McKinsey says Gen AI is stuck: lots of hype, little bottom-line impact. But banks, research firms, and others are quietly breaking through with AI agents and something called the "agentic mesh." Is this AI’s big pivot moment?
Generative AI is everywhere, except where it counts.
In nearly every boardroom today, executives tout their company’s adoption of gen AI. Investments flow, enthusiasm builds, yet something critical remains absent. According to McKinsey’s June 2025 report Seizing the Agentic AI Advantage what’s missing is measurable impact.
"Nearly eight in ten companies report using gen AI,” they say. Yet just as many report no significant bottom-line impact."
The widespread deployment has curiously failed to move financial needles, leaving companies in an awkward position. AI is seen everywhere, but felt nowhere.
At the heart of this paradox, McKinsey argues, lies a misalignment between how AI has been adopted and how it generates value.
Horizontal use cases, notably tools like Microsoft’s Copilot or Google's Workspace AI, proliferate rapidly because they're easy to plug in and intuitive to use. They provide general assistance, they summarize emails, draft notes, simplify meetings, and so on. Yet these horizontal applications scatter their value thinly, spreading incremental productivity improvements so broadly that the total impact fades into insignificance.
As the McKinsey report puts it, these applications deliver "diffuse, hard-to-measure gains."
In sharp contrast, vertical applications (those baked into core business functions) carry the promise of significant value but struggle profoundly to scale. Less than 10 percent of these targeted deployments ever graduate beyond pilot phases, trapped behind technological complexity, organizational inertia, and a lack of mature solutions.
"Vertical use cases seldom make it out of the pilot phase because of technical, organizational, data, and cultural barriers."
This is exactly where AI agents emerge as a solution.
Unlike traditional generative AI, which remains largely reactive (waiting passively for human prompting) agents introduce something entirely different. "AI agents mark a major evolution in enterprise AI—extending gen AI from reactive content generation to autonomous, goal-driven execution,” McKinsey researchers explain.
Agents have autonomy. They plan, retain memory across interactions, orchestrate complex workflows, and adapt to shifting scenarios in real time. They operate less as passive assistants and more as proactive digital collaborators.
But harnessing the potential of agents demands a new type of infrastructure, an entirely new architecture McKinsey terms the "Agentic AI Mesh."
The agentic mesh isn't simply a set of new tools bolted onto existing systems. Instead, it represents a shift in how organizations conceive, deploy, and manage their AI capabilities.
They define the agentic mesh as "a composable, distributed, and vendor-agnostic architectural paradigm that enables multiple agents to reason, collaborate, and act autonomously across a wide array of systems, tools, and language models—securely, at scale, and built to evolve with the technology."
The agentic mesh’s design follows five interconnected principles.
The first is composability, allowing any agent, tool, or LLM to plug seamlessly into the mesh. This modular flexibility eliminates the constant need for disruptive overhauls when integrating new tools.
The second principle is distributed intelligence, where tasks are broken down and collaboratively resolved by networks of agents. This distribution removes bottlenecks that arise when centralizing complex decisions or when empowering AI systems to self-organize around dynamic scenarios.
Third, the mesh employs layered decoupling, meaning logic, memory, orchestration, and interface functions operate independently.
As McKinsey describes, this modular decoupling allows continuous improvement without needing full system redesigns.
Fourth, vendor neutrality prevents lock-ins, as each component can be independently updated or replaced without extensive restructuring.
And finally, governed autonomy ensures that agent behavior remains proactively controlled via embedded policies, permissions, and transparent escalation pathways.
But why such a radical architectural departure from existing frameworks?
Because traditional AI stacks, typically static and LLM-centric, can't adequately manage the unique systemic risks introduced by autonomous agents.
"Agents introduce a new class of systemic risks that traditional gen AI architectures, designed primarily for isolated LLM-centric use cases, were never built to handle: uncontrolled autonomy, fragmented system access, lack of observability and traceability, expanding surface of attack, and agent sprawl and duplication."
These challenges demand a comprehensive governance and management strategy woven directly into the architecture’s fabric.
To address these risks and complexities, the agentic mesh embeds seven core capabilities. Agent and workflow discovery ensures organizations maintain a dynamic catalog of agents, enabling visibility and reuse. An AI asset registry centralizes governance of agent configurations, definitions, instructions, and so on, all of which are crucial for oversight and control.
Observability, they add, offers end-to-end tracing of workflows, vital for transparency and troubleshooting.
Authentication and authorization enforce precise access controls, securing communication channels between agents and external systems.
Further, comprehensive evaluations rigorously test agents, maintaining accuracy and compliance over time. Feedback management incorporates automated improvement loops to evolve agent behavior continuously.
And last but not least, compliance and risk management embed policy controls and ethical guardrails, aligning agent activities with regulatory and organizational standards.
Together, these capabilities form an integrated mesh, robust enough to manage complexity and agile enough to adapt to future advancements.
The broader strategic implications should cause enterprises to rethink their IT architectures entirely and McKinsey notes this shift already underway at major software firms.
"Microsoft is embedding agents into the core of Dynamics 365 and Microsoft 365 via Copilot Studio; Salesforce is expanding Agentforce into a multiagent orchestration layer; SAP is rearchitecting its Business Technology Platform (BTP) to support agent integration through Joule."
These moves signal more than mere technological progression, they foreshadow a rethink of enterprise computing itself, from passive interfaces designed around human navigation to agent-native infrastructures optimized for autonomous digital collaboration.
Yet despite the technological leap, McKinsey emphasizes the most substantial challenge won’t be technical, it will be human. Organizations will have to navigate new complexities around human-agent coexistence, managing autonomy and trust, and preventing agent sprawl.
And for that matter, human-agent cohabitation raises nuanced questions about role boundaries like, hey, when should agents act independently, and when should they defer to human oversight?
McKinsey points out that keeping agents in check isn’t something you set and forget. You’ve got to watch constantly to make sure they’re not overstepping (or worse, sitting idle when something important needs human eyes). Just as crucial is avoiding sprawl, where agents multiply out of control, creating chaos instead of clarity. The goal is an organized, scalable agent ecosystem that stays manageable as it grows.
"The real challenge lies in coordination, judgment, and trust."
Trust, particularly, doesn't come merely from technical performance but emerges from predictable behavior, transparent communication. Just as it does with human beings.
To navigate this transition effectively, CEOs must lead a strategic pivot. McKinsey argues.
"This pivot cannot be delegated—it must be initiated and led by the CEO."
To really make this work, McKinsey says it’s time to wrap up all those scattered AI experiments and get serious about what matters most. That means clear priorities, solid guardrails, and picking high-impact "lighthouse" projects that show how it's done. The agentic mesh isn't just a fancy architecture—it’s a call for leaders to rethink how the whole enterprise runs.
McKinsey highlights several examples where organizations are already making meaningful leaps with AI agents. Take banking, for instance. One bank tackled a massive legacy system overhaul, originally a sluggish and manual process. They deployed agent-driven squads that automated coding, documentation, and testing, cutting the project workload by more than 50%.
Then there’s the retail banking credit-risk process. Previously, relationship managers spent days drafting complex memos manually. Now agents independently handle data extraction, preliminary analysis, and memo drafts, significantly speeding up credit decisions and boosting productivity by as much as 60%.
Another case comes from market research, where agents autonomously detect data anomalies, analyze market shifts, and explain those shifts clearly for human analysts. By handling tedious, error-prone tasks, these agents have the potential to save millions annually and significantly increase analyst productivity.
These aren’t edge cases, they’re clear signs of what happens when organizations embrace agents not just as incremental tools, but as fundamental parts of their business processes.
In the end, the agentic mesh isn’t just another architecture. McKinsey sees it as the bedrock for how enterprises will operate next, not a tweak but a complete shift in how businesses run at their core.
"Agentic AI is not an incremental step—it is the foundation of the next-generation operating model."
Organizations embracing this architecture won't merely capture short-term efficiencies. Instead, they'll redefine how they operate, compete, and create value for years to come.
In classic McKinsey fashion, the message comes through loud and clear: exploring is done, and transforming needs to start now.
If you're still stuck at the edge of the AI paradox…lots of deployments but little payoff…the agentic mesh isn't just an option. It's a strategic necessity. The only real question left is how quickly and boldly you'll step forward.