The AI delusion: why CEOs still don’t get it

AI is being sold as the ultimate cost-cutting machine. Boards want headcount down, margins up, and they want it fast. But what if that entire framing is wrong?

Eric Ries, author of The Lean Startup and long-time startup advisor, argues that our biggest AI fears aren’t really about AI at all. They’re about a corrupted form of capitalism that rewards extraction over value creation—and that same logic is now shaping how we build and deploy AI.

When making money stops creating value

For most of the history of capitalism, there was a clear moral logic: if two people voluntarily trade, fully informed and without coercion, both walk away better off. No new stuff is created in that moment, but new value is.

Today, that logic is breaking. Ries points out that we’ve normalized entire business models that make money without creating value—and often by quietly destroying it. Many of these practices, he notes, “our grandparents would have called crimes.”

Examples include:

• Financial engineering that boosts stock prices while hollowing out the underlying business
• Short-term cost cutting that kills innovation to hit quarterly targets
• Private equity playbooks that “surgically debone” companies and brands people love

We’ve legalized and celebrated behavior that earlier generations would have seen as outright corruption. And that same mindset is now being applied to AI.

Financial gravity: the invisible force inside every company

Ask almost anyone inside a large organization what’s wrong and you’ll hear the same themes: short-termism, cost cutting over innovation, and decisions that make no sense if you care about the long term.

Ries describes a familiar blame chain:

• Employees: “My manager is forcing us to kill good projects to make the quarter.”
• Managers: “I don’t like it either, but my VP wants it.”
• VPs: “The CFO is pushing this.”
• CFO: “The CEO told me we have to.”
• CEO: “The board wants it.”
• Board: “Investors demand it.”
• Investors: “We don’t actually want this.”

No one individually owns the decision, yet the outcome is always the same: squeeze, cut, extract.

Ries calls this force financial gravity. Just as physical gravity pulls objects down whether you believe in it or not, financial gravity pulls organizations toward short-term extraction, even when everyone involved insists they don’t want that.

Why “good governance” is often value-destroying

Modern corporate governance is built around a single idea: shareholder primacy. The purpose of the company, in this view, is to maximize shareholder value—everything else is secondary.

But Ries points out something deeply ironic: companies rated as having the “worst” governance by traditional metrics have actually outperformed those with the “best” governance since 2008. The governance model that’s supposed to maximize shareholder value is, in practice, destroying it.

Why? Because it encodes extraction as a best practice. It rewards:

• Raising prices because “customers won’t notice yet”
• Cutting wages and benefits to hit margins
• Slashing R&D and long-term bets to juice earnings

Over time, this corrodes the most underrated asset in business: trustworthiness.

The Costco and Saul Price playbook: customers first, shareholders last

To understand what resisting financial gravity looks like, Ries tells the story of Saul Price—the largely forgotten father of modern retail and the spiritual ancestor of Costco.

Price started FedMart in the 1950s with a radical idea: as a former lawyer, he believed he had a fiduciary duty to his customers, not just his investors. His hierarchy was simple:

• Customers first
• Employees second
• Shareholders last

He lived this in extreme ways. When competitors sold items below cost to undercut him, he literally posted their ads inside his own stores and told customers to buy from them instead. The result: customers trusted him to look after their financial interests.

FedMart thrived—until investors pushed him out in pursuit of faster, more conventional growth. Within seven years, they had driven the company into liquidation. They killed the goose that laid the golden eggs because they didn’t understand that the real asset was accumulated trust.

Price started over with Price Club. One of his protégés, Jim Sinegal, later merged Price Club with his own company to form what we now know as Costco.

How Costco built a “governance fortress”

Costco combines two things FedMart never had at the same time:

• Ethos: a deep, lived commitment to serving customers and employees first
• Integrity: a legal and governance structure designed to protect that ethos from financial gravity

Some examples of Costco’s approach:

• Margins on goods are capped (around 14%), even though raising them slightly would massively boost profits
• The famous $1.50 hot dog has never increased in price, despite enormous pressure to do so
• They pay suppliers faster and more fairly than they need to
• They conduct an enormous share of food safety inspections in the US—benefiting not just their own customers, but the broader public

On paper, many of these decisions are ROI-negative in the short term. But over decades, they build a moat of trust and loyalty that’s incredibly hard to copy.

To protect this model, Sinegal designed what Ries calls a governance fortress: legal and structural protections that make it hard for future leaders or outside investors to quietly flip Costco into a conventional, extractive retailer.

AI labs as the new corporations—and why their values matter

Ries makes a provocative claim: corporations are humanity’s first artificial intelligences. They’re emergent systems with goals, memory, and behavior that no single person fully controls. AI labs are just the latest, more literal version of this.

That’s why the alignment problem—how to ensure AI systems act in line with human values—isn’t just a technical issue. It’s also a governance issue: who aligns the aligners?

At a Vatican conference on AI governance, Ries noticed something striking: none of the major AI labs (Anthropic, OpenAI, Google, Palantir, Cohere, etc.) use a plain-vanilla, shareholder-primacy structure. Every one of them has tried, in some way, to build in mission protections.

That’s not an accident. Even the people building these systems instinctively feel that handing them to a pure profit-maximizing machine would be dangerous.

The real danger: AI as a cost-cutting weapon

Despite all the talk about superintelligence and existential risk, Ries thinks the most immediate danger is much more mundane: using AI primarily as a cost-cutting tool.

Boards and executives are under huge pressure to:

• Replace workers with AI wherever possible
• Automate away creative and judgment-heavy roles
• Sell AI internally as “efficiency” and “headcount reduction”

That means we’re building systems where humans are effectively aligned to a goal of getting humans out of the loop. In other words, we’re training ourselves and our tools to see human creativity as a cost center to be eliminated.

Ries thinks this is both morally wrong and strategically foolish. The real promise of AI, he argues, is as an amplifier of human creativity, not a replacement for it.

If you’ve ever felt uneasy about headlines like “this is your last chance to get rich before AI replaces you,” you’re not alone. That framing reflects the same extractive logic Ries is warning about—and it’s exactly the mindset that leads to burnout, low-quality work, and a race to the bottom. (For a deeper dive on that mindset, see this analysis of AI hype and get-rich-quick thinking.)

AI as augmentation, not replacement

Used well, large language models (LLMs) can be incredible teaching and augmentation tools. They’re trained on vast swaths of human knowledge and can help fill in gaps in your understanding, challenge your thinking, and speed up tedious parts of creative work.

Ries draws a sharp line between two ways of using AI:

• Replacement mode: “Here are three bullet points, write me a 20-page proposal and send it.”
• Augmentation mode: “Help me think through this proposal, improve my reasoning, and learn how to write it better myself.”

In replacement mode, something disturbing happens. Heavy AI users can develop what he calls a kind of agent psychosis:

• Their skills atrophy because they’re not practicing them
• Their confidence goes up because the AI makes everything look polished
• The gap between what they think they can do and what they can actually do widens

In augmentation mode, the opposite happens: your skills and agency grow. You still get leverage from the tool, but you remain the author of your work.

Ries describes using a custom AI setup while writing his latest book. It didn’t write for him, but it:

• Helped him track where he was in a long list of tasks
• Prompted him back into flow after interruptions
• Acted as a thinking partner, suggesting bad drafts that his human brain could then improve

The result wasn’t AI-written sludge—it was more focused, more deeply considered human work.

How to actually use AI to get smarter

Ries shares a powerful example from a course his AI research lab, Answer.AI, ran with 1,000 students. Instead of just handing them his book to read, they taught students how to do AI-assisted close reading.

Close reading is a classic academic skill: carefully unpacking a text, understanding word choice, structure, references, and context. It usually takes years of training. With an LLM as a reading companion, students could:

• Ask what specific words or phrases meant in context
• Explore references and influences they didn’t recognize
• Analyze why a passage was written the way it was

Many reported it was one of the most satisfying reading experiences they’d ever had. The AI didn’t replace their thinking—it deepened it.

Contrast that with a common pattern Ries sees at work: someone writes three bullet points, uses an AI to inflate them into a 20-page report, emails it to someone else, and that person uses AI to summarize it back down to three bullet points. Millions of tokens, GPU cycles, and gallons of cooling water are burned to accomplish…nothing.

It’s a perfect example of how not to use AI.

AI, engagement, and the social media trap

There’s another danger: many AI products are being built by the same people who optimized social media for engagement at all costs. And they’re bringing the same playbook.

Ries notes that some chatbots are already being tuned like social feeds—to maximize usage, stickiness, and emotional dependence, not user flourishing. That’s how we end up with AI companions that encourage unhealthy behavior or chatbots that subtly nudge users toward addictive patterns. (If you’re curious about how this plays out in synthetic relationships, see this exploration of AI companions and the dark side of synthetic friendship.)

He argues we need an AI Bill of Rights—a clear set of user rights that labs and platforms commit to, including:

• A right to privacy in your AI interactions (for example, not having your chats subpoenaed like unprotected data)
• A right to be free from manipulative engagement-maximizing behavior
• Protection from AI-driven discrimination, fraud, and harm

Before we even argue about government regulation, Ries says, we need to agree on what humans should be entitled to in an AI-saturated world.

Why artisanal work will matter more, not less

With AI able to generate endless content, code, and images, it’s easy to assume human creators are doomed. Ries thinks the opposite is more likely.

As the cost of generic content drops toward zero, high-quality, clearly human work becomes more valuable. He expects:

• Verified human-created content to become a premium category
• True craftsmanship in writing, coding, design, and art to command higher prices
• New discovery tools that help people find the best work for them, not just the most optimized for clicks

We’re in a painful in-between phase right now, where social platforms reward outrage and celebrity over quality. But that equilibrium isn’t stable. As AI floods the zone with cheap content, the market for trusted, human, “artisanal” work is likely to grow.

Breaking the spell of shareholder primacy

Underneath all of this—AI, governance, corporate behavior—is one core idea: shareholder primacy. The belief that a corporation’s sole purpose is to maximize shareholder value.

Ries points out that this idea is surprisingly new. For most of corporate history:

• Companies were chartered for specific, beneficial purposes (like building a railroad)
• States could revoke a company’s charter if it abandoned that purpose in favor of pure enrichment

Only in the late 20th century did “any lawful purpose” quietly get reinterpreted as “maximize shareholder value,” largely through court decisions and academic theories—not democratic debate.

Today, legal scholars defend shareholder primacy by calling it a “normative consensus”: not just what companies do, but what we all supposedly agree they ought to do.

Ries’s key question: are you actually part of that consensus?

Most people he asks say no. They don’t believe companies should be allowed—even obligated—to break rules, harm communities, or lobby for worse laws if it increases profits. But they’ve rarely said that out loud, especially at work.

He argues that the first step in changing our economic system isn’t passing a law—it’s breaking the illusion of consensus. Tell someone you trust: “I don’t believe the purpose of a company is to maximize shareholder value at all costs.” You might be surprised how many people quietly agree.

What leaders can do right now

If you’re a founder, executive, or team lead who feels this tension—between what you’re being pushed to do and what you believe is right—Ries offers a few practical starting points:

• Clarify your real purpose: Write down, in plain language, what human flourishing outcome your organization exists to serve. Not a vague mission statement—an actual purpose.
• Rank your loyalties: Be explicit about your hierarchy (customers, employees, communities, shareholders) and live it in decisions.
• Build structural integrity: Explore governance tools like purpose clauses in your charter, mission guardianship, and director oaths that commit your board to more than extraction.
• Use AI to grow people, not replace them: Encourage AI use that deepens skills, speeds learning, and amplifies creativity—rather than automating away judgment and agency.
• Say the quiet part out loud: In conversations with investors, lawyers, and peers, state clearly that you are not optimizing for shareholder primacy alone.

None of this is easy. Financial gravity is real. But as companies like Costco, Patagonia, and others show, it’s possible to build organizations that resist it—and to be rewarded for doing so.

The real AI delusion isn’t that the machines are coming for us. It’s believing we can bolt AI onto a corrupt economic logic and somehow get a humane future out of it. If we want AI that serves people, we have to fix the system that’s telling CEOs what “success” looks like.