Everything About AI
NextFin News - The question facing corporate leaders is no longer whether artificial intelligence will change work, but whether companies can redesign accountability before the technology changes the risk profile. In an interview published Nov. 12, McKinsey Global Managing Partner Bob Sternfels connected crisis readiness with succession planning, independent governance and faster decision-making, while discussing how AI is reshaping McKinsey’s workforce and what CEOs get wrong when they adopt the technology. The immediate AI cycle may cool as pilots disappoint, but the governance problem is structural.
The interview’s themes join two subjects companies often manage separately. Crisis planning is treated as an emergency exercise; AI is treated as a technology investment. In practice, both determine who makes decisions when information is incomplete, how quickly a company can mobilize, and whether a board can distinguish a temporary shock from a failure in the operating model.
McKinsey’s 2025 global AI survey, summarized in the firm’s published research, found that nearly 90% of respondents said their organizations regularly used AI and almost 70% reported regular use of generative AI. Yet most organizations remained in experimentation or piloting. That gap matters more than the adoption headline: deployment is becoming ordinary before the controls, skills and accountability needed for deployment become ordinary.
The same tension appears in McKinsey’s resilience research. The firm reported that 84% of leaders feel underprepared for future disruptions, while 60% of board members say their companies are not ready for the next major event. Those figures describe a corporate system that recognizes uncertainty but has not consistently converted that recognition into decision rights, tested succession plans or reliable information flows.
The result is a two-speed organization. AI accelerates the production and movement of information, while governance often still moves through periodic committees, unclear ownership and fragmented data. A company can therefore respond faster to a crisis in theory and become more confused in practice.
The Real Crisis Is an Accountability Gap
The first-order effect of AI is familiar: employees receive tools that can summarize, generate, classify and recommend at lower cost and higher speed. The deeper effect is that decisions once separated by function begin to merge. A model used in customer service can alter compliance exposure; an automated procurement recommendation can affect suppliers and geopolitical concentration; an agent connected to internal systems can turn a local error into an operational event.
That is why succession planning and independent governance belong in the same discussion as AI. A crisis exposes the chain of command. If a chief executive is unavailable, a product fails, or a model produces an unsafe output, the organization needs to know who can act, who must be consulted and who owns the consequences. The technology does not remove that need. It makes the answer more urgent because the speed of propagation rises.
McKinsey’s risk and resilience research recommends a cross-functional responsible-AI steering group with at least a monthly cadence. The proposed group includes business and technology leaders alongside data, privacy, legal and compliance functions. Its purpose is not to create another approval layer for every experiment. It is to decide which risks are material, which controls are mandatory and how the company will respond when an AI system behaves outside its intended use.
The distinction is important. A board that asks only how quickly AI can lower costs is measuring the first-order benefit. A board that asks how an AI-enabled process changes liability, concentration, cyber exposure and decision rights is measuring the transmission mechanism.
In a conventional crisis, information is scarce and time is expensive. In an AI-mediated crisis, information may be abundant but ownership is scarce. Dashboards can show anomalies without showing who has authority to stop a process. Models can produce plausible explanations without proving that the underlying data is sound. Automated systems can make a decision repeatable without making it defensible.
That is the structural shift. The crisis playbook is no longer just a document for extraordinary events. It is becoming an operating architecture for systems that make or recommend decisions continuously.
Why the Pilot Economy Has Not Delivered Resilience
The second question is why widespread AI use has not produced equally widespread transformation. The answer is that adoption and institutional change are different processes. A company can add a generative-AI tool to a workflow without changing the workflow’s owner, controls or performance measure.
McKinsey’s survey evidence captures that contradiction: nearly nine in ten respondents reported regular AI use, but most organizations remained in pilots or experiments. The adoption number measures access and activity. It does not establish that AI has changed revenue, margins, service reliability or crisis response.
The pilot model is attractive because it limits immediate disruption. A team can test an assistant without changing the organization chart. But that containment also prevents the company from learning how the tool behaves across functions. A pilot can show that a model writes a useful memo; it cannot by itself show whether the memo changes a regulated decision, duplicates another process or creates a new failure point.
This is where corporate crisis management becomes a practical test. A resilient organization does not define readiness by the number of scenarios in a binder. It defines readiness by whether people can make fast, coordinated decisions under stress. AI pilots that sit outside the core operating model do not test that capability. They test software utility.
McKinsey’s organizations research points to the same bottleneck from another direction. Weak technology and data infrastructure can limit risk forecasting, real-time regional insight and coordinated cross-border decision-making. The problem is not merely that the company lacks a better model. It may lack a common data language, a clear owner and a process for turning a signal into action.
Governance simplification can therefore be more valuable than governance expansion. McKinsey says simplifying governance can free between 20% and 60% of management time for strategic work, based on its experience. The implication is not that companies need fewer controls in every area. It is that they need fewer overlapping decisions and clearer accountability. A committee that adds delay without adding judgment is a crisis risk, whether or not AI is involved.
A global fast-moving consumer-goods company cited in the same research eliminated 70% of duplicated reports and decisions by clarifying accountabilities, cutting data points by 30% and introducing automated dashboards. The example reverses the usual AI narrative. Resilience did not begin with a more sophisticated algorithm. It began with removing duplication and deciding who was responsible.
The immediate AI cycle is cyclical. Companies can abandon weak pilots, consolidate vendors and reduce spending on use cases that do not produce measurable value. But the requirement to redesign decision rights is structural. Even a disappointing model leaves behind new questions about data ownership, human review, model risk and workforce capability.
The Second-Order Effect Runs Through Trade and Workforce Design
The obvious conclusion is that AI changes jobs and increases productivity. The less obvious conclusion is that it changes the geography and vulnerability of corporate operations. Once software agents coordinate tasks across borders, a company becomes more dependent on data infrastructure, model access, cybersecurity and the quality of local information.
That creates a second-order chain. AI adoption can reduce the time required for routine analysis; faster analysis can centralize decisions; centralization can increase the impact of a data outage, cyberattack or policy shock; the resulting concentration can make a company less resilient even as its day-to-day operations become more efficient.
Global trade adds another layer. A company exposed to tariffs, sanctions, logistics disruption or supplier concentration needs local intelligence and cross-border coordination. AI can improve scenario analysis, but only if the data captures regional conditions and the governance model permits rapid action. Otherwise, the company receives a faster version of the wrong answer.
Workforce design follows the same pattern. AI can shift junior employees away from research and drafting toward review, judgment and client or operational ownership. That may raise productivity in the short run. It can also weaken the traditional apprenticeship path if firms automate the tasks through which people learned how to spot errors.
This is why the discussion of AI reshaping McKinsey’s workforce matters beyond consulting. The key resource is not simply access to a model. It is the ability to combine machine output with institutional memory and accountable judgment. If companies remove too many entry-level tasks without replacing the learning system, they may reduce today’s labor cost while creating tomorrow’s shortage of experienced decision-makers.
That shortage would surface during a crisis. A model may identify a pattern, but an experienced operator knows which exception matters, which stakeholder must be called and which apparent anomaly is a data artifact. The value of human expertise rises precisely where automated systems are least reliable: at the boundary between the known process and the unprecedented event.
For companies and their investors, the second-order implication is a widening gap between visible AI spending and less visible resilience spending. Data centers, software contracts and automation projects are easy to count. The costs of governance, training, succession and scenario testing are harder to isolate, even though they determine whether productivity survives a shock.
That gap can distort corporate comparisons. A company with a smaller AI budget but cleaner data, simpler accountability and stronger operational judgment may withstand disruption better than a company with more agents and weaker controls. The competitive advantage will not be measured only by model quality. It will be measured by the speed at which the organization can detect, decide and learn.
The Counter-Thesis: More AI Can Make Crisis Response Worse
The strongest case against this structural reading is that AI may add complexity faster than it adds resilience. New models, vendors and automated agents can expand the attack surface, create opaque dependencies and blur responsibility. A company that installs a steering group and a policy framework may still be slower than a simpler competitor when a crisis arrives.
That counter-thesis deserves more than a compliance response. Agentic AI expands the risk surface because systems can act across workflows rather than merely generate content. A hallucinated answer in a draft is a quality problem; an erroneous agent action in procurement, finance or customer operations can become a control event. The more autonomous the system, the less credible it is to treat human review as a generic safety slogan.
The counterargument also challenges the idea that independent governance automatically improves outcomes. Independent directors can ask harder questions, but they cannot compensate for weak data, unclear operating ownership or a board that receives only curated metrics. Oversight works when information arrives early enough and when someone has authority to act on it.
The answer is not to declare every AI initiative unsafe. It is to separate reversible experimentation from systems that affect customers, capital, employees, safety or regulatory obligations. Low-risk tools can move quickly. High-consequence systems need named owners, documented controls, monitoring, audit trails and a tested shutdown process.
The falsifying test for this structural thesis is observable: if, across the next two reporting cycles, companies that scale AI beyond pilots consistently show no increase in control incidents, cyber events, operational disruptions or management delay relative to less automated peers, then the claim that AI requires a new governance architecture would be overstated. A measurable rise in incidents after deployment without corresponding control investment would instead indicate that the risk surface is expanding faster than resilience.
The point is not that governance should defeat speed. It is that governance must be designed as part of speed. A company that waits for a crisis to decide who owns an AI-enabled process has already surrendered the advantage the technology is meant to create.
What Changes Across the Three Time Horizons
In the short term, the AI story remains cyclical. Executive enthusiasm can move from experimentation to retrenchment as pilots fail to produce measurable value. Vendor consolidation, tighter budgets and skepticism toward broad transformation programs would be normal responses. Crisis-management attention could also rise after a visible operational or leadership event and then fade as immediate pressure recedes.
In the medium term, fundamentals will separate companies that have integrated AI from companies that have merely licensed it. The relevant indicators are not tool counts. They are cycle time, error rates, control incidents, employee capability, data quality and the share of material decisions with a named accountable owner. A company that simplifies governance while improving information flow should be better positioned to convert automation into durable performance.
In the long term, the structural change is organizational. AI will alter how work is divided between people and machines, how leadership pipelines are built and how boards oversee operational risk. The companies most exposed are those whose advantage depends on slow, fragmented decisions or on a narrow group of senior experts. The potential beneficiaries are organizations that combine automation with broad managerial capability, independent challenge and reliable local intelligence.
The base case is uneven scaling: companies move AI into selected core processes while tightening controls and abandoning low-value pilots. The upside case is a productivity-and-resilience combination, triggered if firms simplify governance, improve data infrastructure and demonstrate lower error or disruption rates as automation expands. The downside case is an automation-control gap, triggered if agentic systems multiply operational or cyber incidents faster than boards can identify owners and impose limits.
Three signals deserve attention. The first is whether AI programs move from pilots into audited, revenue- or cost-linked processes. The second is whether companies disclose material control, cyber or operational incidents associated with automated systems. The third is whether leadership succession plans and crisis exercises include AI-enabled operations rather than treating technology as a separate workstream.
Those signals matter because the easy AI narrative is familiar: more adoption should mean more productivity. The harder question is whether productivity remains available when the system is stressed. That is where corporate value, not technological novelty, will be tested.
The interview’s message is ultimately a governance message. Crisis readiness is not a plan stored for emergencies, and AI adoption is not a software purchase. Both are tests of whether an organization can assign responsibility, move reliable information and preserve judgment when conditions change.
AI’s first corporate cycle will be about pilots and budgets; its lasting impact will be decided by who still has authority when the model is wrong.
Explore more exclusive insights at nextfin.ai.
