The Future is Agentic
I've been doing a lot of research lately and I keep coming back to a single thought: the future is agentic. My thought crystalized last week when I came across Andrew Ng explaining the concept to venture capitalists and technologists at Sequoia Capital. At this point I felt confident that my thinking is correct.
So what does it mean - that the future is agentic? It means that our current, collective imagination around "AI" is limited in scope to believing that Chat Bots with RAG (Retrieval Augmented Generative) solutions are the primary pattern to follow. Instead, we should consider GenAI's capabilities to understand instructions and take actions on our behalf. Now imagine an army of Large or Small Language Models - all experts at a single task or set of tasks - working in concert to achieve a complex outcome. It's like Robotic Process Automation (RPA) with intelligent bots. You can set up guardrails, make a mini-RAG for every bot specific to its function, and let it run.
Consider AutoGen from Microsoft Research. In their own words, "AutoGen offers a unified multi-agent conversation framework as a high-level abstraction of using foundation models. It features capable, customizable and conversable agents which integrate LLMs, tools, and humans via automated agent chat. By automating chat among multiple capable agents, one can easily make them collectively perform tasks autonomously or with human feedback, including tasks that require using tools via code." This is incredible.
One application of AutoGen is OptiGuide. Researchers from Microsoft built an agentic framework that works with multiple agents to optimize a supply chain.
If you take this concept to its logical end, you could remove the need for a traditional User Interface and only have the user interacting with a group of agents to orchestrate complex tasks in real time. Your hard-working technology teams could shift to data engineering and platform building to equip the agents with the best data and environments possible to accomplish the business' goals. Done right, it could lead to less customization within departments and streamline shared components and data pipelines.
Just this morning, I had a great discussion with our partner OutSystems that inspired me to write this article. AI Agent Builder can accomplish the type of work I've described above. Consider a low code platform that allows you to quickly wire up a set of agents to perform a series of tasks that can be built quickly. Low code allows engineers to quickly develop applications, closely with business users, that can iterate from prototype to value in a short amount of time.
Thanks for reading. To learn more, check out Digineer and OutSystems .
Trailblazing Human and Entity Identity & Learning Visionary - Created a new legal identity architecture for humans/ AI systems/bots and leveraged this to create a new learning architecture
10 个月Hi Garret, I liked your article. What's missing from it is risk. As risk levels rise, as agents doing tasks, then agents require the following: * Identification it's Agent 12345 and not Agent ABCDE * Authentication at the time its Agent 12345 and not Agent ABCDE * Any credentials the agents might have if required based on risk * New enterprise EMS (Entity Management System) entering the agents into the enterprise authoritative source * Authorization allowing the agent to access data, who they can and can't share it with, what they can do with the data, etc.) * Termination of agents as and when required (or the agent might self-terminate in seconds, days, weeks, years or decades) * AI leveraged contracts doing all the above in seconds You might be very interested in articles and architectures I've created addressing the above. That's what the next messages cover. Guy ??
Director Business Analytics & Infrastructure
10 个月Wow man! Well written and quite engaging…. Very excited to see you follow this path…. Even more so to see the products / tools developed