The hype around GenAI is palpable, but with wins still elusive and costs hard to quantify, defining ROI remains challenging. As AI researchers push for "Agentic" approaches, which focus on individual autonomous agents working on computing around the user. Given my experience as an automation practitioner and advocate for self-healing processes, I see Multi-Agent Frameworks, where multiple agents work collaboratively with employees to drive chains of thought, custody, and reasoning, as a more effective path for enabling sustained success for enterprises and their employees. Do you agree that Multi-Agent Frameworks offer a better internal process handling solution than "Agentic" approaches? What are your thoughts on the future direction of AI in enterprise automation?
Sort By:
Oldest
CEO in Services (non-Government)3 months ago
I am familiar and supportive of multi agent frameworks, when well designed can optimize quality, reduce costs and risk, however … we will always go back to data and its quality and the need to have the human in the loop at the beggining of that multi agent product cycle to fine tune learning. ROI, is the same challenge with new technology, taking an iterative experimentation approach in small batches is a way to 1) show value early 2) learn fast and course correct 3) avoid dumping large amounts at the beginning or committing to plan on still unproven technology and emerging engineering practices. When you learn to ride a bike you use “training wheels” for a reason and you ride in a park under supervision … and you fall, a couple of time until you get it. Same thing with emerging tech and ROI.Executive Manager, Enterprise Architecture and Governance2 months ago
Might be possible. Try it and see. It's too early to have any decent comparison between a multi-agent (at scale) vs. agentic approaches. Also, in my experience, the typical challenges with tech delivery does not disappear, so usual headwinds need solving - integration, precision of understanding the problem, change management and best fit deployment vectors, etc. The data issues bought up by Marcelo can be both a curse and an advantage (depending on the size of your organisation).
One thing is certain - enterprise automation approaches will change. What is still unclear to me is if it's like mainframe to personal computer level of change or cloud software level of change.
Program Director, Intelligent Automation in Healthcare and Biotech2 months ago
From an architecture perspective, my understanding is that multi-agent gives better results since the high level use of a 'dispatcher' chatbot engaging multiple 'department-trained chatbots' will provide more accurate results. The staff member inputs this into the dispatcher chatbot:
"How do I reset my password?"
The dispatcher chatbot then forwards this to the IT, Finance, HR, etc. chatbots.
The IT chatbot comes back and says "to reset your password, please follow the steps on our IT Support Page https://website/reset-password-page
The HR chatbot comes back and says "to get help with access to Peoplesoft, contact HR support"
The Finance chatbot comes back and says "no response".
The dispatcher then defines a confidence score and uses the IT response or maybe the dispatcher combines the top two results if the confidence levels are high enough?