The AI Economy Is Separating Leaders From Spenders - Forbes

In July 2024, The Economist controversially asked, “what happed to the AI revolution?” The article explored the lack of groundbreaking or industry transforming examples. Now, here we are, two years later, asking the same question. The AI economy must now enter its proof-of-work era.

Executives are being handed two very different stories about AI. One says AI is coming for jobs. It will flatten organizations, eliminate entry-level roles, and give companies a new excuse to do what cost-cutting leaders have always done, reduce headcount and call it progress or efficiency gains.

The other says the opposite. A new study covered by the Financial Times found that companies investing most heavily in generative AI are adding workers faster than their peers. In the first two years after adoption, white-collar headcount at the highest-intensity AI adopters increased 10.2% overall. Entry-level employment increased 12%. The research analyzed nearly 22,000 U.S. companies and linked organization-level AI spending with workforce data.

At the same time, Bain & Company’s 2026 Automation and AI Pathfinder Survey found that AI budgets are growing even as many returns are falling short. Nearly 40% of companies that measured AI cost savings landed below 10%, despite targeting 11% to 20%. Yet 90% of those companies are increasing their budgets again.

So, which is it?

Is AI a growth engine or a cost-cutting machine? Is it creating jobs or replacing them? Is it delivering returns or becoming another expensive chapter in the long history of digital ‘transformation’?

The answer is less binary and more revealing: AI is exposing the quality of leadership behind the investment.

Companies that treat AI as a tool tend to get tool-level returns. Companies that treat it as an automation layer tend to get automation-level returns. Companies that treat it as a catalyst to rethink work, value creation, and operating models are starting to produce a very different story.

That is where the real divide is forming.

Spending Is not Strategy

The Financial Times report challenges the assumption that AI adoption automatically leads to broad job losses. The companies spending the most on AI are not necessarily shrinking. In the study, the highest-intensity adopters grew headcount faster than peers, while lower-intensity adopters showed no significant change compared with a control group.

But this does not mean “AI creates jobs” in some universal or automatic way. The findings come with important context. Ramp’s chief economist Ara Kharazian noted that headcount gains did not appear for at least six to 12 months and required a meaningful threshold of investment, not “a couple of dollars a month on ChatGPT.” The FT also noted that most gains were in the tech sector, the study covered white-collar workers, and one labor economist cautioned that it is difficult to separate “intense AI adopters grow faster” from “small fast-growing start-ups buy a lot of AI quite early.”

The lesson is not that AI spending equals hiring. The lesson is that serious AI adoption appears to correlate with companies that are already growing, learning, experimenting, and investing with enough intensity to change how work gets done.

This is the first executive trap.

Many leaders equate AI investment with AI progress. They count licenses, pilots, prompts, copilots, agents, and announcements. These are activity metrics, not transformation metrics. Adoption may show that people have access to AI. It does not prove that AI is changing the economics of the business.

ROI Is not Missing, It Is Often Misdefined

Bain’s research points to the other side of the problem. Companies are spending more, but many are not realizing the returns they expected. The report’s most important conclusion is that the fix is “organizational, not technological.” In other words, the technology may work and still fail to produce enterprise value if the company has not redesigned the work around it.

This is where the AI ROI conversation becomes misleading.

Most organizations try to measure AI through the language of previous automation waves: hours saved, costs reduced, tasks completed, tickets deflected. Those measures are useful, but incomplete. They assume the goal is to make the current business more efficient.

That is necessary. but it makes the company finite in its competitive as more ambitious companies explore innovation and business reinvention with AI.

The bigger opportunity is to ask what AI makes newly possible. Can the company respond faster to market shifts? Can it serve customers in ways that were previously too expensive or too complex? Can it improve decisions, reduce risk, personalize experiences, accelerate product development, or create new revenue streams? Can it turn saved time into new capacity rather than simply treating it as a financial abstraction?

This is why the question “Does AI have ROI?” is too broad to be useful.

AI does not have ROI in the abstract. A workflow has ROI. A customer outcome has ROI. A redesigned operating model has ROI. A new capability has ROI. A business reinvention agenda has ROI.

Without that specificity, leaders end up debating AI as if it were a single investment category instead of a new layer of intelligence that must be intentionally connected to value. This is what Dave Wright and I explore as a new era of ROI in our new book, Infinite: How visionary Leaders Transform Today’s Businesses into AI-Forward Companies. We call it Return on Intelligence. And it’s a timely conversation.

The Cost Problem Is Becoming Harder to Ignore

Ed Zitron’s epic critique of AI economics is intentionally sharp, but there is an executive signal worth taking seriously. His argument is that many organizations have been shielded from the true cost of AI through subscriptions, bundled access, rate limits, and opaque token pricing. As more enterprise use shifts toward token-based billing, companies are becoming more aware of how difficult it can be to measure not only AI’s value, but also the real cost of producing that value.

That does not mean every AI investment is wasteful. It does mean many AI business cases have been built with more optimism than operational discipline.

If a company cannot determine what a task costs, how often it succeeds, how much human review is required, what quality threshold must be met, and how the output contributes to a business outcome, then it does not yet have an AI ROI model. It has an AI usage model.

This will become more important as organizations move from experimentation to scale. In experimentation, ambiguity is tolerable. In scale, ambiguity becomes expensive.

The Autonomy Gap

One of Bain’s most important findings is that only 7% of companies are running fully autonomous agents in production. Most investment cases assume full automation economics, but the operating reality is still far more human. Many companies rely on human approval, guardrails, exception handling, and oversight.

Human oversight is exactly what should exist when AI systems touch customers, employees, compliance, financial decisions, safety, trust, or brand reputation.

The problem is the gap between the business case and the operating model.

If the CFO approved a plan based on full automation, but the business is actually running a human-in-the-loop workflow with escalations, reviews, rework, exception queues, and governance checkpoints, then the company is living with different economics than the spreadsheet promised.

This is one of the reasons AI value feels elusive. The return is calculated as if autonomy has already arrived. The operation is designed as if it has not.

Today, AI is a Circular Bet

Bain also found that 44% of companies plan to fund generative and agentic AI investments from savings generated by prior automation programs. On paper, this sounds disciplined. In practice, Bain warns that it can become a “circular bet with a structural leak” when prior savings came in below target and future investments are sized against projections rather than actual results.

This is how enterprise AI bubbles form inside companies.

Why? Because leaders fail to verify where value was actually realized before reinvesting in the next wave. The risk is not just overspending. The deeper risk is compounding weak assumptions.

A company that funds future AI with unrealized automation savings is pretty much scaling belief.

The Job Debate Is Too Small

The job-loss debate also needs to evolve. The FT notes that academic research has painted a mixed picture, including research showing a 16% reduction in early-career employment in AI-exposed roles and a Harvard paper finding declines in junior employment among AI adopters while senior roles were largely unaffected. The same article also notes that Oracle, Snap, Block, Cisco, and others have linked thousands of job cuts to AI.

This mixed picture is exactly what executives should expect. AI does not affect all work equally. It automates some tasks, augments others, compresses some roles, expands others, and creates entirely new needs around orchestration, supervision, governance, data stewardship, workflow design, and judgment.

The mistake is treating jobs as the primary unit of analysis.

Jobs are bundles of tasks, context, relationships, judgment, accountability, and institutional knowledge. AI can perform tasks and generate outputs. It can assist decisions and increasingly act within boundaries. But companies still need leaders to decide how work should be decomposed, redesigned, governed, measured, and improved.

In Infinite, Wright and I argue that AI-forward companies do not simply automate the work they inherited. They redesign the business around flows of value, with people and intelligent systems working together in new ways. The shift is from the org chart to the work chart, from isolated efficiency projects to enterprise reinvention, from doing the same things faster to asking what should now be done differently.

What Executives Should Do Next to Effectively Steer the AI Economy

Executives need to move AI from budget line to operating discipline.

Before approving the next wave of AI spend, leaders should stop asking, “Where can we use AI?” The better question is, “Which business outcomes must improve, and how should work change to achieve them?”

First, create a baseline before funding the next initiative. For every AI program, document the current cost, cycle time, quality level, error rate, customer impact, employee effort, and revenue or margin contribution of the workflow it intends to improve. Without a baseline, there is no ROI.

Second, assign one accountable business owner to every AI investment. One executive who owns the outcome end to end. If AI is deployed into customer service, claims processing, sales operations, finance, HR, procurement, or software development, someone must be accountable for whether the workflow actually improves.

Third, measure total economics, not just tool usage or performance. AI costs include licenses, tokens, integration, data preparation, human review, exception handling, compliance, rework, training, change management, and governance. Bain notes that only 7% of companies are running fully autonomous agents in production, while many still require human approval, guardrails, and exception handling. Those human-in-the-loop realities must be included in the business case.

Fourth, redesign the workflow before scaling the technology. AI placed on top of a broken process does not fix the broken process nor its performance. It accelerates the dysfunction and makes it harder to unwind. Leaders should map how work flows today, then redesign how it should flow with humans and agents working together. This is the shift Wright and I describe in Infinite as moving from the org chart to the work chart. The org chart tells you who reports to whom. The work chart shows how value is created, where intelligence is needed, where judgment matters, and where AI can responsibly act.

Fifth, convert saved time into captured value. Saving 10 hours does not automatically create 10 hours of value. The organization must decide where that capacity goes. Does it improve customer response times or customer experiences? Increase sales coverage? Reduce risk? Accelerate product launches? Improve quality? Expand service? Create new offerings? Time saved is only potential value until leadership redirects it toward a measurable outcome.

Finally, review AI returns as a learning system. Every major AI workflow should have a monthly operating review that asks: What improved? What did not? What surprised us? What did humans have to fix? What should agents handle next? What should never be automated? What new value became possible?

This is how companies build Return on Intelligence, the ability to turn learning velocity into realized value. ROI in the AI era is not simply return on investment. It is return on better decisions, faster adaptation, improved workflows, stronger customer outcomes, and new capacity for growth.

The executive mandate is clear.

Fund AI where there is a business outcome worth changing. Redesign the work required to change it. Measure the real economics. Assign ownership. Capture the value. Then reinvest what you learn.

That is how AI moves from tech expense to competitive advantage. And that is the difference between companies that spend on AI and companies that become AI-forward.