A wave of research published over the past three weeks has converged on the same uncomfortable truth for corner offices everywhere: artificial intelligence can now draft a market-entry memo, model twelve pricing scenarios, and summarize a thousand pages of competitor filings before lunch, but it still cannot decide what the company is actually for. On October 19, 2026, that tension sits at the center of a growing body of work from Harvard Business Review, McKinsey & Company, Thoughtworks, EY, Deloitte, and the Institute of Chartered Accountants in Ireland, all published or updated within the last three weeks. The consistent message: AI is rewriting how strategy gets built, but the people who own the consequences of a bad call are still, for now, people.
That is not a small distinction. Strategy work used to be bottlenecked by how fast a team could gather data and build a model. Generative AI tools have mostly solved that bottleneck. What is left, according to the latest round of research, is a judgment bottleneck: the scarce human skill of deciding what the data actually means, which assumptions to trust, and who gets blamed if the bet goes wrong.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
The Reports Driving This Week’s AI Strategy Debate
Harvard Business Review published “How AI Is Changing Strategy: Rethinking What Your Business Is Designed to Do” on October 8, 2026, arguing that the function of strategy itself is shifting, not just the tools used to write it. That piece followed an earlier HBR article, “AI Is Revolutionizing Strategic Decision-Making,” published September 1, 2026 and updated October 2, 2026, which laid out the groundwork for treating AI as a participant in strategic analysis rather than a back-office productivity tool.
Thoughtworks added its own data point on October 15, 2026, with “The shift to the AI-first enterprise: Insights from technology leaders forum 2026,” updated two days later on October 17. EY weighed in with “AI in business: How CEOs turn productivity into growth,” first published October 1 and refreshed October 15. Both pieces describe the same pattern from different angles: executives are moving past pilot projects and asking how AI changes the shape of the org chart, the budget, and the decisions a CEO is personally willing to sign off on.
None of this is happening in a vacuum. The debate about AI model capability has been just as busy this quarter as the debate about AI strategy, and the two are feeding each other. Our coverage of the ChatGPT vs Claude vs Gemini comparison and the Grok vs Claude vs DeepSeek pricing gap both point to the same underlying shift: frontier models are now cheap and fast enough that the limiting factor in enterprise AI adoption is no longer model quality. It is organizational trust in what the model’s output should be allowed to decide.
McKinsey Draws a Line: What AI Can Automate, What It Can’t
The sharpest framing of the split comes from McKinsey & Company, which argues that AI can meaningfully accelerate research, interpretation, synthesis, simulation, and communication inside a strategy process. What it cannot yet do, according to McKinsey’s analysis, is interpret the implications of a finding for the business, challenge the assumptions baked into a forecast, weigh trade-offs that have no clean numerical answer, or own a decision once it is made.
McKinsey splits business decisions into two rough buckets. The first is routine, high-volume, and pattern-based: the kind of decision a company makes thousands of times a day where the cost of a single mistake is low and the value of speed is high. The second is the opposite: rare, high-stakes, and dependent on judgment calls that historical data cannot fully capture.
Routine Decisions AI Can Already Handle
McKinsey specifically names dynamic pricing and promotions, demand forecasting and inventory routing, and real-time fraud and credit decisions as examples of decisions that can often be automated with the right guardrails in place. These are decisions companies already run through software, just not always through AI models, and they are the easiest category to hand to a machine because mistakes are cheap, data is plentiful, and the feedback loop is fast.
Strategic Decisions That Still Need a Human Owner
By contrast, McKinsey says decisions like entering a new market, repricing during a disruption, or scenario planning when historical patterns can no longer be trusted should stay human-led, even if AI plays a support role in gathering and summarizing information. These are the moments where a wrong call can sink a quarter or a company, and where the “right answer” often cannot be derived purely from the numbers in front of you.
| Decision Category | Example | Who Leads, Per McKinsey |
|---|---|---|
| Routine, high-volume | Dynamic pricing and promotions | AI, with human-set guardrails |
| Routine, high-volume | Demand forecasting and inventory routing | AI, with human-set guardrails |
| Routine, high-volume | Real-time fraud and credit decisions | AI, with human-set guardrails |
| Rare, high-stakes | Entering a new market | Human-led, AI-supported |
| Rare, high-stakes | Repricing during a disruption | Human-led, AI-supported |
| Rare, high-stakes | Scenario planning under unreliable historical data | Human-led, AI-supported |
What makes this framing useful is that it gives executives a test they can actually run on any given decision: how often does it recur, how reversible is a mistake, and how much does the answer depend on data the company has never seen before. Decisions that score low on all three are candidates for automation. Decisions that score high on any one of them are not.
Nadella: “Copilot, Not Autopilot”
Microsoft CEO Satya Nadella has made the human-in-the-loop argument central to how his company talks about enterprise AI, and his comments are echoing through this fall’s strategy debate. “I think that one big design decision we made was to think about this as a copilot. Not as an autopilot, but as a copilot. Designing in such a way that the human is in control. Human agency, human judgment is what is still the core and then building the product around it,” Nadella said in remarks carried by NPR.
He extended the metaphor further, describing how workers should treat AI-generated output as a starting point rather than a finished product. “You are the editor. That metaphor of thinking of anything that gets generated as a first draft, whether it’s a piece of code or a piece of text or an image, that’s probably a good way for us to think about how things get accelerated. But at the same time, humans have to do their job, too,” Nadella told NPR.
Nadella also described the practical role AI plays inside a strategy workflow in blunter terms, framing the technology as a tool for navigating information rather than a replacement for the people responsible for acting on it. “It’s a reasoning engine that goes and looks at all the data and tries to help you search through it and reason over it,” he said in the same NPR interview.
“Human Capital and Token Capital”: Microsoft’s Framing of the New Balance Sheet
In separate remarks posted to X, Nadella sketched out what he sees as the resource allocation problem every company now faces. “Every company is going to have to build what I think of as human capital and token capital,” he wrote, a framing that recasts corporate budgeting as a balance between people and compute, rather than people replaced by compute.
He was equally direct about what happens when that balance tips too far toward automation without a clear human mandate behind it. “Without human direction, you have compute running in circles,” Nadella wrote, arguing that human goals, relationships, pattern recognition, and judgment are what give AI systems a reason to generate business value in the first place rather than simply generating output.
This is a notably different posture than the “AI replaces workers” narrative that dominated headlines earlier in the AI boom. It also lines up with how Microsoft has positioned its own AI products, including the Copilot brand that is now under regulatory scrutiny in the UK, where the Competition and Markets Authority has flagged Microsoft’s grip on roughly 20 million Copilot users as a potential strategic market status concern.
Mollick’s Formula: Leadership, Lab and Crowd
Wharton professor and AI researcher Ethan Mollick has become one of the most cited voices on how companies actually operationalize AI adoption rather than just talk about it. Mollick’s view, as described in recent coverage, is that effective organizational AI adoption requires “leadership, lab and crowd”: leadership that sets direction and removes obstacles, a dedicated lab function that experiments and builds internal expertise, and a crowd of regular employees empowered to use the tools in their day-to-day work.
That three-part structure matters because it rejects two failure modes that have plagued corporate AI rollouts: a top-down mandate with no grassroots experimentation, and a scattered bottom-up adoption with no leadership backing or coordination. Mollick’s framing suggests the companies seeing real strategic value from AI are the ones running all three layers at once, not picking one and hoping it is enough.
Thoughtworks and the AI-First Enterprise
Thoughtworks’ October 15 report, drawn from its technology leaders forum, describes a cohort of enterprise technology leaders who are no longer debating whether to adopt AI but are wrestling with how deeply to restructure around it. The report’s own language, “the shift to the AI-first enterprise,” signals a move away from treating AI as a bolt-on capability and toward treating it as a design constraint that shapes how teams, workflows, and even org charts get built from scratch.
That shift mirrors what is happening on the model side of the industry. Enterprise buyers now routinely compare multiple vendors before committing to an AI strategy, which is part of why pricing comparisons like Gemini 4 Argon’s public launch and pricing structure draw so much attention from procurement teams, not just developers. The AI-first enterprise framing treats model selection as a strategic decision in its own right, not a technical footnote delegated entirely to IT.
EY: CEOs Turn Productivity Into Growth
EY’s “AI in business: How CEOs turn productivity into growth,” first published October 1, 2026 and updated October 15, tackles a different piece of the puzzle: what happens after a company successfully squeezes productivity gains out of AI tools. EY’s framing treats productivity as a necessary but insufficient outcome. The harder, more strategic question is whether those efficiency gains translate into top-line growth, new markets, or genuinely new products, rather than simply doing the same work with fewer people.
That distinction between cost savings and growth is becoming a dividing line in how boards evaluate AI investment. A company that uses AI purely to cut headcount in support functions is playing a different game than one using AI-accelerated analysis to identify and enter a new market faster than competitors. EY’s research frames the latter as the harder, more valuable, and more strategically defensible use of the technology.
Deloitte’s Caution: Technology Can’t Replace Purpose and Values
Deloitte’s research strikes a more cautionary note than some of the productivity-focused coverage elsewhere this quarter. Deloitte states plainly that technology can accelerate analysis and clarify uncertainty, but it cannot replace human purpose, values, and judgment. That is a different kind of argument than McKinsey’s task-level split between automatable and non-automatable decisions. Deloitte’s point is closer to a philosophical guardrail: even a perfectly calibrated AI model cannot tell a company what it should value, only what the data suggests about a given path.
This matters in practice when AI-generated recommendations conflict with a company’s stated values or long-term identity. A model trained to maximize a narrow metric, like quarterly margin, has no mechanism for weighing reputational risk, employee morale, or long-term brand positioning unless a human explicitly built those constraints into the system. Deloitte’s framing is a reminder that “the model recommended it” is not the same thing as “it is the right call for this business.”
Governance Catches Up: Boards Told to Treat AI as a Core Strategic Issue
While consultancies debate how AI changes strategy work, the governance side of the conversation is moving just as fast. The Institute of Chartered Accountants in Ireland published “The AI Brief, 16 October 2026,” which recommends that corporate boards treat AI as a core strategic issue rather than a delegated technology matter. The recommendation centers on three pillars: focusing on value creation rather than novelty, and establishing clear governance and accountability structures before, not after, AI systems are embedded in decision-making.
That governance push is not happening in isolation. It follows a year of rising scrutiny on how AI companies themselves are governed, including proposals like Anthropic’s pitch for a $1 billion AI safety threshold for lawmakers to consider, which drew its own backlash from parts of the industry. Boards adopting AI into their strategic processes are watching both sets of rules unfold at once: how regulators plan to govern the AI vendors, and how their own internal governance needs to catch up to the pace of adoption.
How the Major Research Houses Compare
Lined up against each other, the major reports published this month are not contradicting each other so much as attacking the same problem from different altitudes. HBR and McKinsey are the most granular, offering frameworks executives can apply directly to specific decisions. Thoughtworks and EY are more structural, focused on how the enterprise itself needs to reorganize to capture AI’s value. Deloitte and the Institute of Chartered Accountants in Ireland are the most cautionary, emphasizing the limits of the technology and the governance structures needed to keep it in check.
| Publisher | Report | Published / Updated | Core Angle |
|---|---|---|---|
| Harvard Business Review | How AI Is Changing Strategy | October 8, 2026 | Strategy’s function is being redesigned, not just its tools |
| Harvard Business Review | AI Is Revolutionizing Strategic Decision-Making | September 1, updated October 2, 2026 | AI as a participant in strategic analysis |
| Thoughtworks | The shift to the AI-first enterprise | October 15, updated October 17, 2026 | Enterprises restructuring around AI by design |
| EY | AI in business: How CEOs turn productivity into growth | October 1, updated October 15, 2026 | Converting productivity gains into real growth |
| McKinsey & Company | The judgment gap in strategic decision-making | 2026 analysis | Splitting decisions into automatable and human-led categories |
| Deloitte | Decision-making with AI | 2026 Human Capital Trends research | Technology accelerates analysis, not purpose or values |
| Institute of Chartered Accountants in Ireland | The AI Brief | October 16, 2026 | Boards must govern AI as a core strategic issue |
None of these reports is arguing that AI’s role in strategy will shrink. All of them, in different language, are arguing that the hardest part of strategy was never the analysis. It was the judgment call at the end, and that part of the job is not going anywhere soon.
From Expert Systems to Large Language Models: A Short History of AI in Corporate Decision-Making
The current debate has echoes of earlier waves of decision-support technology, even if today’s tools are far more capable. In the 1980s, corporate “expert systems” promised to encode human expertise into rule-based software that could replicate specialist judgment. Most of those systems struggled outside narrow, well-defined domains because real business decisions rarely fit the clean rule sets the technology required.
The business intelligence and analytics wave of the 2000s and 2010s shifted the emphasis from encoding expertise to surfacing data, giving executives dashboards and predictive models instead of rule-based recommendations. That wave improved the quality of information available for a decision without fundamentally changing who made the call. Large language models, emerging commercially from 2022 onward and maturing rapidly through 2025 and 2026, represent a third wave: tools that can synthesize unstructured information, draft analysis, and simulate scenarios in natural language rather than structured data alone.
What distinguishes this wave from the prior two, according to the current round of research, is not raw capability so much as scope. Expert systems handled narrow domains. BI dashboards handled data visualization. Today’s models can plausibly participate in almost any knowledge-work task, which is exactly why the judgment question has become so urgent. The tools are no longer limited to one function; the limiting factor is organizational trust in how far to let them go.
What This Means for the Enterprise AI Market
The practical market impact of this debate is already visible in how enterprise AI vendors position their products. Major vendors have increasingly moved away from language suggesting AI will “replace” knowledge workers, instead adopting the copilot framing that Nadella and others have pushed. That shift in marketing language follows the research, not the other way around: companies buying enterprise AI tools have made clear through procurement decisions that they want augmentation, not full automation, for anything touching strategic judgment.
This also reshapes how vendors compete. Instead of competing purely on raw model benchmarks, vendors are increasingly selling governance features, audit trails, and human-approval workflows alongside their core models, a trend reflected in how GitHub Copilot has retired older models in favor of newer ones built with stronger guardrails. The market is rewarding vendors who can demonstrate that their tools keep a human in the approval loop for high-stakes outputs, not just vendors with the fastest or cheapest model.
For enterprise buyers themselves, the research points to a widening gap between companies that have built the “leadership, lab and crowd” structure Mollick describes and companies still running isolated, uncoordinated AI pilots. The former group is starting to report AI-driven changes to how they build strategy itself. The latter group, per EY and Thoughtworks’ findings, is largely still capturing productivity gains without translating them into growth or structural change.
Five Predictions for AI in Business Strategy Through 2027
- Board-level AI governance becomes standard practice. Following the Institute of Chartered Accountants in Ireland’s recommendation, expect more public companies to formalize AI oversight committees or expand audit committee mandates to cover AI governance explicitly, rather than leaving it to IT or a chief AI officer alone.
- The “copilot vs autopilot” framing hardens into a product category. Vendors will increasingly market tiered AI products: fully automated tools for routine, low-stakes decisions like the ones McKinsey describes, and explicitly human-gated tools for anything touching strategy, legal exposure, or public communications.
- Productivity-to-growth conversion becomes the new AI ROI metric. Following EY’s framing, boards will push beyond simple productivity measurement and start demanding evidence that AI-driven efficiency is translating into new revenue, markets, or products, not just lower headcount.
- Mid-sized companies without a “lab” function fall further behind. Mollick’s three-part adoption model favors organizations large enough to fund a dedicated AI experimentation team. Smaller companies relying purely on leadership mandates or grassroots crowd adoption, without a lab layer, are likely to see slower and less structured AI-driven strategic change.
- Human judgment becomes an explicit line item in strategic planning. Expect more companies to formally document which categories of decisions remain human-led, following the McKinsey model, as a defensive measure against regulatory and legal scrutiny of AI-assisted decisions that go wrong.
The Bottom Line for Executives Right Now
Strip away the branding differences between HBR’s, McKinsey’s, EY’s, Thoughtworks’, and Deloitte’s reports, and the practical guidance converges on roughly the same checklist. Use AI aggressively for research, synthesis, and simulation. Build organizational structures, following something close to Mollick’s leadership, lab and crowd model, so adoption is coordinated rather than scattered. Draw an explicit line around which decisions stay human-led, using something like McKinsey’s recurrence-and-reversibility test. And treat AI governance as a board-level responsibility, not a technical afterthought, per the Institute of Chartered Accountants in Ireland’s guidance.
None of the research published this month argues that AI’s role in strategy will plateau. If anything, the opposite: every report describes a technology getting rapidly better at the research, analysis, and simulation work that used to consume weeks of an analyst’s time. The consistent argument is narrower and, in a way, more reassuring for the people doing strategy work today: the judgment at the end of the process, the part where someone has to own the consequences of being wrong, is still a human job.
Frequently Asked Questions
Is AI actually replacing business strategists and executives?
No. The current body of research from McKinsey, Deloitte, HBR, EY, and Thoughtworks consistently frames AI as an accelerant for research, analysis, and simulation, not a replacement for the people who own strategic decisions. McKinsey explicitly states that human judgment remains necessary for interpreting implications, challenging assumptions, weighing trade-offs, and owning decisions.
Which business decisions can safely be automated with AI right now?
According to McKinsey, decisions that are routine, high-volume, and pattern-based are the best candidates for automation with human-set guardrails. Examples cited include dynamic pricing and promotions, demand forecasting and inventory routing, and real-time fraud and credit decisions.
Which decisions should stay human-led even with AI support?
McKinsey names entering a new market, repricing during a disruption, and scenario planning when historical data becomes unreliable as examples of decisions that should remain human-led, even when AI tools assist with the underlying analysis.
What does Satya Nadella mean by “copilot, not autopilot”?
Nadella uses the phrase to describe Microsoft’s product design philosophy for AI tools: systems built so a human stays in control of the final decision, with AI providing accelerated drafts, research, and reasoning support rather than operating autonomously on high-stakes choices.
What is Ethan Mollick’s “leadership, lab and crowd” model?
It is Mollick’s framework for effective organizational AI adoption, built on three layers: leadership that sets direction and clears obstacles, a dedicated lab function that experiments and builds internal expertise, and a crowd of regular employees empowered to apply AI tools in their everyday work.
Why are corporate boards suddenly focused on AI governance?
The Institute of Chartered Accountants in Ireland’s October 16, 2026 brief recommends boards treat AI as a core strategic issue, with clear governance and accountability structures, rather than delegating oversight entirely to technology teams. That recommendation follows a broader pattern of regulators and industry bodies pushing for board-level accountability as AI tools move into higher-stakes business functions.
How is this different from earlier waves of business decision-support technology?
Earlier technologies, like 1980s expert systems and 2000s-era business intelligence dashboards, were limited to narrow domains or structured data. Current large language models can synthesize unstructured information and participate across nearly any knowledge-work task, which is why the current debate centers on scope and trust rather than raw capability alone.
What should a company do differently in response to this research?
The research converges on a few practical steps: use AI aggressively for research and simulation, build coordinated adoption structures rather than scattered pilots, explicitly define which decision categories remain human-led, and treat AI governance as a board-level responsibility rather than a purely technical matter.
