Main Source: Vibe Coding: Building Production-Grade Software With GenAI, Chat, Agents, and Beyond

The Dark Side

While electricity’s tremendous potential was obvious, it wasn’t until twenty years after its invention that factory owners learned to abandon their linear, belt-driven layouts in favor of designs that exploited electric power’s flexibility. (Kim and Yegge 35)

Human institutions are the slowest technology to adapt.

“Science progresses one funeral at a time” - Paul A. Samuelson Source

Org Chart

The New York and Erie Railroad Company that was first designed in 1855 by the railroad’s superintendent, Daniel McCallum. Source

DORA’s 2024 report dropped a surprising finding: Every 25% increase in GenAI adoption correlates with 7% worse stability (more outages and longer recovery times) and a 1.5% slowdown in throughput (deployment frequency and lead times). (Kim and Yegge 39)

The good news is that the same core principles and practices that allow us to deliver software sooner, safer, and happier as we went from one software deployment per year (which was typical in the 2000s) to 136,000 deployments per day (which Amazon achieved in 2015) can be scaled up as we go from generating a hundred lines of code a day to thousands and beyond. (Kim and Yegge 40)

For my part, I see this chart and the above stat as an interlocked issue. The velocity that agentic coding unlocks means organizations must reshape themselves. No longer will top-down decisions work, where each department is focused on its own feature with reckless disregard for the whole. Spheres of operation and cross-departmental cooperation are the new game. In the short term, this means that small startups will gain an advantage, as they can share one mind faster. See The Mythical Man-Month.

Rather than being discouraged by current limitations, successful practitioners adapt their approach to maximize AI’s present capabilities while preparing for its rapid evolution:

  1. Delegate thoughtfully
  2. Supervise appropriately
  3. Establish guardrails
  4. Check work regularly
  5. Create persistent references (Kim and Yegge 40)

If you’re nodding along as you read this last quote, buy the book! I can’t come close to capturing all the wisdom and insights contained within, nor would I want to steal Kim and Yegge’s duly earned credit for contributing to the industry’s knowledge base in such a remarkable way.

AI is Changing all Knowledge Work

AI is a force reshaping how work gets done, and who is doing that work. It’s reconfiguring the jobs themselves, as well as the skills that matter. (Kim and Yegge 43)

A decentralized autonomous organization (DAO), sometimes called a decentralized autonomous corporation (DAC) Source

The entire workforce will be changed by the AI revolution. This is not just an update to the way businesses organize their data; there will be a complete reshaping of the types of roles needed from the front line to the board.

A groundbreaking 2023 study by Dr. Daniel Rock and his colleagues, colloquially called the “OpenAI Jobs Report,” delivered some shocking news: Researchers estimated that 80% of US workers could see AI impact at least 10% of their tasks, potentially more. (Kim and Yegge 44)

This study was groundbreaking in 2023, but the idea that 80% of the workforce may see 10% of their tasks taken over by AI systems now feels like an underestimate, given current technological capabilities. The major bottleneck now is not technical capability, but organizational structure and the time requirements for the human engineers partnering with their AI workforce to build out the future of automated tasks.

Dr. Andrew Ng, one of the founders of Google Brain and now at Stanford University, said, “AI won’t replace people, but maybe people [who] use AI will replace people [who] don’t.” (Kim and Yegge 44)

Beyond the Junior Developer Debate

For decades, we’ve organized software engineering teams in the same way: Senior principal engineers design project architecture, mid-level engineers build complex features, and junior developers learn by handling small, contained tasks. (Kim and Yegge 45)

I can’t speak to the full picture of this process, but for my part, I went from junior to full engineer to lead in under 12 months. It took 70+ hour weeks since February 2025, but it’s doable with the right mindset and by mastering the techniques described in detail in this book.

Every single time we’ve had a significant leap in programming technology, people predict the programmer apocalypse:

“High-level languages will kill assembly programmers!” “Visual Basic will replace professional developers!” “Low-code platforms will make developers obsolete!” “No-code tools mean the end of software engineering!”

However, each time programming got easier, we needed more programmers. (Kim and Yegge 47)

Could AI Lead to Annual 100% Global GDP Growth?

Some economists and AI researchers are making a bold, almost ludicrous claim: that AGI could eventually double global GDP every year. We’re talking about a 100% annual growth rate when the global economy has been puttering along at 2–3% for nearly a century.

Let’s put this into perspective: Before the Industrial Revolution, economic growth barely existed. We had roughly 0.01% annual growth for thousands of years. Then the Industrial Revolution arrived, and growth jumped to 1–2%. That 100–200x increase completely transformed human existence. (Kim and Yegge 49)

AI elevates your ideas, your ambitions. It becomes an amplifier for your creativity. (Kim and Yegge 51)

When put into perspective as a 100x increase in productivity, the idea of 100% GDP growth seems conservative. For the same reason that big tech has compounded GDP growth within the industries it targets, so too will the new AI revolution compound growth across all industries; assuming robotics keeps expanding, the caveat of “except where moving atoms” will shift from relevant to moot.

What Skills to Learn

Because tools will evolve rapidly, core traditional software engineering principles will play at least as large a role, if not larger. Thus, it’s essential to:

  • Create fast and frequent feedback loops for validation and control.
  • Create modularity to reduce complexity, enable parallel work, and explore options.
  • Embrace learning in a world where everything changes fast.
  • Master your craft to thrive in an environment where all knowledge work will be changing in a short timeframe. (Kim and Yegge 61)

In short:

an attitude of openness, eagerness, and lack of preconceptions when studying, even at an advanced level, just as a beginner would. Source

We are all learning in public, and must continue to do so if we want to keep accelerating the adoption rate. This revolution will happen in public and be documented in detail via open source repositories and peer-reviewed papers. See the new engram breakthrough source by DeepSeek, built upon mHC source2.

To achieve FAAFO, you must have the skills and processes to build trust in what your AI collaborator creates. Trust us first: Going fast without feedback is dangerous. (Kim and Yegge 63)

Creating Modularity

Dr. Dan Sturtevant and his colleagues did research that showed how developers working in tangled, non-modular systems are 9x more likely to quit or be fired. And again, The State of DevOps Report showed that a modular architecture was also a top predictor of performance. (Kim and Yegge 63)

To me, this makes sense, as it allows for maximum velocity in development, though it comes with a tradeoff: each shipping team develops deep mastery in its domain, but it becomes harder for someone higher up to maintain the full picture of the entire system. But given the speed at which entire systems are being modified or rebuilt under the AI revolution, maybe that’s no longer a tradeoff worth worrying about.

Alexander Embiricos from the ChatGPT Codex team described how an engineer using AI tools achieved excellent “commit velocity” building a new system from scratch. But when they ported it “into the monolith that is the overall ChatGPT code base that has seen ridiculous hypergrowth” (that is, a system with architectural problems) the results changed dramatically. Despite having the “same engineers, same tooling,” their “commit rate just plummets.” This real-world example shows that even at OpenAI, architectural constraints affect developers using AI too. (Kim and Yegge 63)

Embrace Learning

Learning means doing. It means tackling problems that seem insurmountable. It means taking risks, patiently wading through your mistakes, pushing until you get the outcomes you want, and troubleshooting creatively when things go wrong. (Kim and Yegge 63)

The goal is not to avoid being wrong or to prevent failure. The goal is to ship real systems and learn from feedback from reality.

On May 2, 2014, Zuckerberg announced that the company would be changing its internal motto from “Move fast and break things” to “Move fast with stable infrastructure”. source

There is a balance to be struck here, but any organization that refuses to move for fear of breaking things had better be serving a true societal need, like banking still running on COBOL systems.

On the other hand, the skills crisis is not only about the ability to program in COBOL but also, and more importantly, about the knowledge of legacy applications originally developed decades ago to which layers of changes have been added and whose documentation has been lost. Source

Learning Mindset Shift

Cultivating a learning mindset has nothing to do with innate genius. Learning is about deliberate and intentional practice, much like Dr. Anders Ericsson described for mastering any complex skill.

Expert coaching

The revolution is moving at such velocity that no one person can keep up. And while it is true that being able to self-learn is a required skill to thrive in software engineering, now is not the time to lean on that skill alone. Find those who are ahead of you, and engage them. If for no other reason, they will be able to vet the right learning materials and save you hours or days of wandering down the wrong rabbit holes.

Fast feedback

Take that feedback from your expert and put it into practice. Start small. The goal is to find the shortest feedback cycles possible. I’m talking 30 minutes or less. Then build from there. Each time you do a task, you should be able to find a way to improve by a substantial amount. The time you save by leveraging AI should be fed back into learning, either by getting better at using the tools or by moving up the tech stack and reflecting on how you can solve larger problems.

Intentional practice

You must build. You must find personal projects and ship them. That doesn’t mean you have to build a SaaS product or create an open source project you’ll maintain for others. But you must build and ship at least one tool you’ll use. Otherwise, you fail to achieve the feedback loop described above.

Challenging tasks

Once you can use the tools, you must start building projects that are beyond you, with the goal of shipping. Even projects that seem impossible. If you already know the entire tech stack and systems you want to build for a project, pick a more ambitious one.

In short, achieving FAAFO becomes an exercise in “being a great learner.” (Kim and Yegge 67)