Is Agentic Automation Nearer Than We Realize?

Analysts rely on their instincts. When PAC attended an AI analyst event hosted by McKinsey, we were unsure what to expect. We knew of their compelling thought leadership on the topic, but at the back of our minds were the many insinuations across the industry regarding templated strategies and alpha male attitude. Meeting McKinsey’s QuantumBlack team, we were quickly reminded to reflect on our misconceptions. The executives we met were eager to engage and reminded us more of a startup than a leading business consultancy. When QuantumBlack got acquired in 2015, it had 50 team members. Fast forward to 2025, and there are now 1500 employees serving as the firm’s center of excellence for AI and advanced analytics. Additionally, QuantumBlack Labs, the R&D and software development hub within QuantumBlack, promotes innovation by developing tools and assets that support AI transformations. Not only did we know little about McKinsey’s acquisitions, such as iguazio and IncepTech, that beefed up their technology capabilities, but today, 45% of their work is building digital and technology solutions.

From prompt to agent factory

The presentation that ignited our imagination was a project with a bank, in which QuantumBlack leveraged an agent factory to migrate their client away from a mainframe finally. It stood out because it leveraged a factory approach to building agents to enable the transformation of one of the most persistent technology debt challenges. Looking at it through a different lens, the next frontier of innovation is helping to solve a problem, which is synonymous with a reality check of how far innovation can go. The more we should try to learn from previous secular technology shifts when assessing the progress with Agentic AI.

Adoption of Agentic AI depends on the understanding of agency and autonomy

When analyzing the adoption of Agentic AI, cutting through the market noise is challenging. One has not only to debunk the many marketing claims, but also the luminaries on AI have vastly different opinions. Take Nvidia’s CEO, Jensen Huang, who proclaims that the IT department of every company will be the HR department of AI agents in the future. Yet, Meta’s AI Chief Scientist Yann LeCun cautioned that LLMs epitomize conformism. They regurgitate what is already known in a digestible format. Therefore, he reflected, “That is very useful, but a far cry from a country of geniuses in data centers.”

This boils down to two key issues. First, a more nuanced understanding of agency and autonomy of agents. In our view, at this point, the autonomy is much more restricted than many pundits contend. For details, see our blog on The Hitchhiker’s Guide for Driving Value Realization with Agentic AI. Second, we need to understand the learnings and outcomes from agent orchestration much better. Just pointing to frameworks such as MCP or A2A is completely missing the point of having to design an outcome-centric governance approach that is robust enough to assure the business outcomes of agent orchestration.

From body shops to autonomous systems

For PAC, QuantumBlack’s example of agent orchestration also looped back to the fundamentals of automation and service delivery. When the discussions about RPA started in 2012, we also discussed Autonomic Computing from the likes of IPsoft and Arago. The narratives were about virtual engineers and self-healing systems. Following that, TCS pushed its Machine First framework, where algorithms had the first right of refusal. Simply put, if a task can be automated, a machine gets the first shot at it. As with Agentic AI, the key is understanding the scope of automation and autonomy. While the broader concept of Intelligent Automation did settle on task automation and employee productivity, the ultimate promise of Agentic AI is autonomous services where the primary interaction is through technology, but not any longer with humans. The change that Agentic AI could bring is contextual understanding and automating end-to-end processes. This is why the QuantumBlack example jumped out to us.

This is the context that the infographic in Figure 1 is trying to convey. Delivery models evolved from a focus on labor arbitrage to automation and platform-led services. RPA and platforms such as Salesforce and ServiceNow were key enablers for service delivery for what is now Agentic AI. With the latter, the North Star is the promise of Autonomous Services. In this new paradigm, processes are distributed and self-organized. The culture is predominantly experimental, while the commercial models are shifting to subscription and license-based models. Providers like IBM are discussing consulting-as-code, where even complex transformations will be codified. Accenture, in its 2025 Technology Vision, proclaims that by 2030, agents, not people, will be the primary users of most enterprises’ internal digital systems.

Figure 1: Delivery models are expanding toward autonomous services

Fundamentally, organizations must design a control environment that manages Digital Labor. That means a delivery strategy where humans, bots, and agents coexist and collaborate. What was largely lip service in the early RPA days is now starting to become a reality, but it is also a challenge.

The road ahead is about orchestration and governance

Organizations must define their ambitions in this frenetically evolving technology landscape. Yet, most can’t escape technology, process, and cultural debt. Therefore, the key to progressing with Agentic AI is a holistic approach to orchestration spanning data assets and multiple third-party agents. An outcome-centric approach to governance must underpin this orchestration. Like most secular technology shifts, it is a new complexity that only a few organizations can manage today. We need to learn more from the early deployments, like this one from QuantumBlack.