In the documentary Jiro Dreams of Sushi, there is an instructive detail: for their first few months, incoming apprentices must wring scalding-hot towels, often until their hands turn red and blister. Under strict kitchen rules, beginners are not allowed anywhere near knives or raw fish. Only after long stretches of basic chores are they permitted to handle fish and rice. And the final hurdle before truly graduating to work independently is making tamagoyaki (egg omelet) to standard. It is said that some apprentices tried around two hundred times before finally earning the master’s approval.
Traditional craft had no shortcuts. If you wanted to weed out unreliable candidates, a simple method was to make them repeat the exact same task over and over—until hands grew faster than thoughts and the judgment of quality was drilled into muscle memory. Simply learning to cook a good pot of rice often took apprentices years.
But what if someone brings a smart rice cooker into the restaurant?
Rinse the rice, add water, press the button, and the rice cooker serves up a pot of evenly heated rice every single time. No need to wake up at dawn to watch over the pot, and no risk of ruining the rice because someone was in a bad mood today. What once demanded the apprentice’s deepest dedication—the most critical fundamental—has become the cheapest thing in the kitchen.
Of course, a rice cooker cannot produce a piece of nigiri ready to serve to a customer. It cannot pick the finest catch at the dawn fish market, nor can it adjust the density of a rice ball based on the day’s temperature and humidity. But what it shatters is the very foundation of the system: it flattens the barrier to cooking rice. Once cooking rice becomes effortless, the apprentice ladder and testing rituals that were built on enduring years of basic labor to master rice suddenly lose their footing.
The shock that generative AI has brought to the tech industry today is fundamentally the same.
Many people’s initial reaction was panic, fearing that AI was stealing their livelihood or that software engineers would soon face mass unemployment. But the reality is far more nuanced: AI has not simply carried away entire professions; it has consumed the standardized execution layers within them (a Brookings study quantified which tasks are most vulnerable, and a peer-reviewed study explained why it is precisely these tasks). Just as a rice cooker only cooks white rice, AI takes over repetitive, mechanical steps, not the entire craft.
The problem is that schools and hiring interviews essentially rely on the exact same sieve, with only one filter hole: whether you know how to cook rice. And right now, this sieve sits squarely on the layer most vulnerable to being automated away. Hiring on one side, education on the other—both have run into trouble.
Corporate hiring broke down first. In the past, technical interviews relied heavily on algorithmic problem-solving and writing boilerplate code because it was easy to grade. Now that everyone can casually use AI, this testing format no longer holds up. The Pragmatic Engineer surveyed interviewers at major tech firms and found that 81% suspected candidates were cheating with AI during interviews. When solutions can be generated in seconds, an interviewer on the other side of a screen has a hard time discerning whether a candidate understands foundational principles or is merely skilled at prompting tools.
Once the standard failed, a chain reaction quickly followed on the hiring side. Developer community Stack Overflow cited tracking data from Final Round AI showing that hiring for entry-level tech roles dropped roughly one-fourth (25%) year-over-year. Companies still need junior talent, but the old test papers can no longer distinguish who is genuinely capable (in a survey of employers, about 60% of employers said they had no intention of directly replacing entry-level roles, while 40% were exploring how to augment them with AI).
School education broke down right after. Schools today remain largely assembly lines for standardized knowledge—teaching syntax and common algorithms during term time, and testing the memorization of code snippets in timed finals. At their core, they are teaching students how to cook a standard pot of white rice. Once students can generate well-structured capstone projects in seconds using AI, the discriminatory power of traditional closed-book exams and homework naturally collapses.
Both sides have failed. But the root cause is not what AI took away; it is that this screening mechanism has only ever tested whether you know how to cook rice. Once standardized execution becomes worthless, any grading mechanism relying on mechanical reproduction is bound to fail sooner or later.
When the sieve fails, people tend to panic, largely because many subconsciously assume skill development is a smooth, linear progression: spend a few years writing code, and a junior engineer will naturally mature into a senior expert.
Yet reality is rarely that simple.
From a company’s perspective, fresh graduates often have solid foundations, but carry too much academic mindset and can only handle lower-level positions. It typically takes them three to four years of practical work to understand where senior engineers stand and reach the point where they are ready for promotion.
I have come to realize more and more that progressing from junior to senior is not about doubling or tripling your efficiency at junior-level tasks. If junior engineering work is like cooking rice, a senior engineer is not someone who can cook three pots of rice in one meal. In technology, every step up the ladder almost completely rewrites the job description—it is like taking an entirely new course, or even switching to a different career track altogether, rather than merely executing existing skills faster.
Across the industry, many practitioners emphasize the same thing: the difference between senior and junior is not how fast you write code, but whether you understand business requirements, whether you can foresee potential system failures, whether you can design a sound architecture, and whether you have the courage to take ownership when things go wrong. Promotions and raises do not automatically arrive just because you have accumulated years on the job.
In a well-known empirical study, economist David Deming found that as machines continually replace routine labor, labor market returns to technical execution are actually slowing down, while the rewards for social skills like communication and coordination are growing ever larger. An analysis of earnings calls by Columbia Business School corroborates this: the higher the leadership level, the more vital communication and coordination skills become.
Once you understand this, it is easy to see why traditional education has long harbored a misconception: schools teach students to cook rice not because cooking rice determines their future ceiling, but because cooking rice is the easiest to test, the easiest to grade, and the most effective for getting students past the first round of campus recruiting. Now that AI has shattered this protective umbrella, we are forced to confront an uncomfortable truth: the things that truly matter later on were never seriously taught in school to begin with.
Faced with these changes, it is easy to swing to the opposite extreme. Some claim: “If AI can write code now, do we even need to practice the basics? Shouldn’t we just jump straight to soft skills?”
This actually underestimates the value of tools. A rice cooker eliminates the physical drudgery of watching the fire and controlling the temperature, but it cannot replace your palate for judging good rice. If you cannot even taste whether a bowl of rice is undercooked or mushy, you will have no idea whether what you serve is actually any good. Fundamentals remain essential; tools simply free humans from mechanical repetition so we can finally focus our energy on what matters more.
Everyone has a rice cooker; what truly sets people apart are the five capabilities a rice cooker cannot replicate:
To bring these five capabilities to fruition, both schools and individuals need to adapt their approach.
For schools, the legacy grading and testing system needs to be reshaped. Institutions should increase the weight given to oral exams, project portfolios, and hands-on practice, while integrating the effective use of AI into the curriculum. A committee of experts at MIT pointed out in their official report on AI and education that schools should redesign evaluation mechanisms and advised against relying on AI-detection software. Such detectors often trigger a cat-and-mouse game between students and teachers, carrying a high risk of false accusations. Real assessment should instead center on students explaining their thought processes face-to-face to instructors and defending their own design choices.
For individuals, the most urgent priority is to stop memorizing fragile technical jargon and instead settle down to cultivate genuine engineering judgment. When I talked with students during my university years, I often emphasized this: technical buzzwords might earn you a ticket to an interview, but how far you can go and whether you can advance depends entirely on your judgment. AI has shortened the shelf life of buzzwords, pushing the demand for sound judgment much earlier into one’s career. New concepts pop up every day; a great way to see through them is to distinguish which are the timeless, real problems of engineering and business, and which are mere stopgaps patching current models that are doomed to obsolescence. Focus your energy on the former, and your footing will remain firm.
Yet seeing the road ahead clearly does not mean the journey will be smooth. Between the ideal and reality lie three major hurdles.
The first hurdle is how beginners get their start. Although everyone acknowledges the importance of higher-level skills, experience cannot simply be bypassed. No one gains a global perspective without hands-on involvement in specific tasks. But while experience cannot be skipped, my view is that it can be compressed. Junior engineers need extensive opportunities to operate real systems, and this is where AI can deliver its greatest value—serving as an on-demand sparring partner where failure carries zero consequences. Using AI to simulate a realistic engineering environment, allowing students to experience the full lifecycle from requirements analysis and architectural design to deployment, maintenance, and troubleshooting before entering the workforce, is one of the most effective ways to shorten this trial-and-error period.
The second hurdle stems from a widespread concern: if junior roles are cut, will that block the pipeline to senior talent? This sounds plausible, but it makes the same fundamental mistake: it assumes that doing junior-level work is preparation for senior-level responsibilities. As discussed earlier, the two are largely different tracks. Within the spectrum of junior experience, the portion genuinely useful for senior growth is actually quite small—such as knowing what real systems look like and daring to take ownership of an outcome; the bulk of the rest is execution that AI can already take over. What we should truly worry about is that this small, essential slice gets abandoned. Cutting away the 90% mechanical repetition along with the 10% judgment training—that is what would truly sever the path forward.
The final hurdle is perhaps the most counterintuitive: the breakdown of this sieve is not because AI is too clever and making decisions for humans, but because the next-stage skills—judgment, communication, collaboration, and ownership—are inherently harder to test on legacy exams.
Grading algorithmic problems is effortless: run a test case and see if the output matches. But evaluating how someone makes trade-offs in an ambiguous situation, how they align consensus when a team is in conflict, or whether they step up when an outage strikes—making that kind of assessment objective and scalable comes with an astronomical cost. The sieve breaks down because its mesh only recognizes execution, remaining entirely blind to judgment, collaboration, and ownership.
The rice cooker is here to stay, and the sieve will not fix itself. The urgent imperative is to stop pretending otherwise: in the past, selecting and educating talent relied on a single sieve that only measured execution. Now that it can no longer filter, this is the exact moment to figure out how to teach, practice, and evaluate the capabilities that a rice cooker cannot deliver—judgment, design, collaboration, and ownership. No one can do this step for us, just as no rice cooker can stand before the cutting board in place of the apprentice.