Why vibe coding is a slot machine (and what to do instead) | Eric Ries @ Lean Startup

Eric Ries is best known for The Lean Startup, a book that helped a generation of founders think in experiments, shorten feedback loops, and build with customers instead of assumptions.

But after more than a decade of watching those ideas spread across startups, enterprises, and even governments, he came to a harder realization: teaching people how to build quickly is not the same as teaching them what is worth building, or how to protect it once it starts to matter.

That is the gap behind his newer thinking and his new book, Incorruptible. The central argument is simple, but uncomfortable: many companies know how to create something valuable, yet fail to put the mission, structure, and governance in place to keep that value from being hollowed out later.

That becomes especially relevant in the current AI wave, where speed is intoxicating, demos appear before products, and teams can feel productive while actually learning less.

This conversation moves across all of it: what The Lean Startup left out, why mission protection usually starts too late, how metrics replace meaning, what AI changes about experimentation, and why “just give people what they want” is still a surprisingly difficult operating principle.

Why Lean Startup still matters

One of the reasons The Lean Startup has held up so well is also one of the reasons some people found it frustrating.

It was written from first principles, not as a tactical manual for one specific stage, function, or technology stack. Ries deliberately avoided filling it with trendy tools, tactical checklists, and detailed implementation recipes that would age quickly. Instead, he focused on durable ideas people could adapt in their own context.

That choice made the book useful far beyond early software startups. The ideas spread into:

• Large companies
• Small teams
• Nonprofits
• Government organizations
• Different industries with very different operating constraints

But the tradeoff is obvious. If someone wants a highly prescriptive framework for exactly what to do next, a first-principles book can feel incomplete. It tells you how to think and test, not how to copy a script.

Ries sees that as part of the point. If the method claims to be scientific, people should not be asked to accept it on faith. They should get enough tools to test the ideas themselves and adapt them to local conditions.

Even so, he believes something important was missing.

What Lean Startup left out

Over time, Ries wrote more about scale, organizational design, and innovation inside larger companies. That led to books like The Startup Way and later leadership-focused work.

Still, there was a deeper omission.

For years, the startup world repeated a familiar aspiration: founders should build products that change the world. The phrase sounded noble enough that few people stopped to define the direction of that change. The implicit assumption was that of course it meant improvement.

But in practice, many technology companies changed the world in ways that made life worse, not better. Some created enormous financial value while leaving the people involved rich and miserable. Others optimized themselves into mediocrity, drifting far from the original product vision that made them worth building in the first place.

Ries’s own critique of his earlier work is sharp: he encouraged people to build things worth protecting, but did not spend enough time teaching them how to protect those things.

And that matters because anything genuinely valuable will eventually come under pressure.

The problem with “change the world” without “for the better”

It is easy to dismiss mission as soft language for idealists. But Ries’s point is not that every founder must become a social philosopher. It is that failing to define what the company is for creates practical damage.

Even founders who are not especially driven by grand moral narratives usually carry some internal standard about what kind of company they want to build. They may not call it mission. They may not like terms such as “stakeholders” or “purpose.” But they still have convictions.

For example, they may believe:

• Quality should never be sacrificed for a small margin gain
• The product should solve a real problem rather than exploit attention
• The company should be the best possible place for a certain kind of talent to do meaningful work
• Customer trust matters more than short-term extraction

Those beliefs are not decorative. They shape decisions under pressure. And pressure always comes.

Without a clearly articulated purpose, teams slide toward whatever can be measured, priced, sold, or justified in a spreadsheet. What starts as a product company becomes a financial optimization machine.

Why it is always “too early” until it is too late

One of the strongest ideas in this conversation is also one of the simplest: founders are constantly told to postpone governance, culture, and mission protection.

The advice usually sounds reasonable.

• Wait until product-market fit
• Wait until the next round
• Wait until the company is larger
• Wait until there is more to govern
• Wait until there is more leverage

But the pattern repeats. There is always another milestone that makes the discussion feel premature. Then one day, the company is too entangled, too financed, too structurally committed, and too constrained to make the changes it wishes it had made from the start.

That is why Ries argues that mission protection should begin at the earliest stage, not after success arrives.

Paradoxically, the fact that a very young company has little to govern is exactly what makes governance easiest. There are fewer stakeholders to negotiate with, fewer systems to unwind, and fewer precedents to reverse.

The job early on is not to create bureaucracy. It is to set intention.

The builder’s intuition most founders already have

Ries shares a useful example of a founder who rejected the language of ESG, purpose, and stakeholder capitalism. He described himself in far more hard-nosed terms: he wanted to make money for investors and disliked mission-heavy vocabulary.

Yet when pressed on what kind of company he wanted to create, he became deeply animated. His goal was to build the best company engineers had ever worked at. He cared intensely about engineering standards and product quality. He would not compromise the craft just to unlock an extra 10% of margin.

That is mission, whether or not someone likes the word.

Ries calls this the builder’s intuition: the belief that the best path to durable value is to create something genuinely useful, meaningful, and loved, and then capture a fair share of the value created.

What many founders do not realize is how much this intuition collides with the dominant logic of modern finance.

The moment quality becomes the true north, shareholder value is no longer the only true north. That conflict often stays hidden until the company becomes large enough for the pressure to turn explicit.

Mission vs. shareholder primacy

This is where good companies often begin to fracture.

A team may sincerely believe its purpose is to build an excellent product, serve customers well, and protect quality. But if the formal structure of the company recognizes only maximizing shareholder value, the mission is vulnerable whenever tradeoffs arise.

And those tradeoffs rarely arrive in dramatic form. They usually come disguised as sensible optimizations:

• Reduce quality slightly to improve margin
• Add more friction to increase lock-in
• Increase ad load to boost monetization
• Make the product more addictive because retention numbers improve
• Cut corners customers may not notice immediately

Each decision can be defended individually. Over time, they accumulate into a business that no longer resembles its original promise.

That is why Ries argues that values cannot live only in the founder’s head. They need structural protection.

What founders can do from day one

Ries separates this into two parts.

First, a company needs ethos: its character, or what it does when nobody is watching. This is the invisible leader inside the organization, the shared sense of purpose that guides decisions when no one is there to police them.

Second, if that ethos is real, it should be protected by structure.

One of the most straightforward tools he recommends is converting to a Public Benefit Corporation, or PBC, in jurisdictions where that option exists. In many places, this can be done through a very simple filing. The point is to place the company’s purpose directly into its charter.

That purpose does not have to be grandiose. It can be specific and practical:

• Build safe and beneficial technology
• Create the highest-quality product in a category
• Serve a particular community well
• Protect long-term trust in a market

Ries views this as an obvious move for many founders because it creates protection against future scenarios where the company may otherwise be pushed to compromise mission in favor of short-term financial logic.

He also mentions founder-specific share structures, such as founders’ preferred shares, as another way to preserve enough control early on to guide the company toward a more durable governance model later.

The goal is not permanent founder dictatorship. In fact, Ries explicitly rejects that. Founder control should function as a bridge, not an endpoint. The long-term aim is an institutional architecture that preserves continuity, succession, and values beyond any one individual.

Culture matters because temptation is constant

When Ries talks about ethos, he is talking about the repeated moments in which a company has a chance to do the easy wrong thing.

Those moments happen constantly:

• Can the company get away with shortchanging customers?
• Can it make the product slightly worse while raising monetization?
• Can it lobby for an advantage rather than earn one through excellence?
• Can it exploit user behavior instead of serving user interests?

In larger organizations, there are often people Ries describes as torchbearers. These are the ones who still care deeply about the product, the user, and the mission. They are the people who repeatedly have to say no to spreadsheet-led ideas that make the product worse in exchange for local gains.

That is exhausting when the company lacks a clear operating system for defending principles.

What matters most is not whether people can produce a return-on-investment justification for every principled decision. It is whether the company understands that trust, product integrity, and customer respect are the source of the return in the first place.

Surrogation: when metrics replace the mission

One of the most useful concepts in the discussion is surrogation. This is what happens when a metric becomes a substitute for the thing it was supposed to represent.

A company might begin with the real goal of delighting customers. Because delighted customers stay longer, retention improves. Because retention improves, revenue grows. Because revenue grows, investors benefit.

Then the chain gets collapsed.

Instead of asking how to delight customers, the company asks only how to make retention go up. That opens the door to manipulative strategies that produce the number while undermining the underlying reality.

And the same pattern repeats one layer up:

• Retention replaces delight
• Revenue replaces retention
• Stock price replaces revenue

As long as the chosen number rises, people convince themselves the business must be healthy. In the short term, almost any metric can be inflated by consuming a more valuable hidden asset: trustworthiness.

That is why companies can appear successful for years while hollowing themselves out.

Why principle-driven leaders can seem difficult

Mission-driven leadership is often mistaken for stubbornness or impracticality. In reality, principle usually creates friction because it refuses the easiest compromise.

Ries points to several examples.

At Costco, the famous low-priced hot dog is not simply a pricing tactic. It reflects a deeper operating commitment to member trust and disciplined value. Raising the price may be rational in isolation, but irrational relative to the promise the company wants to keep.

At Patagonia, quality was not a slogan. It was an obsession. Product standards were argued over intensely. Even commercial opportunities that looked financially attractive could be rejected if they sent the wrong signal about the brand’s standards.

From the outside, this can look irritating, rigid, or overly idealistic. But the deeper logic is that principles force teams to confront reality creatively. Instead of compromising at the first obstacle, they have to find a better answer.

That is often where the breakthrough comes from.

Companies without deeply embedded principles simply take the easy tradeoff and move on.

The AI mania and the return of demos without products

Ries is enthusiastic about AI, but he is also skeptical of the current hype cycle.

One thing he finds striking is the return of something software had largely outgrown: flashy demos detached from real products. In earlier eras, companies often raised money on impressive presentations and never delivered something useful. SaaS helped reduce that pattern for a while.

Now it is back.

In the AI boom, many companies showcase demos, release models, or attract attention without having a meaningful product in the hands of customers. That is partly a consequence of hype and capital chasing returns, but it also creates confusion about what progress actually looks like.

Why Lean Startup becomes even more relevant in AI

Despite the noise, Ries believes the core Lean Startup logic is more relevant than ever in AI.

Its sweet spot has always been environments with:

• High uncertainty
• Fast cycle times

AI intensifies both.

Industries are being reshaped quickly, which drives uncertainty through the roof. At the same time, teams with new tools can build and test dramatically faster than incumbents. That combination creates the perfect environment for experimentation, learning, and rapid iteration.

Ries offers a vivid example from outside software: Samsung’s entry into the US appliance market.

While established manufacturers worked on long product cycles, Samsung shipped refrigerators and other appliances at a much faster cadence, often with strange features and uneven quality early on. Incumbents dismissed many of those early products as cheap and odd. But over several years, Samsung learned faster. Its cycle time advantage became a compounding competitive advantage. By the time traditional players recognized what was happening, they were already behind.

The lesson is broader than appliances. Whenever a newcomer can learn much faster than incumbents in a changing market, the competitive balance shifts quickly.

The danger of vibe coding

Ries’s warning on AI is not that it lacks power. It is that people are using it in ways that feel productive while reducing actual learning.

That is especially true in what is often called vibe coding: prompting tools to generate large amounts of code or content with minimal human understanding of what is being produced.

The seduction is obvious. If the artifact appears quickly, it feels like work has accelerated. But in startups, the artifact itself is rarely the most important output. The real value lies in learning:

• What the customer actually needs
• What product behavior matters
• What design choices work
• What assumptions were wrong

If AI generates the artifact without deepening human understanding, the team may produce more stuff while learning less. That can leave them with low-quality output and weak judgment at the same time.

The study that showed AI felt faster while making developers slower

Ries references a particularly striking study involving open-source developers. Participants were asked to randomly decide whether to use AI on a task or not, then report their experience.

The self-assessments suggested that using AI improved productivity by about 20%.

But when actual output was measured, the opposite happened: productivity was roughly 19% lower with AI.

That gap between felt productivity and real productivity is enormous. It suggests a phenomenon where using the tool creates a compelling sensation of momentum even when performance declines.

Ries compares that state to a kind of dark flow, similar to what happens in gambling systems. The interaction becomes absorbing and rewarding in itself. It feels like progress because the machine keeps producing something. But not all production is progress.

Why AI can become a slot machine

The slot machine metaphor fits because the feedback loop is emotionally powerful.

You enter prompts. The system instantly returns outputs. Some are surprisingly good. Some are weird. Some are almost right. The randomness itself is stimulating. It invites another try, then another tweak, then another revision.

That experience can be fun enough to disguise poor judgment.

Instead of carefully designing and validating work, people can drift into repeatedly asking the machine to improve itself according to its own standards. The result may be longer codebases, longer documents, and more artifacts that nobody fully understands or wants to review.

Ries describes exactly that trap: generating a large artifact, feeling productive, then realizing that neither he nor anyone else really wanted to read, audit, or take ownership of it.

Once that happens, the tool has not accelerated the project. It has blocked it.

What to do instead: use AI to amplify human creativity

The alternative is not to reject AI. It is to use it in ways that strengthen human learning rather than replace it.

Ries’s standard is clear: use tools that keep humans in the loop and enhance human creativity.

That means:

• Using AI to accelerate understanding, not bypass it
• Keeping design and judgment with people
• Structuring tasks so the tool supports, rather than substitutes for, thinking
• Paying attention to where the real bottlenecks are in a workflow

He also points out a common analytical mistake. Teams often measure how much faster AI makes one step in a workflow and then assume the entire workflow will speed up proportionally. But that step may not be the true bottleneck.

In software, writing code matters, but it is only one part of a larger system that includes design, understanding, architecture, debugging, communication, and iteration. Speeding up one piece does not automatically speed up the whole.

When AI is given good structure and oversight, however, it can be genuinely useful. Much like a junior employee, it performs better when the problem is well designed and the context is clear.

How leaders should actually learn AI

For companies trying to adopt AI well, Ries offers advice that is not particularly glamorous, but likely correct: leaders need direct understanding of the technology.

Not necessarily expert-level mastery, but enough technical literacy to separate capability from salesmanship.

That matters for two reasons.

First, AI is genuinely transformative in some contexts, so delegating all understanding is risky.

Second, hype cycles attract a great deal of snake oil. There are products being sold aggressively into organizations that the actual end users do not find useful at all.

The best defense is not cynicism. It is informed judgment.

Ries recommends spending real time learning how modern AI systems work at a conceptual level. For non-technical leaders, that might mean:

• Reading foundational material slowly
• Using AI itself as a study partner
• Taking a serious course on deep learning basics
• Asking naive questions until the system becomes less mysterious

He argues that if boards and executives truly believe missing the AI wave is existential, then investing a few hours or a few weeks in understanding the technology should be an obvious priority.

Why teams need time to learn, not just pressure to adopt

There is also an organizational lesson here.

Many companies say they want people to use AI, but do not make space for learning. They add expectations without reducing workload, which creates a predictable result: shallow adoption, tool fatigue, and frustrated teams.

A better approach is to deliberately allocate work time for exploration. That turns learning into an accepted part of the job rather than a side project employees are expected to absorb on top of everything else.

Ries praises that kind of institutional support because it signals that learning itself is a value, not an extracurricular activity.

And in a fast-moving technological shift, that matters enormously.

There is more happening in tech than AI

Even while deeply engaged with AI, Ries is careful not to let it eclipse everything else.

He points to a range of areas where important innovation is happening:

• Advanced aerospace and ultra-fast transport
• Climate and geoengineering technologies
• Weather modification
• Bioinformatics and computational biology
• New approaches to healthcare and pharmaceutical access
• Fintech models inspired by trust-based businesses like Costco

The broader point is that the venture and media conversation may currently be dominated by AI, but meaningful innovation is still happening across many domains. Staying curious outside the dominant narrative is part of avoiding strategic blindness.

The real hack: just give people what they want

At the end of the discussion, Ries offers a “hack” that sounds almost too obvious to count as one.

But that is exactly why it matters.

If founders want to build companies that endure, they should begin by asking what people actually want from them.

Employees want to work somewhere their humanity is respected, where the work is meaningful, and where mutual respect is real.

Customers want products and brands they can trust.

Partners want to work with companies that behave fairly and are not needlessly extractive.

Investors, at their best, want exposure to businesses with real transformative potential, not just shallow optimization games.

That sounds like common sense because it is common sense. The problem is that many companies talk themselves out of it. They replace it with more sophisticated-sounding strategies, MBA logic, dashboard worship, and financial engineering.

Meanwhile, the more durable strategy remains stubbornly simple:

• Make a great product
• Be trustworthy
• Treat people with respect
• Stand for something real
• Protect the mission early, before it becomes hard