Generative AI What Is Agentic AI, and How Will It Change Work?
Summary.
From the early days of mechanical automatons to more recent conversational bots, scientists and engineers have dreamed of a future where AI systems can work and act intelligently and independently. Recent advances in agentic AI bring that autonomous future
...The way humans interact and collaborate with AI is taking a dramatic leap forward with agentic AI. Think: AI-powered agents that can plan your next trip overseas and make all the travel arrangements; humanlike bots that act as virtual caregivers for the elderly; or AI-powered supply-chain specialists that can optimize inventories on the fly in response to fluctuations in real-time demand. These are just some of the possibilities opened up by the coming era of agentic AI.
While previous AI assistants were rules-based and had limited ability to act independently, agentic AI will be empowered to do more on our behalf. But what, exactly, is agentic AI? “You can define agentic AI with one word: proactiveness,” said Enver Cetin, an AI expert at global Experience Engineering firm Ciklum, whom I interviewed. “It refers to AI systems and models that can act autonomously to achieve goals without the need for constant human guidance. The agentic AI system understands what the goal or vision of the user is and the context to the problem they are trying to solve.”
To achieve this level of autonomous decision-making and action, agentic AI relies on a complex ensemble of different machine learning, natural language processing, and automation technologies. While agentic AI systems harness the creative abilities of generative AI models such as ChatGPT, they differ in several ways. First, they are focused on making decisions rather than on creating content. Second, they do not rely on human prompts, but rather are set to optimize particular goals or objectives, such as maximizing sales, customer satisfaction scores, or efficiency in supply-chain processes. And third, unlike generative AI, they can also carry out complex sequences of activities, independently searching databases or triggering workflows to complete activities.
The Benefits of Working with Agentic AI
With their supercharged reasoning and execution capabilities, agentic AI systems promise to transform many aspects of human-machine collaboration, especially in areas of work that were previously insulated from AI-led automation, such as proactively managing complex IT systems to pre-empt outages; dynamically re-configuring supply chains in response to geopolitical or weather disruptions; or engaging in realistic interactions with patients or customers to resolve issues. Three of the main benefits will be greater workforce specialization, greater informational trustworthiness, and enhanced innovation.
Greater specialization
The importance of workforce specialization — the “division of labor” — has been understood since Adam Smith’s celebrated pin-factory visit in the opening paragraphs of The Wealth of Nations. Smith observed how one worker “draws out the wire, another straights [sic] it, a third cuts it; a fourth points it…” such that “the important business of making a pin is, in this manner, divided into about eighteen distinct operations.” Specialization brings greater efficiency, learning by doing, and innovation — but can be difficult to implement as businesses run up against workforce shortages and mismatches between roles and available human skills. Because agentic models are explicitly designed to carry out very granular tasks, they enable much greater specialization of roles compared with previous broad-brush automation systems. What’s more, multiple agentic roles can be created rapidly. In knowledge work, for example, agents can be created for information retrieval, analysis, workflow generation, and employee assistance — all working in tandem. Some AI agents will also work “behind the scenes”, orchestrating the work of other agents, just as human managers do for their teams.
Innovation
With their enhanced judgement and powers of execution, Agentic AI systems are ideal for experimentation and innovation. For example, ChemCrow, an AI-powered chemistry agent, has been used to plan and synthesize a new insect repellent as well as to create novel organic compounds. Multi-agent AI models can also scan and analyse vast research spaces — such as scientific articles and databases — in a fraction of the time it would take teams of human scientists and researchers. SciAgents — a multi-agent model developed by researchers at MIT — includes not only robot scientists to develop research plans, but a Critic Agent to review these and suggest improvements. Working together, the team of AI agents was able to identify a novel bio-material combining silk and dandelion-based pigments that had better mechanical and optical properties compared with similar materials, with a reduced energy input to boot.
Greater trustworthiness
The greater cognitive reasoning of agentic AI systems means that they are less likely to suffer from the so-called hallucinations (or invented information) common to generative AI systems. Agentic AI systems also have significantly greater ability to sift and differentiate information sources for quality and reliability, increasing the degree of trust in their decisions. For example, while customer information is often scattered in different formats across different parts of a business — emails, databases, spreadsheets, and the like — an agentic AI system can quickly discern that the most reliable and up-to-date information is likely to be in the firm’s customer relationship management (CRM) systems. Agentic systems are also designed to quickly learn a company’s human and brand values, ensuring that these are aligned with decisions and actions.
Potential Use Cases
While many applications of agentic AI are still experimental in nature or at a pilot stage, the broad contours of potential use cases are already starting to emerge across different industries and functions. Some examples include:
Customer service
In contrast to traditional automated customer bots that were pre-programmed with a limited range of responses and actions, agentic customer service agents can quickly grasp customer intents and emotions and take independent steps to resolve queries and problems. For example, an agentic customer service agent could predictively assess whether a customer delivery is going to be late, inform the customer of the delay, and proactively offer a discount to sweeten the disappointment. Ema, an AI startup based in California, offers agentic AI chatbots that can dynamically trawl thousands of different databases and apps to resolve customer queries and complaints, learning from each customer interaction and identifying recommended actions for human agents. Ema also audits its content for accuracy and compliance purposes, while also making recommendations to improve the customer knowledge base.
Manufacturing
From controlling the flow of production lines to customizing products to making suggestions for improved product design, agentic AI is likely to have multiple applications in smart manufacturing. Data from sensors attached to machines, components, and other physical assets in factories and transportation can be analyzed by agentic AI system to predict wear-and-tear and production outages, avoiding unscheduled downtime and associated costs to manufacturers. German AI start-up Juna.ai deploys AI agents to run virtual factories, with the aim of maximizing productivity and quality while reducing energy consumption and carbon emissions. It even offers agents tailored to specific goals, such as production agents and quality agents.
Sales support
For sales agents, the critical goal of finding and developing sales leads can often be swamped by a mass of emails, paperwork, and other mundane but necessary administrative tasks. Agentic AI systems could dramatically liberate sales teams from much of this time-consuming activity. CRM technology titan Salesforce, for example, recently introduced its Agent Force Service Development Rep to assist the work of human sales teams. Powered by large-language models (LLM), the agent can interpret customer messages, recommend follow-up actions, book meetings, answer questions, and generate responses that are attuned to the company’s brand voice. Complementing these activities is the Agent Force Sales Coach, providing personalized feedback to human agents and opportunities for learning through virtual role-play sessions.
Health and social care
Their ability to adapt to different settings, interpret human emotions, and show empathy makes agentic AI systems ideal for non-routine, soft-skills work in areas such as healthcare and caregiving. Hippocratic AI, an agentic AI healthcare company based in California, has created a phalanx of AI agents tailored to different areas of healthcare and social support. The team counts among its ranks Sarah, an AI agent who “radiates warmth and understanding” while providing help with assisted living. Sarah can ask patients about their day, organize menus and transport, and regularly remind patients to take their medication. Judy, another AI-powered agent, helps patients with pre-operative procedures, for example by reminding patients about arrival time and locations, or advising on pre-op fasting or stopping medications.
The Challenges Ahead
Despite significant potential to transform human-machine collaboration and drive greater efficiency and business growth, agentic AI systems are still at a relatively early stage of development. Moreover, despite their greater powers of reasoning and execution, they do not remove traditional workforce management challenges; instead, they change them. Just as in traditional, human workforce settings, managers must still pay heed to issues of team composition and role selection, and they must set the right overall goals to ensure that agentic AI or hybrid teams can be successful. They must also carefully calibrate the conditions under which agentic AI systems can be trusted to make decisions and the circumstances in which human decision makers need to intervene.
Imperatives for Success
To capitalize on the opportunities of agentic AI while mitigating the risks, managers should consider the following imperatives:
Set SMART goals
Just as the performance of human teams can be stymied by poorly defined or badly articulated goals, so too can agentic AI systems go off track if goals are not set clearly. In fact, goal-setting becomes even more important for agentic AI, as the systems initially lack the contextual information — such as organizational and market context, company values, and so forth — that is often tacitly understood by human workers. Ciklum’s Cetin underlines the importance of comprehensive goal setting: “For agentic AI to succeed, the models must have SMART (specific, measurable, achievable, relevant, time-bound) goals and sub-goals and know how to measure them. They must have the right contextual information — why are these goals important to the company, how do they drive revenues, etc. Finally, as managers, we need to establish feedback loops to adjust the models as we learn more about their performance.”
Pay attention to team selection
Compared to generative AI — which is largely based on prompting large-language models with singular questions — Agentic AI is much more of a team endeavor, making use of multiple AI agents, all of whom have specific roles to play in achieving a greater goal, be it maximizing customer experience or innovating a lower-cost business process. Just as in human teams, problems of coordination, conflict, and resource management are likely to arise. Managers using agentic AI systems will need to pay careful attention to team selection, ensuring that they have the right combination of agentic roles carrying out the right tasks, in an efficient way. Furthermore, they will need to carefully consider how agentic teams interplay with human workers to achieve trust and efficiency in activities.
Scaffold the decision space
While agentic AI models are explicitly designed to evaluate decision choices and carry out complex sequences of actions, they are not foolproof and can still make mistakes, just as humans do. Learning science highlights the importance of “scaffolding” in learning, giving learners exposure to real-world practice with safeguards — supervision, well-defined limits, etc. — which are then progressively withdrawn as experience grows. Such scaffolding will be essential as agentic AI systems are applied to different tasks and business areas, with decision-makers constructing appropriate scaffolding for these models based on factors such as the criticality of the decision, the consequences of mistakes, the degree of confidence in the data used to train the models, the degree of human supervision, and the experience profile of the humans who work alongside these systems (for more on the role of experience, see the author’s previous HBR article with Mark Williams on “How AI Can Help Leaders Make Better Decisions Under Pressure”).
. . .
From the early days of mechanical automatons to more recent conversational bots, scientists and engineers have dreamed of a future where AI systems can work and act intelligently and independently. Recent advances in agentic AI bring that autonomous future a step closer to reality. The agentic AI prize could be great, with the promise of greater productivity, innovation, and insights for the human workforce. But so, too, are the risks: the potential for bias, mistakes, and inappropriate use. Early action by business and government leaders now will help set the right course for agentic AI development, so that its benefits can be achieved safely and fairly.
Mark Purdy is a co-founder and director of Beacon Thought Leadership, an independent advisory firm focused on content development and training services.
Source:
https://hbr.org/2024/12/what-is-agentic-ai-and-how-will-it-change-work
.jpeg)
Comentários
Enviar um comentário