Ratio Revolution via Gen AI? Meet the Agentic Workflow
A human wrote this. ??
If you know all about generative AI agents and agentic workflows, please stop reading and accept my apologies for the interruption.
If, however, this is a new topic – and if your AI interests, like mine, run toward business value, i.e., how AI might make things faster, cheaper, and smarter – I invite you to read on.
?
Let us talk, for a moment, about generative AI agents and agentic workflows.
Say it again, and slowly:? generative AI agents and agentic workflows.
These look to pave a path (the path?) to scalable and sustainable enterprise value creation through generative artificial intelligence.
A path to partial or full automation and transformation of business processes – including ones currently touched by machine/deep learning AI algorithms.
A path to make human decision-making faster, cheaper, and smarter.
A path to new benchmarks for financial performance. (Will your firm measure up?)
?
How soon? At today’s rate of AI innovation, this looks to be feasibly implementable (and scalable) within three years. Which means the topic of generative AI agents and agentic workflows belongs on innovation study lists, vendor scouting, and consultancy agendas beginning this year.
?
At the risk of error, allow me to share a layperson’s introduction.
It begins with two elements – a generative AI agent, and a workflow broken down into fine-grained discrete steps.
A generative AI agent is, according to Nvidia, a software entity with complex reasoning capabilities, memory, and the autonomous means to execute specific tasks.? In short, it is trained to ingest and analyze relevant data and act accordingly. Its tasks can range from applying logic (time to reorder), evaluating answers (current stock insufficient to meet anticipated demand), and – important -- coordinating activities across multiple agents.
Envision, next, the second element: a workflow that has been de-composed into multiple, well-defined steps. Each step demands a decision and offers a set of potential actions. Each step is informed by new data; analysis of that data (and data that precedes it) leads to new knowledge; new knowledge leads to a decision, which leads to selection of an action, and then on and on.
Now bring the two together. Put a generative AI agent to work in one or more steps in the workflow. The agent is guided by natural language instructions. (You write a text prompt.) ?It has been trained to do specific tasks, to assess selected data, to make decisions, to take actions,?
It understands (and can deliver on) complex requests. It ingests not only internal ERP data but relevant external data, such as weather, catchment area traffic, trends on Pinterest or Instagram.
It continually learns from its data and could – in time – continually seek routes to greater efficiency. which should lead to optimized outcomes.
Faster, better, cheaper 24/7/365. With never a bad Monday.
And far, far more agile – adapting not only to tomorrow, but tomorrow’s tomorrow.
?
Now take a fourth step: place task-specific agents across the segments of the workflow. And let the many agents begin to work together. And then, in harmony. Analysis at workflow A step 3 informs step A-4 – and informs step 5 in workflow B (store fulfillment.) ??And step 7 in workflow Q (marketing). And the human staffers in customer service (workflow H) and finance (workflow Y).
A Generative AI Network (GAIN) forms, step by step, workflow by workflow. And the business – step by step, workflow by workflow – begins, at an accelerating pace, to know sooner, decide smarter, and act faster.
What benefits might it bring? Let’s turn to this recent article from Medium:
·??????? Increased efficiency within and across workflows. With its ability to connect multiple systems, and connect to multiple data sources, complex and interwoven workflows can be automated.
·??????? Timesaving. Mundane, repetitive work is automated; humans – with human intuition and creativity – can focus on higher-value tasks.
·??????? Adaptable, dynamic operation. Unlike hard-coded programs or even conventional AI, agentic AI has the ability to adapt to evolving circumstances, adjusting plans and decisions in real time.
·??????? Workflow optimization. Agentic AI not only completes tasks, but as noted above, seeks greater efficiency in doing them.
·??????? Enhanced decision-making. Here, the power of vast and deep data analysis.
?
A level of skepticism is merited. It's not today, it's tomorrow. And none of this will be easy. Agentic workflows will be no better than the data they access. And, for most enterprises, an agentic workflow will demand serious integrations into business systems of record, as well as the ability to ingest external, unstructured data.
领英推荐
That being said:
·??????? Global consulting giant McKinsey & Company is today implementing generative AI agent pilots in customer service environments, as recent advances in agent memory structures enable personalization of interactions with both external and internal users.? The results are very promising.
·??????? Microsoft’s Copilot studio, per this InfoWorld note, looks to be evolving beyond chatbots to AI-orchestrated workflows.?
·??????? Salesforce’s recent London event, per colleague Andrew Grant, showed a next-generation Gen AI Einstein, with an agentic workflow driving dramatic customer service value. ?
·??????? An ecosystem of generative AI agents is growing quickly, with firms identified in more than a dozen functions. According to AI seer Ollie Forsyth’s excellent AI Agent Market Map, there are now agents with specific capabilities in data analysis, sales, design and marketing, personal assistance, compliance, workflow automation, voice & video, research, legal, hiring, and customer support – not to mention a host of general purpose agents and firms that offer dev tools and build-your-own agent frameworks and services.?
?
Bottom line:
Generative AI agents and agentic workflows are fast approaching – sooner to us than we to them. They will evolve, improve, and no doubt rise and fall and rise again on the hype curve as the months (nay, weeks) pass by.
And when they arrive, they will transform. They will change financial performance and enterprise ratios.
Irrevocably.
?
I suspect that the best minds in the best companies are already – in silence – working on this.
I also believe that those who wave a dismissive hand will be left further behind.
?
?
Here are some of the sources that informed this article.
?
Bisson, “Copilot Studio turns to AI-powered workflows,” InfoWorld, June 6, 2024
Forsyth, “AI Agents: Part 1,” [email protected], June 5, 2024
Gill, “Agentic AI Driving the Enterprise Workflow: A Paradigm Shift Toward Autonomy and Intelligence,” Akira, May 17, 2024
Gill, “Revolutionizing Artificial Intelligence Through Agentic Workflows,” Xenonstack,com, May 24, 2024
Grant, DX Cubed, private conversations, May-June 2024
Ramlochan, “Agentic Workflows: The Power of Agent Collaboration,” PromptEngineering.com, April 12, 2024
Varshney, “Introduction to LLM Agents,” Nvidia, November 30, 2023
Zhukov, “What is Agentic AI? Understanding agentic AI,” Medium, April 23, 2024
“The promise and reality of gen AI agents in the enterprise,” McKinsey & Company, May 17, 2024
With research assistance from Perplexity.ai.
?
?
I am Jon Stine, 35+ years in retail business and technology. Most recently in conversational AI.
I read, I listen, I observe. I think, I write, I advise.
?+1 503 449 4628.
Lotas Productions-global audio production & voice-over casting. Managed synthetic voice for actors & developers.
3 个月Chin up and eyes forward like riding a bike. From an audio production POV, embracing these technologies will lead us to more engaging ads, efficient content creation, and innovative customer experiences, while maintaining ethical standards will be the key to long-term success. It's always a pleasure Jon.
Data, database, and data warehouse architect (See my website)
3 个月Stimulating thoughts. "A software entity with complex reasoning capabilities, memory, and the autonomous means to execute specific tasks." Two words stand out to me. "Software" distinguishes these entities from humans. (The gap is narrowing.) "Specific" (which Nvidia did not include, but I think you correctly infer) implies that the entity is limited/controlled in some manner. Ala Asimov. As a programmer, I can see how we might enforce such limits. But we will forever be wanting to reduce/remove limits so as to maximize value.
IT Executive Leadership | Customer Centric | Unified Commerce |Strategic Solutions & Delivery | Private Equity | Strategy & Advisory
3 个月There are a number of approaches to this and one of the opportunity areas is in an organization's integration framework. Providers such as Workato, Snap Logic, Mulesoft and others are introducing these capabilities.
Vice President | Sr. Product Manager - Cobrowse, Live Chat, Smart Assistant at U.S. Bank
3 个月Thanks for sharing. Some thoughts that might be valuable. Communication (either through voice or text) requires NLU/NLP which is different than Generative AI (rather a part of) and should not be over looked. Often this is the area domain expertise is required to train and evolve the model(s) essentially providing context to the data consumed. That said, using GenAI for avatar creation (not suggesting you implied this based on article but the picture in post is confusing a bit) inevitably requires explainability, especially if the design is personalized to the individual user based on data known about them. Generating the likeness of a human based on inferences gets dicey to say the least and gets into taboo territory quickly.
Senior Vice President, Data Sciences at Target
3 个月Jon, I don't know if you remember a talk we had on the patio at Target North campus about eight years ago about the future of agents and what each of us might pay to have one working on our behalf (rather than as a trojan horse vacuuming our data).....it's happening. Finally.