KEY TAKEAWAYS:

• GenAI has improved enterprise automation, but it remains assistive, not autonomous. Chatbots and co-pilots can increase productivity in specific tasks, but they require human oversight and still struggle with inaccuracies, limiting impact.
• Agentic AI steps beyond GenAI by fully executing workflows. Unlike GenAI, agents are goal-driven and capable of adapting, learning, iterating, collaborating with other systems and humans, and completing tasks end-to-end.
• The future of enterprise AI is a collaborative Agentic AI systems, promising a more transformative leap than Gen AI, which has added incremental value so far. As AI agency evolves, Agentic AI will rehaul current workflows and redefine how humans and technology work together.

On October 7, 1913, at the Highland Park factory in Michigan, USA, Henry Ford introduced one of the world’s first moving assembly lines to produce the Ford Model T. Instead of workers putting together entire cars by hand, the process had now been broken down into 45 distinct tasks. Workers stood in line as a moving conveyor belt brought car parts to them, and this dramatically sped up production as everyone performed their specialized tasks over and over.

What was taking more than 12 hours could now be done in 93 minutes. The production cost became low enough that the car could be sold to America’s middle-class, not just its elites. The Ford assembly line revolutionized not just the automobile industry, but also how work would be done by humans and technology forever.

In every industry from manufacturing to services, work is now split into specialized, repeatable functions—such as production, finance, marketing, and sales. Only now, instead of a conveyor belt, the technology that assists humans has evolved. Employees use various SaaS tools to execute their tasks, but human coordination remains necessary. Tasks have manual and digital components, with humans still at the center, coordinating across workflows, planning ahead, making decisions.

This system has its inefficiencies, despite rapid software advancements. Workers spend hours switching between systems, searching for the right data, and navigating multiple data silos. My decade in the world of Robotic Process Automation (RPA) revealed how automation could make a significant difference to productivity—but the scripted and brittle nature of RPA also became a constraint. Our hope was that more cognitive technologies would provide the gains of automation while also being fluid and intelligent.

That said, even with GenAI, while productivity seems to have improved within specific workflows, enterprises still resemble the early assembly line in their fragmented, vertically segregated operations. Human beings are managing layers of ‘digital conveyor belts’ across them.

It is the next evolution of AI—Agentic AI—that can truly step-change the technological component of modern work. Autonomous, collaborative AI agents will not just assist human employees, but actually complete workflows across the enterprise, from decision-making through execution. And this will be as transformative for white-collar work as the assembly line was for industrial production, reshaping how businesses operate and work gets done.

GenAI: Two Years On, Its Impact and Limitations

ChatGPT and the slew of GenAI products that followed it through late 2022 grabbed our collective attention and imagination. Millions of people were suddenly using an interactive technology that could understand and speak just like them, producing outputs that were thought to be intrinsically human, such as code, writing, images, videos, even diagnostic treatments.

This watershed moment in the history of technology led to adoption and innovation in both consumer and enterprise sectors. In a 2024 McKinsey survey on the state of AI, two-thirds of over 1,300 respondents reported that their organizations were regularly using GenAI, nearly double the percentage from just ten months before. With GenAI estimated to add $4.4 trillion to the global economy, corporate leaders are eager to capture this value.

But two years on from ChatGPT’s “wow moment”, rising adoption of GenAI has also led to the understanding that it can bring along with significant risks. Organizations are still early in their GenAI journeys, but they have first-hand learnings on what works and what doesn’t now, separate from the general hype around the technology.

There have been material benefits from deploying GenAI within customer support, legal, and sales and marketing use-cases, via both cost reduction and increased revenue, but research also highlights the many challenges that have come along with it:

• Inaccuracies and Hallucinations: A common finding across studies by McKinsey, Deloitte, Asana, and Anthropic is that inaccurate outputs from AI are a major concern for enterprises. There is a need for high-quality training data and for customized models, leading to an increased emphasis on data governance as well as solutions that can be tailored to the enterprise. Hallucinations are occasionally tolerable within internal use-cases, but with customer-facing applications, they stand to risk reputation, trust, and revenue.
• Scalability and Deployment Challenges: A majority of respondents across studies were unable to get most of their GenAI experiments into production, indicating that deployment and scalability issues plague these projects. Most planned AI investments are stalled by delays, as integrating AI within existing systems, without a disruption to current operations, is a challenge.
• Measurement and AI Literacy Gap: A 2024 Deloitte survey of more than 2770 enterprise leaders found that 40% of them were struggling to define and measure the impact of their GenAI initiatives. Although most were seeking productivity gains, less than half were actually tracking changes along relevant metrics. Most companies do not have a formal AI strategy in place, leading to misalignment between business goals and AI projects, which in turn can create sub-optimal RoI.

In addition to these issues of inaccuracy and hallucination, scalability and deployment challenges, and vague RoI, businesses are dealing with concerns about data privacy, cybersecurity, explainability, and the impact of AI on organization structure and culture. Given how much there is to navigate with a technology evolving as rapidly as AI, we believe the issues with GenAI will not stop at the above.

The Real Gap: GenAI Doesn’t Do Work. Agentic AI does.

If the modern enterprise needs to undergo a major overhaul in order to incorporate a new but promising technology, it might as well evolve in the best possible way, so as to result in tangible RoI, aligned with strategic business goals.

And that’s where GenAI stops short, and Agentic AI goes long. While co-pilots and chatbots paved the way in generating text, images, and code, their role remains largely assistive to humans. They perform tasks at the “prompt” of a human, and their outputs are within the realm of existing vertical functions, be it customer support or sales, relying on human oversight for improvement, coordination, and completion.

The real transformative leap for businesses comes with Agentic AI, which is autonomous and can own and execute workflows end-to-end. AI agents are goal-driven software entities that can reflect and iterate on their own outputs, use external tooling to generate better results, identify goals and plan ahead, and also collaborate with other agents to execute complex workflows from start to finish. A caveat is that most early enterprise AI experiments are built off of limited, off-the-shelf GenAI solutions, leading to hallucinations and data privacy concerns. But mature Agentic AI providers such as Ema custom-train models on enterprise data, performing queries with the highest accuracy and at the lowest costs, while tackling security and data privacy via top-tier encryption and data governance policies.

According to a recent Garter webinar, Agentic AI has been gaining momentum through 2024, and is capable of delivering transformational value to businesses. Unlike GenAI, which has been delivering incremental value so far, Agentic AI systems promise to automate the work of entire roles, be it business analysts or quality assurance specialists or marketing. Indeed, the major leap between GenAI and Agentic AI is the latter’s ability to tackle complexity and multiplicity in the nature of tasks, learn from its own outputs, connect to other tools, plan ahead, and collaborate with other agents and also humans.

By 2028, Gartner estimates that 33% of enterprise software will include Agentic AI, up from less than 1% today. These agentic systems will autonomously handle 15% of day-to-day work and decision-making, offering a real boost in efficiency for businesses.

But the promise of Agentic AI doesn’t stop with individual AI agents. Systems of AI agents will function best together as an intelligent, adaptive ecosystem, breaking down larger business goals into smaller problems and sub-tasks, executing and optimizing workflows end-to-end.

Just as building a birdhouse requires the intelligent application of multiple tools—not just a hammer, but also nails, wood, and glue applied towards a visual plan of the birdhouse—effective automation in the enterprise demands multiple agents specializing in different parts of large and complicated workflows, aiding and collaborating with each other towards bigger goals.

To unlock this transformative potential, businesses must first understand their internal processes and data. AI agents are waiting to be trained and deployed, but their success depends on the clarity of work assigned to them, and the quality of the data they’re trained on. Enterprises need to define their business goals, codify their current workflows, and create data readiness and systems where collections of agents can work and drive business value together.

The Future of Agentic AI in the Enterprise

Given the pressures to stay AI-forward, leaders must avoid the trap of adopting tech for tech’s sake. The right strategy is to start by identifying pressing business goals and challenges, and working backwards from there to see how tech, and AI, can help.

A “tiger team” of AI agents can be applied towards meaningful, measurable business strategies and outcomes. But this does not necessarily mean jettisoning the technologies in use before. For example, Intelligent Document Processing (IDP) may still read data from documents, Robotic Process Automation (RPA) may access data and trigger next steps, while an Agentic AI layer on top can use these outputs for another workflow, while also evaluating the quality of their outputs, and ensuring that they are in line with larger strategic goals and decisions. A “human-in-the-loop” control mechanism should be deployed, where human specialists validate outputs for accuracy, brand goals, and compliance, creating a flywheel for the ongoing improvement of Agentic AI systems.

Furthermore, there are benefits to thinking of agents, or a mesh of agents, as adding more workers to the organization. These agents will adapt, learn, and scale, meaning business processes will have to be rearranged as AI takes on tasks that previously required more manual coordination across departments. The shift from human-led decision-making to AI-driven execution will streamline operations and change entire job functions. But they will also create new efficiencies and possibilities.

The technology is still evolving, and AI agency lies on a spectrum. On the one end are LLM- or GenAI-based assistants that perform specific tasks under defined conditions. At the other end are Agentic AI systems that learn from their environment, can make decisions, collaborate with others, and perform tasks independently.

As we move towards the latter end of the spectrum, enterprises will have to mitigate new challenges, just as adding new workers and restructuring the organization would involve. But one thing is for sure: Just as Ford atomized work down to speed up production, Agentic AI will unleash new innovations at both the task-level and in how work gets allocated. Technology will no longer simply assist humans, as it has for more than a century. Instead, automation and humans will together form a large, self-sustaining workforce — one that makes optimal decisions, drives productivity, and aligns with meaningful, strategic goals.

Ian Barkin is a seasoned innovator in digital operations and intelligent automation, and serves as an advisor to several companies in the field, including Ema.