The Value Of Vibe Coding (or, the Good FAAFO)

The following is an excerpt from the forthcoming book Vibe Coding: Building Production-Grade Software With GenAI, Chat, Agents, and Beyond by Gene Kim and Steve Yegge.

Sure, vibe coding makes you code faster—that’s the obvious selling point. But if you think speed is the whole story, you’re missing out on the juicy stuff. We’ve discovered that vibe coding creates value across five dimensions, which we’ve named FAAFO—fast, ambitious, autonomous, fun, and optionality. We explored them briefly in the Introduction, but we will go into more detail in this chapter.

Think of FAAFO as your new superpowers. You’re coding faster, and you’re now bold enough to risk projects you’d have laughed off as impossible before. You’re working solo on stuff that used to need teams. And because you’re lowering the cost of coordination, and the “people can’t read my mind” tax inherent in any collaboration, you and your team can work more autonomously. You’re having fun again, like when you first learned to code. And most powerful of all, you’re exploring multiple solutions simultaneously, picking the best one instead of committing to the first idea that seems workable. 

Write Code Faster

While speed is a clear value of vibe coding, it is arguably one of the most superficial benefits. It’s impressive, but we’ve had a lot of speedups before. The main value of going faster is the extent to which it multiplies the value in the other dimensions of FAAFO.

Consider the video excerpt tool that Steve helped Gene create (mentioned in the Introduction), which generated clips from podcasts and videos. They built the first working version in forty-seven minutes of pair programming using only chat coding, no agentic AI assistance. That’s pretty fast. Gene estimated that it would have taken them two to three days to write it by hand.

The key lesson we learned during that session: type less, lean on AI more. 

But we also found that sometimes AI can make things maddeningly slower and more frustrating. We’ve each experienced this firsthand. Gene spent hours going in circles with AI trying to get ffmpeg to properly position captions and images in video files. Steve wasted an afternoon wrestling with an AI collaborator that confidently insisted on different approaches, all of them wrong, to parsing command line arguments in Gradle build scripts.

It can take both vigilance and good judgment to recognize when you’re being led down a rabbit hole and need to change course. Vibe coders must learn to notice when AI is heading confidently down a wrong path and decide when to redirect or abandon unproductive approaches.

Despite these occasional challenges, we still love it. And when vibe coding isn’t possible (e.g., no internet connection or local LLM), many developers like us now choose not to code at all. Old-style coding by hand seems pointless. It’s like needing to get down a seventy-mile desert road, but you won’t have a car for a couple hours. It’s less work to wait for the car to come get you, as opposed to walking part of the way. It’s not worth the bother.

Who wants to write code by hand like some relic from 2010? Not us.

Be More Ambitious

Recall Gene’s first working version of the video excerpt tool, which previously would have taken days. Because of the time and effort required, he had originally deferred trying. This happens in organizations too. There could be many reasons why projects are never started: Perhaps the perceived benefit wasn’t high enough to warrant the work, or maybe the difficulty made the payoff not worth the investment, or possibly another opportunity offered a higher, more immediate return.

With vibe coding, Gene was able to complete work that otherwise would never have been undertaken. Projects that once seemed too difficult or time-consuming become feasible, opening new possibilities for what can be accomplished. Vibe coding reshapes the spectrum of what can be built, letting you be more ambitious.

Seemingly impossible projects move into the realm of possibility. Applications that would have required specialist knowledge across multiple domains can now be built by developers with AI assistance filling their knowledge gaps. Five-month projects become five-week projects, or sometimes five days. Ideas once considered too ambitious get tossed onto your to-do list without a care in the world.

Small-ish low-return jobs become quick wins, because it can be easier to do the work than to create the task. Documentation, tests, minor UI improvements, and small refactorings that were perpetually pushed aside can now take seconds or minutes instead of hours or days. These tasks get done, rather than accumulating in ever-growing “broken windows syndrome” backlogs. You can fix every window in town and keep them fixed for once.

As Cat Wu, product manager for Anthropic’s Claude Code team, observed: “Sometimes customer support will post ‘Hey, this app has this bug’ and then 10 minutes later one of the engineers will be like ‘Claude Code made a fix for it.’ Without Claude Code, I probably wouldn’t have done that…It would have just ended up in this long backlog.” There has always been a category of work where it was easier to fix than to record and prioritize. That category is bigger now with AI.)

This expanded capability leads directly to our next important dimension of value…

Be More Autonomous

In June 2024, Sourcegraph’s then Head of AI, Rishabh Mehrotra, showed Steve a demo of a multi-class prediction model he had created—from concept to deployment—in half a day using vibe coding. He told Steve it would have been a whole summer intern project, or perhaps six weeks for a superstar intern, as recently as a year prior. Rishabh was shocked that he had completed it alone in a few hours. 

Rishabh had only discovered it was easy because he didn’t have the budget to hire an intern. So, in desperation, he figured he’d try it alone with AI. He finished so fast he—an AI expert—was flabbergasted.

This illustrates the third dimension of value that vibe coding enables. Developers (and teams) can accomplish tasks autonomously (and in some cases, alone) that otherwise would have required help from other developers or sometimes teams. Working with multiple people introduces significant challenges—communication and coordination, competing priorities, merging work—and the more people involved, the less time you spend solving the problem. 

Working autonomously frees you to do the work you need to do, enabling independence of action. (This is a term we’ll use throughout the book). Steve experienced this firsthand as a leader of one of Amazon’s first “two pizza teams” created to reduce customer contacts per order. The mandate was simple: give small, cross-functional teams complete ownership of their problem space with full capability to deploy solutions without navigating layers of dependencies and approvals. If reducing customer contacts means changing the checkout flow, rewriting the help system, or building new infrastructure, the team could do it all. No waiting for the UX team’s roadmap. No negotiating with the infrastructure team’s priorities. No endless meetings to align seventeen different stakeholders. 

This radical autonomy and independence of action transformed how fast Amazon could move from identifying problems to shipping solutions. Now, with AI as your tireless collaborator, you can achieve this same independence of action as an individual developer.

Beyond eliminating organizational friction, AI also helps solve an equally difficult problem: the “mind reading” tax inherent in collaboration. Let’s face it—no matter how skilled our teammates are, something inevitably gets lost when we try to convey what’s in our heads. When vibe coding autonomously, this universal challenge becomes less of a problem. You can implement what you envision because there’s no gap between your idea and its execution. You know it’s right when you see it because it matches the picture in your head.

The consequences of these two taxes show up across every domain where experts and novices collaborate. For fifteen years, Dr. Matt Beane studied this phenomenon, with surgical robotics providing a compelling example. Traditionally, junior surgeons learned by necessity—procedures required three or more hands, making their participation essential while creating natural apprenticeship moments. However, when surgical robots enabled senior surgeons to operate independently, these teaching opportunities disappeared despite training remaining an official responsibility. 

The senior surgeons, given the choice, overwhelmingly chose to work alone. This wasn’t because they didn’t value teaching; it was because coordination costs are often higher than we acknowledge. Every explanation, every correction, every moment spent bringing someone else up to speed represents time not spent on the primary task. When the surgical robots removed the physical necessity of assistants, the true cost of coordination became visible through the seniors’ behavior.

This same pattern appears in software development. If it’s possible to create things without external dependencies, without any need to communicate and coordinate with others to get what we need, the advantages multiply rapidly. The constant back-and-forth of explaining requirements, correcting misunderstandings, and reconciling different mental models disappears.

Economist Dr. Daniel Rock (famous for his work on the OpenAI jobs report) calls this the Drift, borrowing from the movie Pacific Rim, where two pilots mentally connect to operate giant mechs. When you and your team vibe code, you can create that kind of mind-meld with AI assistants, reducing the coordination costs that typically slow down multi-human teams.

With the Drift active, a product owner can directly work with the code base through AI rather than writing a detailed products requirement document (PRD). A developer can evolve the database schema without a database specialist. As Dr. Rock demonstrated with his three-person team that built a GitHub app in forty-eight hours, this shared mental model accelerates development in ways that traditional human-to-human coordination cannot match. Being autonomous with AI means being unblocked—free to move at your own pace without constant negotiation and handoffs.

Scott Belsky, Chief Product Officer at Adobe, describes this as “collapsing the stack,” illustrating the benefits of the same person owning more of the process. When that happens, they not only generate better results, but it’s also more fun. Which leads to our next dimension of value…

Have More Fun

While writing code faster, tackling more ambitious projects, and eliminating coordination costs are fantastic benefits, vibe coding delivers another fundamental transformation that shouldn’t be underestimated: programming becomes more fun.

Traditional programming involves many tedious tasks that few developers enjoy. Fixing syntax and type checking errors, wrestling with unfamiliar package managers, writing boilerplate code, searching for documentation, and so on. Vibe coding eliminates these pain points, shifting focus from implementation details to building things.

A randomized controlled trial of GenAI coding tools found that 84% of developers reported positive changes in their daily work practices after using AI tools. They reported being more excited to code than ever before, feeling less stressed, and even enjoying writing documentation.

At Adidas, where seven hundred developers now use GitHub Copilot daily, 91% of developers reported that they wouldn’t want to work without it. Fernando Cornago, SVP of Digital Technology at Adidas, described how vibe coding resulted in developers spending 50% more time in what they called “happy time,” productive time when they were mastering their craft. This is the opposite of “annoying time,” such as struggling with brittle tests and meetings. (We cover more of this story in Part 4.)

Building cool things is addictive. Vibe coding, especially with agents, turns your keyboard into a slot machine. You “pull the lever,” and out comes a payout—a chunk of working code, a generated test, or a refactoring. Each little payout delivers a tiny dopamine hit, a neurochemical reward that makes us feel good and encourages us to pull the lever again. 

It’s fun and pulls you in. We’ve both found ourselves so thrilled and engrossed by what we’re creating that time melts away. It’s driven by that exhilarating “Let’s just do one more thing!” feeling, and the sheer fun of seeing ideas take shape. But unlike the tedious all-nighters of traditional debugging sessions, these jam sessions of pure creation. But perhaps the most powerful benefit of all is yet to come: vibe coding increases your ability to explore options and mitigate risks before committing to decisions.

Explore More Options

The fifth dimension of value that vibe coding creates may be its most profound: expanding your ability to explore multiple options before committing to decisions. In traditional development, choosing a technology stack often means making nearly irreversible commitments with limited information. These architectural decisions became what Amazon called “one-way doors”—once you walk through, turning back becomes almost impossible (or inconveniently expensive).

Vibe coding reduces the cost of exploring multiple paths in parallel. You can experience this firsthand while building a project in your preferred language. During a forty-five-minute walk with your dog, you can have a voice conversation with an AI assistant that thoroughly evaluates your options for complex libraries or frameworks. What might usually require days of research is compressed into minutes, providing detailed insights into each option’s trade-offs without writing a single line of code.

This is a capability that we never had before as programmers: The luxury of trying something five or ten different ways at once for practically free. And it extends beyond research to implementation. You can prototype the same API using three different architectural patterns in a single afternoon—say, RESTful, GraphQL, and gRPC. You can implement core endpoints using each approach, complete with serialization, error handling, and client integration. What previously might have required weeks of effort for a single implementation can now be comparatively evaluated through hands-on experience with all three options.

This concept of optionality was formalized in finance theory in the 1970s: An option is defined as the right, but not the obligation, to make a future decision. This concept is powerful in software development because software begins as pure thought—software is infinitely malleable until deployment creates real-world constraints. Every architectural choice, every library selection, every design pattern traditionally forced us to pay the full cost upfront without knowing whether we’d chosen correctly. 

The higher the uncertainty, and the higher the risk/reward ratio, the more valuable options are. If there is no uncertainty, we don’t need options—we pick the best choice, certain that our answer is correct. However, when things are highly uncertain (such as in the AI field right now), options become extremely valuable. (Another corollary: In times of high uncertainty, avoid making long-term decisions, which deprive you of options.)

Vibe coding changes the economics of software creation: Instead of betting everything on our first guess, we can place small bets across many possibilities and double down only on what works.

Toyota discovered how significant option value was decades ago in manufacturing. While American manufacturers focused on standardization and rigidity, Toyota built systems that enabled flexibility and adaptation. Their modular production lines, frequent experimentation, and rapid feedback cycles (including four thousand daily Andon cord pulls stopping production) created an option-rich system.

They could manufacture multiple model years simultaneously on the same production line, implement dozens of production changes daily, and exploit option value in many other ways that created a durable, lasting competitive advantage. Seventy years later, automakers around the world are still copying this strategy.

It is almost impossible to overstate the value that optionality creates. Over two hours, the two of us were tutored by one of the premier economics scholars, Dr. Carliss Baldwin, William L. White Professor of Business Administration, Emerita at Harvard Business School. She has written extensively about how the ability to parallelize experimentation, enabled by modularity, creates so much surplus value that it can blow companies and industries apart. 

This explains how Amazon’s microservices rearchitecture in the early 2000s (which Steve was a part of) allowed them to rapidly experiment with new business models, eventually spinning AWS into a more than $100 billion business that competitors couldn’t match because their architecture prevented exploration.

AI can drive down the cost of change, and can decrease the time and cost to explore options. That is, if you have a modular architecture that enables it. We’ll explain how to create this later in the book. Organizations that take advantage of creating option value will be orders of magnitude more competitive than those that don’t. (We explore this in more detail in Parts 3 and 4.)

AI As Your Ultimate Concierge

As a head chef running a world-class restaurant, you will run into many problems that aren’t strictly culinary. As it happens, however, your sous chef is also a sommelier, detective, accountant, rat catcher, master plumber, award-winning author, and tax planner. Remarkably, it is also a surgeon, taxidermist, and a lawyer. We think of AI as a concierge who is available to you 24×7, literally on a moment’s notice, happy to take a phone call with any of your questions or whims.

Your AI collaborator is more than a code generator. It can help you with your toughest problems. Sometimes, it’s your personal detective that you send to root through labyrinthine Git histories. You only need say, “I lost some test files somewhere between commit 200 and commit 100,” and not only will it find it (“Found it, it was 43 commits back.”) but it will track them down and stitch them back into your code (“I extracted out the tests, also the build configuration that refers to them.”)

We’ve handed AI enormous, nested structure dumps and said, “Find that one little detail buried ten layers deep,” and it came back in seconds with: (“It’s [‘server’][‘cluster’][‘node_13’][‘overrides’][‘sandbox’][‘temporary’]”). 

We also love using AI as a design partner—a quick collaborator who’s awake at any hour you’re inspired to work. It’s the extra pair of hands that can validate your ideas or debug that sneaky performance glitch you’ve been chasing for days. 

We’ve mentioned a few of the many kinds of messes that AIs can produce—or more accurately, messes that you produce using AI. It turns out your AI concierge is great for helping you get out of those messes as well, as long as you use the disciplined approach of only tackling small tasks at a time and tracking your progress carefully (which we cover in a future chapter).

Conclusion

We’ve seen how vibe coding rapidly accelerates your workflow, turning multi-day chores into lunchtime wins—like Gene and Steve hacking together the video excerpt tool in less time than it takes to cook a decent chili. Sure, sometimes your AI sous chefs misinterpret recipes (looking at you, captioning nightmare with ffmpeg), and you’ll occasionally need to step in yourself, but the net result is still far quicker than manual coding.

However, as we showed you, speed is the least interesting part. Vibe coding creates value along five distinct dimensions or FAAFO: fast, ambitious, autonomous, fun, and optionality.

• Fast feedback loops and high velocity make more projects feasible: AI’s speed enables all the other dimensions of FAAFO.
• Ambition reshapes your project landscape: “Not quite worth it” tasks become quick wins, and impossible dreams land on your to-do list.
• Autonomy eliminates friction: Work at your own pace without constant negotiation, handoffs, and the coordination costs that slow traditional teams.
• Fun drives engagement: Programming becomes addictive again when you’re building rather than debugging, creating rather than wrestling with syntax.
• Options create competitive advantage: Explore multiple approaches in parallel, turning one-way doors into reversible experiments.

In the next chapter, we’ll show some of the risks of vibe coding, and what you can do to mitigate them.

Stay tuned for more exclusive excerpts from the upcoming book Vibe Coding: Building Production-Grade Software With GenAI, Chat, Agents, and Beyond by Gene Kim and Steve Yegge on this blog or by signing up for the IT Revolution newsletter.

- About The Authors

Gene Kim

Gene Kim has been studying high-performing technology organizations since 1999. He was the founder and CTO of Tripwire, Inc., an enterprise security software company, where he served for 13 years. His books have sold over 1 million copies—he is the WSJ bestselling author of Wiring the Winning Organization, The Unicorn Project, and co-author of The Phoenix Project, The DevOps Handbook, and the Shingo Publication Award-winning Accelerate. Since 2014, he has been the organizer of DevOps Enterprise Summit (now Enterprise Technology Leadership Summit), studying the technology transformations of large, complex organizations.

Follow Gene on Social Media

• 
• 

Steve Yegge

Steve Yegge is an American computer programmer and blogger known for writing about programming languages, productivity, and software culture for two decades. He has spent over thirty years in the industry, split evenly between dev and leadership roles, including nineteen years combined at Google and Amazon. Steve has written over a million lines of production code in a dozen languages, has helped build and launch many large production systems at big tech companies, has led multiple teams of up to 150 people, and has spent much of his career relentlessly focused on making himself and other developers faster and better. He is currently an Engineer at Sourcegraph working on AI coding assistants.

Follow Steve on Social Media

•