The rate at which agentic AI is entering the contemporary workplace is, bluntly, staggering it is, in fact, unprecedented in modern corporate history, remarks Simon Cockshell, partner for The Future of Management and Organization Practice at McKinsey. What started as a tentative foray into using generative AI has morphed, quite literally, into an operational transformation, leaving even the most seasoned corporate leaders stunned. In fact, even AI-initiated restructuring at McKinsey itself is squarely within a sector that is due for a shake-up.
Agentic AI represents a level of advancement upon traditional robotic process automation (RPA). While traditional RPA has represented a highly rule-bound task, agentic AI integrates large language models capable of interpreting ambiguous data, making dynamic decisions, and engaging with autonomous multi-step processes. In the consultancy domain, where managing client fulfillment requires an immediate synthesis of cross-section data, agentic AI can now undertake the heavyweight analysis task associated with preparing reports, scenario analysis, and even making strategic decisions at a margin cost tending to zero. As UiPath President Daniel Dines has declared, With AI agents, RPA, and human guidance, the entire process can now be automated by AI.
Acceleration in this progress is achieved with advancements in chain of thought (CoT) reasoning in AI systems. Historically, generative AIs “hallucinated” facts. However, CoT networks minimize this with AIs auditing themselves in chain of thought reasoning with steep declines in logic and math errors. For example, in programming code, Granite Instruct models of AI in programming are fine-tuned with exemplars of chain of thought with multimodal CoT AIs handling numeric, image, and text data in a holistic analysis. This has resulted in “vibe coding”, which uses natural language programming, where anyone other than software professionals can create and implement functional tools for consultants and analysts. Replit Agent platforms allow sub-agents to be controlled automatically and separately in distinct areas of a project of interest.
The fallout from all of this remains startling. Goldman Sachs has predicted the potential elimination of 6-7% of jobs in the US, and the hiring of people considered to be “AI-exposed” has dropped by 13% since the widespread use of LLMs. According to models used internally by McKinsey, there exists the potential to displace or elimenate half of all white-collar jobs for entry-level workers in the fields of technology, finance, legal, and consulting. Indeed, all of this makes sense considering the reports of Y Combinator startups, whose start-up code has come to include the programming ability of AI, which was the end result of teams of humans five short years ago.
Speaking about the engineering side, the infrastructure for the deployment of the technology is actually helping its adoption. The creation of low-code and no-code technologies will allow different people from an organization to implement AI agents. The implementation of code by companies like Anthropic, Cursor, and Replit will allow developers to implement code from natural language descriptions to the development stage in a matter of a few hours. This development of code will also affect the structure of the organization.
The consulting approach and, more specifically, that approach that has traditionally relied on a staff of young analysts is even more vulnerable.
Agentic AI has the ability to automate much of the efforts involved in data analysis tasks and even building the slide deck that is the meat and potatoes of the work involved in the entry-level analyst position. As pointed out by Anthropics Dario Amodei, Its going to happen in a small amount of time as little as a couple of years or less. However, the diffusion of this technology at a very rapid rate also has strategic implications. A conceptual form of McKinseys agentic organization, in line with the above scenario, argues that the combination of human team members and other agents acting in a highly integrated manner could potentially substitute the traditional functional structure with self-contained units focused on the result.
One could envision, as an example, a consulting project in which overnight market analyses are carried out by agents such as market agents, while compliance agents verify regulatory compliance and creative agents compose winning proposals, which would require human consultants to work on storytelling and stakeholder engagement. The problem in this area is clearly a double-edged sword for the executives: the problem of how they should respond in the short term to the problem of the sector they are in as a result of the labor market shock as well as how they are going to prepare their own company in the productivity increase.
It is an area in which the response of the policymaker in this scenario is likely to lag behind. It has been acknowledged in the area of business policy changes: Policymakers will have a very limited ability to do anything here unless its through subsidies or tax policy, according to a quote from a research paper at Harvard Universitys Christopher Stanton. It is up to the companies in these areas to upgrade their employees and the nature of their own business as a whole through the inclusion of AI literacy.
The warning signs are there. The chain-of-thought model of artificial intelligence has led the margins of errors down. The agentic architectures of the technology are setting the parameters of the use of the technology. Ease of deployment is clearing the technology barriers. The scenario in the area of consulting as well as in the various white-collar businesses is less a future threat than a reality of the present moment which requires an instantaneous response.
