Eight in ten companies say they’ve adopted generative AI. Eight in ten also admit it hasn’t improved their bottom line.
Those figures aren’t just opinions. They come from a June 2025 McKinsey study that calls this the “gen AI disconnect”: companies are spending billions on AI, but struggling to turn it into measurable business value.
The issue isn’t enthusiasm. The problem lies in how generative AI is being deployed. Businesses are scaling the easy, low-impact use cases, while the harder, industry-changing projects stall.
Where Does AI Actually Move the Needle?
McKinsey draws a line between “horizontal” and “vertical” AI applications.
- Horizontal AI (like copilots in Office apps, customer chatbots, or brainstorming assistants) is quick to roll out across thousands of employees. The productivity benefits are scattered and difficult to measure.
- Vertical AI is where the real transformation happens. For a law firm, that could mean automated contract review. For logistics, dynamic fleet optimisation. For retail, demand forecasting. These use cases are harder, slower and often industry-specific – but the payoffs are tangible.
Most firms have chased the easy wins. The result: widespread adoption, but limited business impact.
The Rise of Agentic AI – and Why It’s Not Plug-and-Play
McKinsey highlights the next wave: agentic AI. Unlike copilots, agents don’t just suggest – they take action. They can update records, trigger workflows, even coordinate with other agents.
But here’s the catch: agents can’t simply be dropped into existing processes. To unlock real value, businesses must rethink workflows and operating models. McKinsey is blunt: many “AI pilots” so far have been surface-level experiments that don’t change how work actually gets done.
Avoiding “Agent Sprawl”: Why Orchestration Matters
Autonomous agents bring risks too. McKinsey warns of “agent sprawl” and “autonomy drift,” where systems multiply and act unpredictably – echoing the problems of the Robotic Process Automation boom, only faster.
The answer? An agentic AI mesh: a coordinated architecture to orchestrate agents, enforce governance, and keep them aligned with business goals.
Why Trust Governance Are the Real Bottlenecks
Perhaps the biggest takeaway is that technology isn’t the barrier – trust is.
Agents are powerful because they act independently. But that independence also makes them harder to trust. To succeed, executives need to invest not just in the tech, but in governance, risk management, and cross-departmental coordination.
McKinsey calls this building “new operating models and new trust frameworks.” Training, oversight, and clear accountability will decide whether AI agents become business assets or liabilities.
What Business Leaders Should Do Next
Based on McKinsey’s research, here are five practical lessons:
- Don’t mistake adoption for impact. Just rolling out copilots doesn’t guarantee financial returns.
- Prioritise vertical use cases. The harder, industry-specific projects are where AI proves its worth.
- Treat agents as transformation projects. They require rethinking workflows and business models.
- Plan for orchestration. Without an “agentic mesh,” AI adoption risks sliding into chaos.
- Put governance front and centre. Trust frameworks and accountability are as important as the tech itself.
Final Thought
Generative AI is everywhere, but returns remain elusive. The issue isn’t potential – it’s integration. To move from hype to measurable outcomes, businesses must go beyond shiny copilots and redesign workflows, operating models, and governance structures.
The next phase of AI won’t be measured by how many agents a company deploys, but by whether those agents are embedded in ways that truly shift performance and profitability.
Read the full McKinsey report here: Seizing the Agentic AI Advantage
