[{"content":"Genesis 2:3 says: “Then God blessed the seventh day and made it holy, because on it he rested from all his work of creation.”\nOver the first six days, we have traveled a long road:\nWe saw through the sugar coating of natural language , and recognized the review tax that overloads the brain ;\nWe paused amid the Brownian motion of code , and recognized the ghost of round-trip engineering from thirty years ago ;\nAfter the illusion of the Happy Path shattered, we saw more clearly the essential difficulties and new hope in the real world .\nOn the seventh day, when we took our hands away from the blind clatter of the keyboard, our eyes began to see a new world.\nThe Renaissance in fifteenth-century Europe broke through the predicament facing medieval craftsmen. As AWS CTO Werner Vogels has pointed out , AI does not bring a wave of unemployment, but a “Renaissance Developer ” to the software world.\nThe “code farmers” whose sole job was translating requirements into code are leaving the stage. Next will come generalist creators who combine technical depth, business insight, and humanistic aesthetics.\nBut this does not mean we can throw away the fundamentals.\nLeonardo was Leonardo because he had dissected real muscles and bones before putting pen to paper; future engineers will become generalist creators because, before directing AI, they will already understand system constraints, scalability, and information security inside and out.\nWriting code may gradually become cheap “manual labor,” while “thinking” will transform into another form.\nThe seventh day is about to pass, and tomorrow will be a brand-new Monday. Do not treat these seven days of discussion as an endpoint; they may not even qualify as a prelude.\nOur exploration and construction are only now beginning to dawn.\n","permalink":"https://drakeguan.org/en/blog/2026/09/the-renaissance-developer/","summary":"\u003cp\u003eGenesis 2:3 says: “Then God blessed the seventh day and made it holy, because on it he rested from all his work of creation.”\u003c/p\u003e\n\u003cp\u003eOver the first six days, we have traveled a long road:\u003c/p\u003e\n\u003cp\u003eWe saw through \u003ca href=\"/en/blog/2026/08/the-english-programming-illusion/\"\u003ethe sugar coating of natural language\u003c/a\u003e\n, and recognized \u003ca href=\"/en/blog/2026/09/the-code-review-tax/\"\u003ethe review tax that overloads the brain\u003c/a\u003e\n;\u003c/p\u003e\n\u003cp\u003eWe paused amid \u003ca href=\"/en/blog/2026/09/code-churn-as-brownian-motion/\"\u003ethe Brownian motion of code\u003c/a\u003e\n, and recognized \u003ca href=\"/en/blog/2026/09/the-round-trip-curse/\"\u003ethe ghost of round-trip engineering from thirty years ago\u003c/a\u003e\n;\u003c/p\u003e","title":"Day Seven: The Dawn of the Renaissance Developer"},{"content":"If, after reading the first five posts, you think this is a farewell ceremony for software engineering, with engineers about to lose their jobs , think again.\nWhenever revolutionary automation tools have appeared throughout history, prophets and the media have always been unable to resist saying, “Such-and-such is going to disappear; such-and-such is going to become unemployed.” But seventy years have passed, and these predictions have almost never come true.\nWhen Fortran appeared, the priests of assembly language were told they were finished; when VisiCalc, the first electronic spreadsheet in history, was born , Wall Street declared that accountants were no longer needed.\nWhat happened? Demand for accountants multiplied several times over. They simply stopped spending all day adding up ledgers by hand and became financial analysts who helped steer businesses.\nEconomics has a famous concept called “Jevons paradox ”: when the efficiency of using a resource surges and the cost of production plummets, total consumption does not shrink. Instead, it expands wildly by orders of magnitude.\nThe same is true of AI and code.\nAs AI drives down the marginal cost of generating code, humanity’s ambition to build complex software systems will also expand at ten times the speed—and go beyond what came before.\nSo the crisis has never been that the tools are too powerful. It is that we mistake “incidental implementation” for “the essence of engineering.”\nIn his landmark 1987 essay “No Silver Bullet ,” Turing Award winner Fred Brooks made this point long ago: AI eliminates only the “accidental difficulties” of looking up APIs, typing parentheses, and formatting. But the “essential difficulties” of dealing with real-world exceptions, security, surprises, and state will always remain, waiting to be solved.\nDefining a system’s defensive boundaries, balancing technical debt against performance, and imposing order on chaotic business requirements… these essential complexities have always been the exclusive domain of human thinkers.\nDo not listen to the marketing pitch that “everyone is an engineer,” and do not fear the apocalyptic funeral bell proclaiming that “engineers are dead.” Do not let marketing rhetoric stoke your anxiety, but neither should you turn a blind eye to technological progress.\nThe priests never disappeared. They simply changed names.\nThe next golden age of software engineering, belonging to true thinkers, is only just beginning… ","permalink":"https://drakeguan.org/en/blog/2026/09/the-elevated-priesthood/","summary":"\u003cp\u003eIf, after reading the first five posts, you think this is \u003ca href=\"/en/blog/2026/09/the-default-insecurity/\"\u003ea farewell ceremony for software engineering, with engineers about to lose their jobs\u003c/a\u003e\n, think again.\u003c/p\u003e\n\u003cp\u003eWhenever revolutionary automation tools have appeared throughout history, prophets and the media have always been unable to resist saying, “Such-and-such is going to disappear; such-and-such is going to become unemployed.” But seventy years have passed, and these predictions have almost never come true.\u003c/p\u003e\n\u003cp\u003eWhen Fortran appeared, the priests of assembly language were told they were finished; \u003ca href=\"https://www.npr.org/sections/money/2017/05/17/528807590/episode-606-spreadsheets\"\u003ewhen VisiCalc, the first electronic spreadsheet in history, was born\u003c/a\u003e\n, Wall Street declared that accountants were no longer needed.\u003c/p\u003e","title":"Day Six: The Priests Never Disappeared; They Just Changed Names"},{"content":"A broken system is not frightening; you can fix it. More frightening is a system that starts up and runs smoothly, while everyone in the company knows only the rough outline and nobody knows exactly how it came to work.\nThe words “it works ” are often one of the most devious disguises worn by a cybersecurity disaster.\nOver the past year or so, real incidents have taught us a harsh lesson about “default insecurity.”\nFrom a popular vibe-coding platform that left out RLS security rules (CVE-2025-48757) , leaving hundreds of applications exposed, to a dating app that configured Firebase for public reads, allowing tens of thousands of identity documents to be downloaded directly ; and even an autonomous agent that, while optimizing automatically, formatted the production database .\nMaking a system “work” merely means taking a stroll down the calm “Happy Path ,” which probably accounts for only 10% of the effort in software development. The remaining 90% of genuine expertise lies in reasoning through every kind of situation: What happens when the network goes down? How do you prevent an outage when traffic surges? When a malicious attacker enters strange characters, how do you stop the system from handing over the entire database?\nTyping has become cheaper and code generation has become faster, but the cost of security and defense has not decreased at all.\nTools have not eliminated software engineering.\nThey have simply pushed the small group of “modern priests” who truly understand architecture, performance, defense, and edge cases onto an even higher altar. ","permalink":"https://drakeguan.org/en/blog/2026/09/the-default-insecurity/","summary":"\u003cp\u003eA broken system is not frightening; you can fix it. More frightening is a system that starts up and runs smoothly, while everyone in the company knows only the rough outline and nobody knows exactly how it came to work.\u003c/p\u003e\n\u003cp\u003eThe words “\u003ca href=\"/en/blog/2026/09/the-round-trip-curse/\"\u003eit works\u003c/a\u003e\n” are often one of the most devious disguises worn by a cybersecurity disaster.\u003c/p\u003e\n\u003cp\u003eOver the past year or so, real incidents have taught us a harsh lesson about “default insecurity.”\u003c/p\u003e","title":"Day Five: The Three Most Dangerous Words: “It Works”"},{"content":"Vibe coding delivers enormous productivity , and it feels great. But what makes it most maddening is not that AI cannot write, or writes badly.\nIt is that you personally fixed a fiendish bug last Friday night. On Monday morning, in a good mood, you ask AI to casually add a button—and it “helpfully” wipes out your handwritten code wholesale, as if it were an obsolete draft.\nThat feeling of being stabbed in the back by a teammate is not unique to AI.\nEngineers experienced it thirty years ago.\nIn the 1980s and 1990s, the giants of “computer-aided software engineering (CASE),” led by KnowledgeWare , sold the exact same beautiful dream to the world: analysts would only need to draw diagrams on a screen, and the system would automatically generate code. At the time, everyone thought programmers were genuinely about to lose their jobs.\nThat billion-dollar bubble ultimately died from an unsolvable problem: round-trip engineering .\nMachine-generated code looked beautiful, but as soon as a human engineer entered the real world and manually changed a few lines of boundary conditions or business requirements in the source code, problems began to emerge. There is a difficult-to-handle “loss of semantics” between natural language (or an architectural diagram) and the underlying implementation code. AI has no idea why you manually changed those two lines; from its probabilistic perspective, your hard-won work may be nothing more than disposable noise.\nWhen manual intervention destroys the perfect automated loop, we are forced into a devil’s bargain: do you spend three times the effort fighting AI for control? Or do you simply stop resisting and hand everything over to the black box?\nSo more and more people choose compromise, and even learn to console themselves:\n“It’s okay if I don’t understand the underlying code. It works now.”\nBut every engineer who has lived through a P0 knows in their heart:\nThose three words—“it works”—are often the sweetest, and most dangerous, beginning of every disaster… ","permalink":"https://drakeguan.org/en/blog/2026/09/the-round-trip-curse/","summary":"\u003cp\u003e\u003ca href=\"/en/blog/2026/09/code-churn-as-brownian-motion/\"\u003eVibe coding delivers enormous productivity\u003c/a\u003e\n, and it feels great. But what makes it most maddening is not that AI cannot write, or writes badly.\u003c/p\u003e\n\u003cp\u003eIt is that you personally fixed a fiendish bug last Friday night. On Monday morning, in a good mood, you ask AI to casually add a button—and it “helpfully” wipes out your handwritten code wholesale, as if it were an obsolete draft.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThat feeling of being stabbed in the back by a teammate is not unique to AI.\u003c/strong\u003e\u003c/p\u003e","title":"Day Four: The Ghost of “Round-Trip Engineering” Returns"},{"content":"Running on a treadmill at the gym at least burns a few calories; but sprinting on an AI coding treadmill is more likely to leave you exhausted, uncertain, and buried in technical debt.\n“I wrote two thousand lines of code with AI today!” may be the most addictive temptation facing developers right now.\nAfter analyzing more than 200 million lines of code , the software analytics team GitClear found that the real thing exploding after the spread of AI was not quality refactoring, but “code churn.” The proportion of freshly written code that is repeatedly overturned, modified, or simply deleted within two weeks has reached a historic high.\nIn physics, Brownian motion describes pollen particles being violently battered by water molecules from every direction, producing intense, random movement. But if you extend the time frame and calculate the vector, the probability of moving in any direction is equal, so the “average displacement” is actually zero.\nThe Turing Award-winning computer science pioneer Edsger W. Dijkstra offered a similar warning in a famous 1988 memorandum :\n“If we wish to count lines of code, we should not regard them as ‘produced assets’ but as ‘spent costs.’”\nAI really does let us produce (squander) code at ten times the speed, but unfortunately, it has not bought you ten times the distance forward.\nAs we gradually see this frantic running in place for what it is…\nWhy does AI’s coding loop start producing bugs—or even lose control—whenever I step in and change just a few lines of code? ","permalink":"https://drakeguan.org/en/blog/2026/09/code-churn-as-brownian-motion/","summary":"\u003cp\u003eRunning on a treadmill at the gym at least burns a few calories; but sprinting on an \u003ca href=\"/en/blog/2026/09/the-code-review-tax/\"\u003eAI coding treadmill\u003c/a\u003e\n is more likely to leave you exhausted, uncertain, and buried in technical debt.\u003c/p\u003e\n\u003cp\u003e“I wrote two thousand lines of code with AI today!” may be the most addictive temptation facing developers right now.\u003c/p\u003e\n\u003cp\u003eAfter analyzing \u003ca href=\"https://www.gitclear.com/ai_assistant_code_quality_2025_research\"\u003emore than 200 million lines of code\u003c/a\u003e\n, the software analytics team \u003ca href=\"https://www.gitclear.com/ai_assistant_code_quality_2025_research\"\u003eGitClear\u003c/a\u003e\n found that the real thing exploding after the spread of AI was not quality refactoring, but “code churn.” The proportion of freshly written code that is repeatedly overturned, modified, or simply deleted within two weeks has reached a historic high.\u003c/p\u003e","title":"Day Three: The 10x Engineer on a Treadmill"},{"content":"What was the bill mentioned in the previous post ? Who was it sent to? Who pays it?\nThe answer may be a little ironic:\nThe bill was sent directly to our eyes and brains.\nThe software engineering world has held a consensus for decades: typing was never the bottleneck.\nPython became one of the world’s most popular languages in part because its creator, Guido van Rossum, designed it around the idea that “code is read much more often than it is written .”\nUncle Bob gave us an even more striking ratio in his book Clean Code: engineers spend ten times as long reading old code as writing new code .\nThe miracle of AI is that it reduces the 10% cost of typing to almost nothing. But the trap of AI is that it dumps the remaining 90% of cognitive capacity on you, until you crash or give up.\nFor the past two or three decades, the entire software world has been trying every possible way to reduce the “burden of reading and understanding.” Now AI has reversed that direction.\nThis is not productivity. It is a “review tax.”\nWhen we spend huge amounts of time every day racing through the endless cycle of reading, debugging, and telling AI to rewrite things, are we really any closer to our destination?\nOr are we simply running, increasingly out of breath, on a treadmill that keeps getting faster? ","permalink":"https://drakeguan.org/en/blog/2026/09/the-code-review-tax/","summary":"\u003cp\u003eWhat was the bill mentioned in the \u003c!-- WARNING: Missing reference to 08/2026-08-31.md --\u003e\u003ca href=\"/08/2026-08-31\"\u003eprevious post\u003c/a\u003e\n? Who was it sent to? Who pays it?\u003c/p\u003e\n\u003cp\u003eThe answer may be a little ironic:\u003c/p\u003e\n\u003cp\u003eThe bill was sent directly to our eyes and brains.\u003c/p\u003e\n\u003cp\u003eThe software engineering world has held a consensus for decades: typing was never the bottleneck.\u003c/p\u003e\n\u003cp\u003ePython became one of the world’s most popular languages in part because its creator, Guido van Rossum, designed it around the idea that “\u003ca href=\"https://peps.python.org/pep-0008/\"\u003ecode is read much more often than it is written\u003c/a\u003e\n.”\u003c/p\u003e","title":"Day Two: Typing Was Never the Bottleneck; Reading Is"},{"content":"In February 2025, Andrej Karpathy, a founding member of OpenAI and former director of artificial intelligence at Tesla, made a memorable remark :\n“The hottest new programming language is English.”\nIt sounds a lot like a new myth Silicon Valley has just invented.\nBut if “giving computers instructions in human language” could eliminate engineers and software engineering, then the discipline should have been declared finished back in 1959. That was the year Grace Hopper championed COBOL . The promise she made to the business world was remarkably similar to today’s vibe coding:\n“Managers would only need to write sentences in everyday English, and the system would run automatically.”\nSix decades later, we still need engineers. Because programming languages and syntax were never an imposing wall blocking development; they were more like a safety net. Removing parentheses, types, and syntax does not make the complexity disappear along with them. You simply have to put the logic back into natural language using more—and more cumbersome—adjectives and clauses.\nComputers can finally understand English and Chinese, but the cost of thinking has not been discounted as a result.\nNatural language makes programming easier, but it does not make thinking easier.\nThe parentheses are gone, but not the bill. ","permalink":"https://drakeguan.org/en/blog/2026/08/the-english-programming-illusion/","summary":"\u003cp\u003eIn February 2025, Andrej Karpathy, a founding member of OpenAI and former director of artificial intelligence at Tesla, made a \u003ca href=\"https://x.com/karpathy/status/1886192184808149383\"\u003ememorable remark\u003c/a\u003e\n:\u003c/p\u003e\n\u003cblockquote\u003e\n\u003cp\u003e“The hottest new programming language is English.”\u003c/p\u003e\n\u003c/blockquote\u003e\n\u003cp\u003eIt sounds a lot like a new myth Silicon Valley has just invented.\u003c/p\u003e\n\u003cp\u003eBut if “giving computers instructions in human language” could eliminate engineers and software engineering, then the discipline should have been declared finished back in 1959. That was the year \u003ca href=\"https://www.wikiwand.com/en/Grace_Hopper\"\u003eGrace Hopper\u003c/a\u003e\n championed \u003ca href=\"https://www.wikiwand.com/zh-hant/COBOL\"\u003eCOBOL\u003c/a\u003e\n. The \u003ca href=\"https://blakecrosley.com/zh-Hant/blog/engineering-philosophy-grace-hopper\" title=\"Engineering Philosophy: Grace Hopper, Teaching Computers to Speak Human Language\"\u003epromise\u003c/a\u003e\n she made to the business world was remarkably similar to today’s vibe coding:\u003c/p\u003e","title":"Day One: “Programming in English” Is a 70-Year-Old Promise"},{"content":"This is Shuen-Huei Guan, or just Drake.\n","permalink":"https://drakeguan.org/en/about/","summary":"\u003cp\u003eThis is Shuen-Huei Guan, or just Drake.\u003c/p\u003e\n\u003cp\u003e\u003cimg alt=\"gravatr\" loading=\"lazy\" src=\"https://www.gravatar.com/avatar/f9005f9e68699af4039f6cb718f932dc?s=400\\\"\"\u003e\u003c/p\u003e\n\u003cp\u003eSomething about computer graphics, computer animation, renderman,\npython programming, and random thoughts,\nthose are what you can get here.\nAnd surprisingly, I keep moving forward to some new fields,\nlike video-on-demand streaming, adaptive streaming, content platform.\nThey are related to cultivating our mind at the same time\nentertaining our life~\u003c/p\u003e\n\u003cp\u003eFor further info, refer to \u003ca href=\"http://about.me/drakeguan\"\u003ehttp://about.me/drakeguan\u003c/a\u003e\n\u003c/p\u003e\n\u003cp\u003eCheer~\u003c/p\u003e\n\u003cscript type=\"application/ld+json\"\u003e\n{\n  \"@context\" : \"http://schema.org\",\n  \"@type\" : \"Person\",\n  \"name\" : \"Shuen-Huei Guan\",\n  \"url\" : \"https://drakeguan.org\",\n  \"sameAs\" : [\n    \"https://about.me/drakeguan\",\n    \"https://www.linkedin.com/in/drakeguan\",\n    \"https://github.com/drakeguan/\",\n    \"https://twitter.com/drakeguan\",\n    \"http://stackoverflow.com/users/261718/drake?tab=profile\"\n  ]\n}\n\u003c/script\u003e","title":"About Me"}]