that unstructured data or translate to structured data for humans and systems
This is a fascinating perspective on the future of Agentic AI! It’s exciting to think about how this approach will unlock the next level of productivity and AI innovation in enterprise environments!
Co-Founder of Altrosyn and DIrector at CDTECH | Inventor | Manufacturer
2 周The emphasis on capturing enterprise exceptions for AI training is crucial, as these often reveal the nuances of human decision-making that deterministic systems struggle with. I think bridging this gap between rigid IT systems and the adaptability of human expertise is key to truly effective Agentic AI. How do you envision incorporating real-time feedback loops from exception handling into the continuous learning process of these AI agents?