Post
JA EN

Why Japan's Engineering Managers Burn Out: Three Separations You Must Make in the AI Era

Posted
By
45 min read
Why Japan's Engineering Managers Burn Out: Three Separations You Must Make in the AI Era
  • Who this is for: Engineering managers and tech leads, senior engineers being courted into management, and HR or executive teams designing organizations.
  • Assumed background: Familiarity with common team-management vocabulary (1:1s, performance reviews, PIPs).
  • Reading time: about 60 minutes.

Overview

“Engineering management is a punishment game.” In Japan this phrase has crossed over from grumbling into a structural diagnosis. It is backed by data, not just sentiment: 96% of Japanese line managers also do individual-contributor work1; only 21.4% of non-managers say they want to become a manager—the lowest level among comparable economies2; and Gallup’s most recent global poll puts Japanese employee engagement at 6%, the worst in the developed world3. In April 2026, Nikkei Business argued explicitly that the absence of US-style technical career tracks (Engineer → Senior → Staff → Principal → Distinguished) is what funnels every senior engineer through a single management gate4.

A short note on language for international readers: the phrase translated here as “punishment game” (batsu-ge-mu, 罰ゲーム) is a Japanese workplace idiom for a role nobody wants because it grinds people down. It was popularized by the labor researcher Yuji Kobayashi in his 2024 award-winning book Kanrishoku ga Batsu-Game ni Naru (“Management as a Punishment Game”). “Playing manager” (pureiingu manēja) is the local term for a manager who must keep delivering individual contributor output on top of management duties—what English-speaking readers might call a “working manager,” but with stronger expectations on the IC side. Both terms appear repeatedly in the data below.

The thesis of this article is simple: the punishment-game effect is what happens when an organization refuses to separate the distinct functions that the word “management” papers over, and instead jams them all into a single EM role. “Management” as a job title hides activities that differ in what they target, what domain expertise they require, and what kind of claim they make about the world. Treat them as one bundle and the EM is structurally guaranteed to drown.

Three separations, in particular, are non-negotiable.

SeparationWhat gets separated
1. ObjectManaging AI (a personal skill everyone needs) vs. managing humans (a manager’s job)
2. DomainOperational management vs. specialized interpersonal management
3. Discussion layerLegal facts vs. psychological tendencies vs. personal values

The first thing to underline is that managing AI is not a skill specific to EMs—it is a personal skill every working professional needs in the AI era. Engineers, designers, marketers, and operations staff all increasingly brief, delegate to, and review the work of AI agents. Camille Fournier’s playbook for delegation, review, and feedback5 and Andy Grove’s idea of task-relevant maturity6 map cleanly onto how you brief an agent, calibrate trust, and review output—but they apply as personal productivity skills for everyone, not as a new managerial responsibility. Human motivation, careers, and mental health, by contrast, sit in a body of expertise (Self-Determination Theory7, the Job Demands-Resources model8) that an engineer will not absorb by reading a book over the weekend—and this is where the EM’s actual battlefield lies. Confusing the two—mixing “AI management” with “people management”—is what loads disproportionate responsibility onto a single EM.

Until an organization institutionalizes these three separations, no amount of individual heroics will dissolve the punishment-game structure. Japan layers four additional structural amplifiers on top of that base problem—missing professional managers, normalized playing-manager work, the labor-law structure, and the calcification of misfit senior layers. This article documents the symptoms, defends the necessity of each separation with academic and field evidence, names the four amplifiers, and ends with prescriptions for both the organization and the individual EM. Once you let go of four illusions at once—”AI management is a manager’s job,” “everyone can keep growing,” “if you’re a good engineer you must also be a good people manager,” and “if we just create the environment, everyone will change”—a humane design starts to come into focus.

The Symptoms: Numbers Behind “Punishment Game”

Multiple independent surveys point in the same direction.

Yuji Kobayashi of Persol Research and Consulting—whose book Kanrishoku ga Batsu-Game ni Naru won the 2024 HR Award and whose December 2025 column updates the picture—identifies a “quadruple bind” in Japanese middle management2:

Pain pointShare of managers reporting it
Increasing workload52.5%
Difficulty developing reports37.5%
No successor in sight56.2%

In international comparisons, Japan’s pool of would-be managers is the smallest: only 21.4% of non-managers want a management track (versus more than 80% in India and Vietnam). Japan’s manager headcount roughly halved between 1995 and 2015, and the typical age of first promotion to kacho (section manager) is around 40, against 28–34 in peer economies2. Gallup’s State of the Global Workplace puts engaged Japanese employees at 6%, well below the East-Asia average of 18% and the global average of 23%3.

A 2019 Sanno Institute study found that about 96% of department heads in listed Japanese companies are playing managers, with individual-contributor work taking a weighted-average 39.9% of their time1. The management-research literature converges on the view that high-performing playing managers should keep IC work below roughly 30% of their time. Japan’s current average is already brushing the upper bound at which the role can still produce results.

Together these numbers describe an EM job that has been engineered to exceed what individual effort can absorb.

Core Diagnosis: Three Separations Are Failing

Why does this load show up here, and at this scale? The diagnosis at the heart of this article is that the word “management” smuggles together activities that are fundamentally different in kind.

Japan’s punishment-game effect arises because those activities are never deliberately separated; they are stacked into one EM role. AI management in particular—which is supposed to be a personal skill for everyone—often gets reframed as “the manager’s new job” and piled on top of the existing interpersonal load.

flowchart TB
  AI["Briefing / delegating to / reviewing AI<br>(should be everyone's personal skill)"]
  WORK["Operational feedback"]
  MENTAL["Mental-health support"]
  CAREER["Career redesign"]
  VALUE["Transmitting values<br>('you should grow')"]
  EM(("One EM<br>holds it all"))
  PUNISH["Punishment game"]
  AI --> EM
  WORK --> EM
  MENTAL --> EM
  CAREER --> EM
  VALUE --> EM
  EM --> PUNISH

These are different activities, with different objects, domains, and epistemic status. Refusing to separate them is itself the trigger for burnout. The rest of this article unbundles them along three axes.

flowchart TB
  TASK(("The work we call<br>'management'"))
  TASK --> S1["Separation 1: Object<br>AI (everyone's personal skill) vs.<br>humans (a manager's job)"]
  TASK --> S2["Separation 2: Domain<br>Operational vs. specialist"]
  TASK --> S3["Separation 3: Discussion layer<br>Legal / psychological / values"]

Separation 1: Object — AI Management Is Everyone’s Personal Skill; People Management Is the Manager’s Job

The first separation is whether the object of management is an AI agent or a human being. The claim to underline here is precise: managing AI is not an EM-specific competency—it is a personal skill that every working professional needs in the AI era. The EM punishment-game problem and the AI-management discussion live on different layers, and conflating them is itself part of the problem.

AI management is everyone’s core skill

Some commentators argue that “management skills will become irrelevant in the AI era.” That’s only half right. Briefing AI agents, delegating design, reviewing output, calibrating trust—the manager-of-AI skill set—is becoming a core competency for everyone, not an obsolete one. For engineers, “designing what an AI agent should write and reviewing the output” is rapidly becoming more central than writing code by hand. Designers, marketers, researchers, and even operations staff increasingly need to delegate to, review, and re-instruct AI agents. The skill is role-agnostic.

Existing management literature is highly reusable as a curriculum for this skill. The 1:1, feedback, delegation, and capability-development practices in Camille Fournier’s The Manager’s Path5, the team-sizing and technical-decision frameworks in Will Larson’s An Elegant Puzzle9, and Andy Grove’s notion of task-relevant maturity (TRM)—adjusting how prescriptive your instructions are based on the operator’s mastery of the task6—all transfer cleanly to AI agents. Designing task chunks for an agent, judging what the model can and cannot reliably do, reviewing output, redelegating with corrected context, and giving explicit feedback are now part of the daily craft of every working professional, not just senior engineers and certainly not just EMs.

The point worth driving home is this: don’t mistake “AI management” for an EM-exclusive skill just because it shares the word “management.” AI management is a personal productivity skill. There is no reason it must reside with an EM, and treating it as a senior-engineer market-value asset that everyone should grow—regardless of job title—is the right framing.

Human management is a different sport—using the AI playbook on humans backfires

The trap is assuming AI-management practices transfer to humans. With an AI, “tighten the spec, tighten the review” raises quality. With a human report, the same intervention often does the opposite. Self-Determination Theory (Deci & Ryan) has spent four decades demonstrating that intrinsic motivation rests on three basic psychological needs: autonomy, competence, and relatedness7. Gagné and Deci’s foundational application to work (2005, J. Organ. Behav., DOI 10.1002/job.322, with more than 6,980 citations) experimentally established that threats, deadlines, directive evaluation, and imposed goals reduce intrinsic motivation7. The well-meaning push of “I want to see more growth from you” is, by SDT’s reading, more likely to suppress motivation than spark it.

In other words, the methods of AI management and human management diverge at the level of mechanism. Human management is, properly, the EM’s actual battlefield.

The connection to EM burnout

This is where the link to the punishment-game effect comes into focus. Part of why EMs burn out is that the line between “AI management = personal skill” and “human management = managerial specialty” is left fuzzy inside the organization, and both loads end up on the EM’s shoulders.

  • AI management arrives as “the manager’s new responsibility”: something that should be a personal skill cultivated across the whole organization is added to the EM’s job description.
  • People who are good at human management short-circuit to “so AI management is also the EM’s job”: the two are different in object but both called “management,” and the equivalence is taken for granted.
  • And in the other direction: success at AI management is exported to humans—running reports with “instructions and strict review”—which damages intrinsic motivation in SDT terms.

If the two are not deliberately distinguished, an EM exhausts their reports by importing AI-management instincts into human relationships, or burns out on human management with no bandwidth left to grow AI skills, or carries both at once and burns out themselves. Making “AI management is everyone’s; human management is the manager’s” an explicit organizational frame is the first move in unwinding the punishment-game structure.

A further wrinkle: even within “managing humans,” there is a second internal split between what the EM can carry and what should be handed to a specialist. That is Separation 2.

Separation 2: Domain — Operational vs. Specialized Interpersonal Management

The second separation, inside the human-management domain, is the line between what an EM can legitimately handle and what should be handed to a specialist.

The case for a hard boundary here comes straight from the Job Demands-Resources model of Demerouti and colleagues, which shows that burnout emerges from an imbalance between the demands a role places on you and the resources you have to meet them8. The Japanese EM is asked to absorb a near-pure stack of demands—40% IC work1, stagnating reports, a 56.2%-rate of “no successor”2—while resources (a real professional management track4, real authority) are anemic. That is a textbook recipe for burnout. Without bounding the domain the EM owns, demands will simply expand to fill the role.

What an EM can own

  • Setting operational expectations and giving operational feedback
  • Assigning tasks and adjusting scope
  • Running 1:1s focused on progress and obstacles
  • Skill development on the job (code review, pairing, internal study groups)
  • Designing team communication norms
  • Tooling and AI/dev environment hygiene

These behaviors sit on the same continuum as the ten manager behaviors Google’s Project Oxygen identified—coaching, empowerment, clear vision, technical guidance—and are accessible to anyone with strong engineering judgment10.

What should be handed to a specialist

  • Assessing and treating mental health → occupational physician, occupational counselor
  • Career redesign and role-fit reconsideration → career counselor, HR
  • Long-running underperformance (PIPs, reassignment, separation) → HR
  • Harassment investigations and remediation → compliance, legal, HR
  • Transfers out of the team → HR
  • Deep, structural motivational issues (long-term unmet SDT needs) → external coach, OD specialist

The reason for the line is not that handing things off makes the EM’s life easier. It is that diagnosis itself is specialist work.

Diagnosing “someone who isn’t growing” is a specialist skill

One of the judgment calls EMs face most often is: “this report’s growth curve has flattened—what now?” Reducing that to a single axis (“low motivation”) is dangerous. Beneath the surface there are at least five distinct factors at play (this matrix is my own consolidation, mapping SDT7, occupational-health frameworks, and career research onto a practical decision tool):

FactorContentCan the EM handle it alone?
Capability profileDifferences in learning rate or abstraction; neurodiversity (ADHD, ASD, etc.); skill-role mismatchNo—occupational physician / specialist territory
SituationalCaregiving, child-rearing, health, mental-health onset, financial or physical pressureNo—HR / occupational physician / counselor territory
Career stanceDeliberate choice to treat work as a means of life; phase-of-life prioritization (e.g., new parent, late-career stabilization)Partly—dialogue between the person and HR
EnvironmentalLack of learning opportunities, low psychological safety, no role modelsYes—EM + organization
Motivational structureLong-term failure of SDT’s three needs (autonomy, competence, relatedness)Partly—EM owns environment, specialists own the deep layer

Many cases that an EM reads as “lack of motivation” are in reality capability profile (e.g., undiagnosed neurodiversity), situational (a caregiving load, early-stage depression), or career-stance (a rational decision to optimize for life outside work) phenomena. Telling these apart requires distinct kinds of specialist knowledge—occupational medicine, occupational counseling, career counseling, organizational development.

Handing specialist interpersonal management to specialists is not, therefore, an exercise in offloading effort; it is acknowledgment that an engineer’s amateur diagnosis carries an unacceptable misdiagnosis risk. Even a Performance Improvement Plan—the apparently mechanical case—has reported success rates clustering in the 40–60% range, with the higher end requiring genuine support structure11. In Japan, Article 16 of the Labor Contract Act prevents a US-style PIP from functioning as a “termination warning”; instead it must serve as documentation of long-term improvement support11. That is HR/legal territory, full stop.

Span of control reinforces the boundary

Harvard Business Review’s research on management span shows the average CEO’s direct reports doubling from about 5 in the 1980s to roughly 10 in the 2000s12. For knowledge workers a span of 15–20 can be workable, but for complex roles 3–6 is the safe range12. The moment one of your reports is in any of the five-factor categories above, your effective span shrinks dramatically. Japanese EMs are quietly expected to maintain heavy IC work and a wide span at the same time.

There is no design path that lets an unbounded EM absorb this load. That is the second separation.

Separation 3: Discussion Layer — Legal Facts, Psychological Tendencies, Personal Values

The third separation cuts the other way. It says: when you talk to a report (or a peer, or yourself) about “growth,” “learning,” or “ambition,” do not blend three distinct epistemic registers. Mixing them is what produces the unproductive “should people grow / shouldn’t they grow?” arguments that EMs find themselves trapped in.

LayerNatureExamples
Legal factsConstraints that hold regardless of personal opinionEmployers cannot mandate off-hours study; Article 16 of the Labor Contract Act limits dismissal; duty of safe-care obligations
Psychological tendenciesEvidence-based probabilistic patterns; individual variation expectedSDT, JD-R, Project Oxygen behaviors
Personal valuesMultiple legitimate answers; the individual choosesWhether to study off-hours, whether to seek management, whether to optimize for growth or stability

Under Japanese labor law, employers cannot compel off-the-clock activity. The widely-cited blog post by Ayumu Yonemura at Axia—arguing that the unspoken expectation of off-hours study is “corporate arrogance”—is best read not as a values stance but as a clean restatement of the legal default13. Article 16 of the Labor Contract Act, which requires “objective and reasonable grounds” for dismissal, is similarly a fact about the operating environment11.

Layer 2: psychological tendencies

What SDT and JD-R provide are probabilistic claims about the average effect of an intervention78. “External pressure tends to undermine intrinsic motivation”; “high demand plus low resources tends to produce burnout.” These are tendencies, not laws. People who thrive under pressure and people who never burn out at high demand exist. The right way to use psychological tendencies is as a sensible default for system design, not as a verdict on any specific person.

Layer 3: personal values

“Should I study off the clock?” “Do I want to grow at all costs?” “Do I want to be a manager?” These are value questions in the strict sense. There is no externally-correct answer. Some engineers want to keep learning during evenings and weekends; others want clean separation. Neither preference is wrong.

Refusing to require off-hours learning is legally and psychologically defensible, but that is not the same as denying that some employees genuinely want to study off the clock. The problem is making their personal choice into the implicit baseline for everyone.

What this means in practice

Before an EM tells a report “I want to see more growth from you,” it is worth pausing to ask which layer the statement lives in:

  • Legal/contractual expectation (“here is what the role requires within working hours”) → fine to state explicitly
  • Information about a psychological tendency (“over the long run, people who keep learning tend to do well”) → fine as information, not as instruction
  • Intervention into someone’s personal values (“I want you to be more ambitious”) → handle with care; this is encroachment on someone’s life choices

Mixing the layers is how an EM ends up using business-expectation language to lecture someone on their life. It is also how an EM ends up importing infinite “developmental responsibility” into their own role.

The Reality of Off-Hours Learning: Ideal, Illusion, and Selection

Once the three separations are in place, one practical question still bites: what about off-hours learning? Stopping at “you can’t legally require it” is a cop-out, because it ignores the brutal pace of the field.

What the data says about study volume and outcomes

IT skills decay fast. Multiple industry analyses converge on a half-life of about five years for general professional skills and under three years for specific IT technical skills14. The World Economic Forum’s Future of Jobs Report 2025 projects that 39% of skills will become obsolete or transformed between 2025 and 203015.

Japan’s self-study numbers do not match that pace:

IndicatorValueSource
Self-development participation (all workers)36.8%MHLW Basic Survey on Human Resources Development 202416
Self-development participation (international, Japan)27.2% (lower tier of seven countries)Recruit Works, Global Career Survey 202417
Adult learning among high-skilled adults52% (OECD average 70%)OECD PIAAC 202318
Corporate education spend / GDP0.1% (Western peers above 1%)MHLW analysis16

Globally, the Stack Overflow Developer Survey 2024 reports that 68% of developers code as a hobby and 40% study outside work specifically to advance their career19. The world’s top engineers learn off the clock as a matter of course. The gap between Japan’s self-development rate and the global developer norm is large and real.

In other words, the assumption that “in-hours learning alone will keep you at industry par” is empirically false. That is not a moral observation; it is a structural one driven by skill half-lives and global learning norms. People who learn voluntarily outperform on average—both research and field experience back this up.

The ideal model: growth through environment

If off-hours learning cannot be required, what can an organization do? The theoretically supported route is the natural development of intrinsic motivation through environment.

  • Peer effects. Mas & Moretti (2009, American Economic Review 99(1), 112–145, DOI 10.1257/aer.99.1.112) show in a natural experiment with supermarket cashiers that working alongside more productive peers raises an individual’s productivity—a 10% rise in coworker permanent productivity yields about a 1.5% rise in the focal worker’s productivity (an elasticity of about 0.15)20.
  • Communities of practice. Lave and Wenger’s Situated Learning (1991) frames learning as legitimate peripheral participation in a practicing community21.
  • Internalization in SDT. Deci and Ryan describe motivation moving along a spectrum—external regulation → introjected → identified → integrated → intrinsic—with autonomy, competence, and relatedness as the catalysts that drive movement up the chain7.
  • Four-phase interest development. Hidi and Renninger (2006, Educational Psychologist 41(2), DOI 10.1207/s15326985ep4102_4) describe interest forming in four stages—triggered situational → maintained situational → emerging individual → well-developed individual22. The “triggered situational” sparks an organization can provide become the long-run soil for individual interest.

The shared message is that there is a real, theory-backed path from “people around me are learning” to “I find this fun” without anyone being coerced. Csikszentmihalyi’s flow theory23—the absorption that happens when challenge and skill are balanced—rises from exactly this kind of soil. An organization’s responsibility, on this view, is to provide the opportunities: internal study groups, pair programming, learning time during work hours, book budgets, conference travel, exploratory time slots.

The reality: this works for fewer people than we want to admit

The ideal model does not work uniformly.

  • The Mas–Moretti peer effect is real but modest—an elasticity of about 0.1520.
  • Japan’s PIAAC 2023 numbers say that even within a learning culture, roughly half of high-skilled adults do not participate18.
  • Stack Overflow’s 68% “code as a hobby” implies a 32% who do not, even among the global elite19.
  • The SDT internalization literature itself notes that not everyone reaches the intrinsic end of the spectrum7.

So: the path “discover the joy and grow” exists, but the share of people who walk it is a minority. The research punctures both the organization-side fantasy (“if we just create the environment, everyone will change”) and the individual-side fantasy (“if you’re pressured enough, you’ll grow”).

Person-Job Fit: not “competent vs. incompetent” but right role vs. wrong role

A further misconception worth dispatching: arguments about “what to do with someone who isn’t learning” tend to slide into a binary of competent vs. incompetent. That is wrong, both scientifically and practically.

Person-Job Fit research—the organizational-psychology literature on how well a person’s profile matches a role—shows that the same individual can shift dramatically in satisfaction, commitment, and performance simply by changing role. The Kristof-Brown, Zimmerman & Johnson meta-analysis (172 studies, Personnel Psychology 58) reports ρ = .56 between Person-Job Fit and job satisfaction, ρ = .47 with organizational commitment, and ρ = .20 with task performance—medium-to-large effects24. The verdict “doesn’t learn = incompetent” is not supported by the data. People who look like they aren’t growing in one role often have strengths that surface in a different role.

Buckingham and Clifton’s Now, Discover Your Strengths (2001) makes the case that engagement and productivity rise when people are placed in roles aligned with their strengths25. Wrzesniewski and Dutton’s job-crafting paper in Academy of Management Review 26 reframes the relationship: don’t only fit people to jobs, let people reshape jobs around their strengths26. The arrow runs in both directions.

There is also a wall that motivation alone cannot scale: competence, the second of SDT’s three needs7. Cognitive style, neurodiversity, life stage, and physical condition are not under the individual’s control, and they shape which roles they can succeed in. The five-factor matrix above maps onto exactly these. Reading “low motivation” off someone whose actual situation is capability or life-stage misalignment hurts both the person and the organization—the person is told they have a character flaw, and the organization loses the chance to redeploy them well. “They’re not incompetent; they are in the wrong seat” is the realistic prescription.

Conclusion: filter at the door, fit existing employees in place

This produces a split in policy by audience.

  1. Provide the learning environment (organizational responsibility): study groups, pairing, in-hours learning, book budgets, conference attendance. This is the floor.
  2. At hiring time, filter: for senior roles that genuinely require autonomous learning, listing “habit of self-directed learning” as a hiring criterion is legal, fair, and unambiguous. Drawing the line at the hiring contract—between “in-hours-complete roles” and “self-driven-learning roles”—prevents most of the downstream conflicts we’ll see in a moment.
  3. For existing employees, place them well: forcing off-hours learning on a current employee is legally and psychologically a non-starter, but Person-Job Fit research24 tells us that the same person in a different role can perform very differently. Don’t reach for “underperformer” as a verdict; reach for “is there a role inside the organization—maintenance, operations, documentation, support, QA—where this person’s profile pays off?”

When fit can’t be found, the process must still be gradual. The Bloomberg L.P. case (Tokyo High Court, 2013) ruled that even after three rounds of PIPs, dismissal for incompetence was invalid; the court required both “a degree of capability decline that makes continued employment impractical” and “no prospect of improvement”27. Rights to assign reassignments are bounded by the Toa Paint case (Supreme Court, 1986) and its three tests: business necessity, absence of improper motive, and no unreasonable burden on the employee28. The Shiga Prefecture Social Welfare Council case (Supreme Court, April 2024) added a first-of-its-kind ruling: for employees with a contractually-limited job category, the employer cannot reassign them outside that category without their consent29. Whether the original contract limited the role to “engineer” matters enormously here.

The scientific backbone of running such processes is organizational justice. Colquitt’s J. Appl. Psychol. meta-analytic work establishes that the four dimensions of justice—distributive, procedural, interpersonal, and informational—materially shape commitment and performance30. For an outcome (a reassignment, a separation) to be experienced as “the consequence of the person’s own choices” rather than an arbitrary blow, four conditions must hold: (1) opportunities were objectively offered, (2) the evaluation process was transparent, (3) outcomes were predictable, and (4) the person was given an explanation and a hearing. Fail those and you accumulate both a commitment hit and a litigation risk.

Net: organizations carry both a “provide opportunity” responsibility and a “search for fit” responsibility. The individual carries the choice of whether to learn, and that choice changes which roles they fit. None of this asks anyone to break the law; it asks the organization to find a workable balance between autonomy and institutional accountability inside the legal constraint.

“Everyone learns and loves it” is a worthy north star. It is also a low-probability outcome at the population level. Doing filtering at hiring and fit-search for current employees as separate policies is what lets you keep the organization’s skill bar high without grinding people down.

Japan-Specific Structural Amplifiers

The three separations and the off-hours-learning question are international concerns. Japan adds four amplifiers on top: a missing institution (no professional manager track), a normalized operating mode (playing manager as the default), a labor-law structure (dismissal restrictions and membership-type employment), and a cumulative outcome of those (calcified misfit senior layers).

Amplifier 1: missing professional manager track

Takeshi Kimura at Nikkei Business points out that the US-style technical career ladder (Engineer → Senior → Staff → Principal → Distinguished, with compensation matching the management track) has not taken root in Japanese companies4. In Japan, “I want to remain a technologist; I do not want to be a manager” is read as “no ambition.” Even at large system integrators that nominally have a specialist track, it usually does not function in practice.

There are exceptions—Omron’s four-tier specialist program (introduced in 2005, capped by a “Fellow” rank with compensation at or above the management track) is the most-cited31—but the diffusion is narrow. Allen and Katz’s dual-ladder research adds a structural caveat: the technical track tends to lose status and pay battles to the management track, and the gatekeeper effect of the supervisor on promotion paths is large32. Drawing the ladder isn’t enough; you have to make the rungs real.

Amplifier 2: normalized playing-manager work

96% of Japanese managers are playing managers, and the 39.9% IC share is already past the recommended ceiling of about 30%1. Adding specialized interpersonal management to a manager already at saturation is structural breakage, and the math is plain. AI-management—the personal skill every working professional now needs—has to be picked up by EMs as individuals on top of all this; the IC content shifts but the time pressure does not, it becomes more demanding, not less. The crucial point is that this should not be foisted onto EMs as a new managerial responsibility; it should be framed as a personal skill the whole organization is investing in.

Amplifier 3: labor-law structure

The OECD’s Employment Protection Legislation index places Japan’s protection of regular employment in the relatively strict tier among OECD countries33. The OECD’s 2007 analysis estimated that a one-unit change in EPL is associated with a statistically-significant decline of about 0.02 percentage points in annual labor productivity growth and about 0.04 points in TFP growth33. The share of Japanese employees with under one year of tenure is only 7.3%, an extraordinarily low level of mobility internationally33.

Keiichiro Hamaguchi’s What is Job-Type Employment Society? (Iwanami Shinsho, 2021) frames the Japanese norm not as “job-based” employment but as “membership-based” employment—you are hired into the company as a member, not into a specific job. The four-element redundancy-dismissal doctrine, he argues, was built up in the courts as a follow-on to membership-type hiring; the result is a self-reinforcing loop in which the doctrine entrenches membership-type hiring, which in turn entrenches the doctrine34. In a job-based regime, the disappearance of a job is a legitimate cause for dismissal. In a membership regime, the obligation is to find the person another job inside the company—which is exactly why reassignment authority is broad and dismissal-for-incompetence is hard to win.

The connection to the punishment-game effect is direct. An EM cannot release a non-functioning report not because of personal soft-heartedness but because the legal regime obligates the organization to find another role. And the pool of “another role” is bounded by the Shiga case29 when the employee has a job-limited contract. The result is the EM operating under the triple constraint of “can’t release, can’t reassign easily, can’t develop fully.”

A caveat: the OECD’s deeper criticism is not the strictness of regular-employee protection per se, but its role in producing a sharp regular/non-regular dualism33. “Just deregulate it” oversimplifies the picture. This article does not argue for legal reform; it argues that recognizing how the current legal architecture amplifies the punishment-game structure is necessary for any organization-level or individual-level prescription to land.

Amplifier 4: misfit senior layers entrenched (Peter Principle × seniority)

The fourth amplifier is what membership-type employment plus seniority promotion produces over time: a tendency to lock people who don’t fit a role into the role above them. This presses down on EMs from above.

Laurence Peter and Raymond Hull’s The Peter Principle (1969) gave the structural observation: “in a hierarchy, every employee tends to rise to their level of incompetence”35. Promote on past results, plateau on present results, and the organization fills with people who can no longer do the role they hold.

Benson, Li, and Shue empirically confirmed this in The Quarterly Journal of Economics 134 (2019, DOI 10.1093/qje/qjz022) using data on 38,843 sales workers across 131 US firms36. Top sales producers were promoted into management—and sales-production rank failed to predict managerial performance. “A great engineer becomes a great people manager” is not a guarantee. That observation is the empirical backbone of this article’s central claim that personal skills (engineering, AI management) and people-management as a specialty are different sports.

In Japan’s membership-type regime the effect intensifies. With no real specialist track4, the promotion path narrows to a single channel: management. First kacho promotion happens around 40 (vs. 28–34 elsewhere)2, and a “your turn came around” pattern based on tenure becomes the operating reality. The Cabinet Office’s Japan Management and Organizational Practices Survey (the Japanese counterpart of Bloom and Van Reenen’s World Management Survey) finds Japanese firms scoring above-average on monitoring and target-setting but below-average on incentives—consistent with selection by tenure rather than by demonstrated capability37.

That layered above the field EM produces five secondary loads:

  1. Lower-quality technical decisions. AI investment, team structure, technology choice get overridden by senior-layer judgments that disregard field-level reasoning.
  2. Outdated evaluation axes. “Time in seat” beats “writes code well” or “keeps learning” in evaluation reproduction.
  3. Specialist-track absence is reproduced. Senior leaders who do not see management as a specialist profession do not invest in building real specialist tracks.
  4. AI transformation lag. Senior leaders unfamiliar with AI management (everyone’s personal skill in the AI era) slow the organization’s overall adaptation; field EMs absorb the additional load.
  5. “The boss themselves isn’t growing” cases. The five-factor matrix above doesn’t apply only to reports—it applies to seniors. When seniors carry capability-profile or situational issues, the field cannot diagnose, name, or push back; it just gets buffeted.

So: “no specialist manager track” is not just a missing feature; it is the cumulative outcome of a hiring-and-promotion regime that calcifies misfit senior layers in place. Untangling the EM punishment game without applying Person-Job Fit24 to senior layers themselves is not realistic.

The naïve fix—replace the senior layer—has its own failure mode. Long tenure builds context knowledge and relationship capital that are real organizational assets; tearing them out creates a different problem. The right design separates “manager fit” from “context-knowledge bearer”: a real specialist track (organizational principle 4 below) is not just a humane outlet for engineers, it is the prescription for the senior fit problem too. Senior leaders who carry deep technical or organizational knowledge but lack management fit can move into Fellow / advisory / mentor roles where their knowledge is preserved without forcing them into a role they don’t fit.

Four amplifiers, compounding

Stack the four amplifiers and the picture is sharp: the Japanese EM is asked to do the work that needs the most separation, in the configuration that allows the least separation. No specialist track. Playing-manager work as the default. Dismissal restrictions that obligate retention of misfit reports. Misfit senior layers above. Any one of those is survivable; together they compound.

Prescription (Organizational): Four Principles to Institutionalize Separation

Given the diagnosis, the organizational prescription writes itself: make the three separations institutional, not heroic.

Principle 1: explicitly split management into three layers

Stop bundling AI management, operational management, and specialist interpersonal management into a single role. The most important thing to avoid is loading AI management onto the EM as a “new responsibility.”

LayerOwnerCore skill
AI management (personal skill)Everyone (regardless of role)Task-chunk design, output review, TRM evaluation
Operational managementEM1:1s, evaluation, operational feedback
Specialist interpersonal managementHR pros, occupational physicians, external coaches, ODMental-health assessment, career redesign, long-running underperformance

The EM owns layer 2. The organization owns the handoff path to layer 3. Layer 1 (AI management) is a personal skill that every engineer and every working professional should grow, not just EMs, and the organization owes the whole workforce—not just managers—training opportunities, tooling, and shared best practices. Avoid the trap of “EM is responsible for catching up reports who are slow on AI”; treat AI fluency as an organization-wide development program, not as a managerial deliverable.

Principle 2: institutionalize escalation criteria

Specify when the EM stops carrying something and escalates. Escalation is not abdication; it is the EM doing their job correctly.

  • Same issue surfaces in 1:1s three times in a row → loop in HR
  • Mental-health warning signs (sleep, appetite, focus changes; rising self-criticism) → schedule occupational-physician consultation immediately
  • Performance stagnation lasting more than six months → confer with HR on reassignment / PIP
  • Suspected harassment → immediate compliance handoff (the EM is not the investigator)

Principle 3: drive playing-manager ratio down structurally

The recommended IC share for high-performing playing managers sits below ~30%; Japan’s weighted average is already 39.9%1. There is no slack to add interpersonal management on top. Treat this as a policy commitment: cap IC work at under 30% by (1) reviewing managers’ coverage, (2) right-sizing teams (3–6 reports for complex roles12), (3) delegating technical decisions to lead engineers, and (4) staffing the human-domain side with HR and OD specialists.

Principle 4: make the specialist track real

Build a real specialist track—four levels, compensation at or above management, a dedicated research budget, the works—on the model of Omron’s Fellow program31. The dual-ladder research warns that building the track on paper is not enough32. The operating culture (specialist-track moves are not demotions), the evaluation criteria (no implicit penalty for absence of management experience), and the existence of real upper rungs (actual Distinguished Engineers and Fellows) all need to be in place before the track does any work.

Prescription (Individual): Embedding Three Separations in Daily Behavior

Institutional change takes time. In parallel, an EM can embed the three separations into daily behavior.

Practicing Separation 1 (object): grow your AI-management skill on purpose—but don’t carry it as an EM-specific duty

Train yourself, deliberately, in delegating to AI agents, reviewing their output, and evaluating task-relevant maturity. This isn’t required because you’re an EM; it’s required because every working professional needs it in the AI era, and it directly compounds your market value as an engineer. Shifting your IC component from “writing code directly” to “designing for AI delegation and reviewing output” lets you maintain your IC share while staying technically engaged.

What to watch out for: don’t slide into the mindset of “I, as the EM, am responsible for raising the AI fluency of every report who’s behind.” AI management is a personal skill, and individual employees own the responsibility to learn it on the clock. The EM is responsible for environment—offering learning opportunities, encouraging in-hours learning—but is not responsible for tracking and lifting each report’s individual AI proficiency. Conflating the two converts AI-management load into people-management load and accelerates the punishment-game spiral.

Practicing Separation 2 (domain): use the five-factor matrix as a triage tool

When a report’s growth has flattened, run them through the matrix—capability profile, situational, career stance, environmental, motivational—and identify which factor is in play. The moment a factor falls outside what you can handle, hand off to HR, an occupational physician, or an external coach without hesitation.

Don’t take “I should have developed them” personally. Carrying it personally is the seed of the punishment-game spiral.

Practicing Separation 3 (layer): self-audit your statements in three layers

Before saying something to a report, ask which layer the statement belongs to. Legal/contractual expectation? Information about a psychological tendency? Or an intervention into their personal values? If it’s the third, slow down and pick your words with care.

Statements like “I want you to be more ambitious” or “I want to see you study more outside work” are typical value-intervention markers. Confusing them for operational expectations is how you simultaneously raise pressure on the report and load infinite developmental responsibility onto yourself.

Beyond separations: increase your luck surface area through volume of action

The three separations are defensive practice. Alongside them sits an offensive practice: probabilistically increasing your chances of meeting work that absorbs you and people who fit the role.

Luck exists for both individuals and managers. The individual’s luck is whether they meet a flow-state-inducing piece of work23. The manager’s luck is whether they can find people who fit a given role (the Person-Job Fit search)24. But this kind of luck is not random.

Krumboltz’s Planned Happenstance Theory (Journal of Career Assessment 17, 2009) argues that career serendipity is deliberately cultivable. The key is five behavioral traits: curiosity, persistence, flexibility, optimism, and risk-taking38. Granovetter’s classic “Strength of Weak Ties” (American Journal of Sociology 78, 1973) shows that opportunity flows disproportionately through the wide perimeter of weak ties, not the dense core of strong ones39.

Implementation is mundane. As an EM-individual: show up to internal study groups, try new technologies casually, attend external community events, broaden your small talk. Reframe these not as “forced learning” but as luck-cultivation behavior. As an organization: hiring great people once is not enough—build varied roles, keep the door wide, and increase the rate of internal mobility. Volume of action raises the probability of fit-matching.

Volume of action improves luck. The principle holds for individuals and organizations equally. The “three separations,” “responses to four amplifiers,” “fit-based placement,” and “opportunity provision” laid out above can all be re-read as a systematic way to increase the rate at which the right things happen by accident.

Conclusion

Why does engineering management in Japan turn into a punishment game? Not because anyone failed in spirit, but because data and structure say so. 96% playing managers1, 21.4% management aspiration2, 6% engagement3. The work that requires the most separation has been pushed into the role with the least.

Three separations are the way out.

  1. Object. Grow AI management as everyone’s personal skill (not as an EM-specific duty); treat human management as a different sport that belongs to managers and tech leads.
  2. Domain. Operational management is the EM’s; specialist interpersonal management belongs to specialists. Use the five-factor matrix to avoid misdiagnosis.
  3. Discussion layer. Don’t blend legal facts, psychological tendencies, and personal values.

On top of that, recognize the four Japan-specific amplifiers: missing specialist track4, normalized playing-manager work1, the labor-law structure (dismissal restrictions and membership-type employment)3334, and calcified misfit senior layers (Peter Principle × seniority)3536. Don’t forget that the five-factor matrix applies upward as well as downward.

The prescription is four organizational principles (three-layer split, escalation criteria, IC ratio under 30%, real specialist track) and three personal practices for the EM’s daily behavior.

“Should I learn off the clock?” “Should I be a manager?” “Should I keep growing?” — these are ultimately personal and organizational choices, with no single right answer. But the structural reality is hard: IT-skill half-life under three years14 and 39% of skills displaced or transformed by 203015. People who learn voluntarily perform better on average. The organizational answer is to discharge the opportunity-provision responsibility while applying filter-at-hiring / fit-for-existing-employees as separate policies. Person-Job Fit research24 tells us that “doesn’t learn = incompetent” misreads most cases; “wrong seat” reads them more accurately.

Drop four illusions at the same time: “AI management is a manager’s job,” “everyone can keep growing,” “if you’re a good engineer you must also be a good people manager,” and “if we just create the environment, everyone will change.” Then a humane design becomes visible.

Grow AI management as everyone’s personal skill. Hand human-specialist management to specialists. Respect personal values as personal. Making those three boundaries explicit—at the organizational level—is the starting point for unwinding the punishment-game structure. And alongside the boundaries: whether you meet absorbing work and people who fit a role has a luck component, and that luck is reachable through volume of action23383922. Structure-fixing and luck-cultivation are the two wheels of the cart out of the punishment game.

References

  1. Over 90% of department heads are playing managers — HR Pro / Sanno Institute survey (conducted 2019, published 2020, target: department heads at listed companies with 100+ employees). ↩︎ ↩︎2 ↩︎3 ↩︎4 ↩︎5 ↩︎6 ↩︎7

  2. Management as a Punishment Game: Can the “Strong Middle” Recover? — Yuji Kobayashi, Persol Research and Consulting (December 15, 2025). ↩︎ ↩︎2 ↩︎3 ↩︎4 ↩︎5 ↩︎6

  3. State of the Global Workplace 2024 — Gallup (2024 edition). Japan engagement 6%, East Asia average 18%, global average 23%. ↩︎ ↩︎2 ↩︎3

  4. Foolish Japanese Companies That Promote Engineers Into Management: Tako-tsubo Organizations Block DX — Takeshi Kimura, Nikkei Business (April 16, 2026). ↩︎ ↩︎2 ↩︎3 ↩︎4 ↩︎5

  5. The Manager’s Path — Camille Fournier, O’Reilly Media (2017). ↩︎ ↩︎2

  6. High Output Management — Andrew S. Grove, Random House (1983, revised editions). ↩︎ ↩︎2

  7. Self-Determination Theory and the Facilitation of Intrinsic Motivation, Social Development, and Well-Being — Ryan & Deci, American Psychologist 55(1), 68–78 (2000). Related: Self-determination theory and work motivation — Gagné & Deci, Journal of Organizational Behavior 26(4), 331–362 (2005, DOI 10.1002/job.322). ↩︎ ↩︎2 ↩︎3 ↩︎4 ↩︎5 ↩︎6 ↩︎7 ↩︎8

  8. The job demands-resources model of burnout — Demerouti, Bakker, Nachreiner & Schaufeli, Journal of Applied Psychology 86(3), 499–512 (2001, DOI 10.1037/0021-9010.86.3.499). ↩︎ ↩︎2 ↩︎3

  9. An Elegant Puzzle / Staff Engineer: Leadership beyond the management track — Will Larson, Stripe Press (2019/2020). ↩︎

  10. Google’s Project Oxygen — Harvard Business School Case (started 2008, expanded to ten behaviors in 2018). ↩︎

  11. Performance Improvement Plan Success Rate — Umbrex (HR-industry KPI overview). PIP completion / success rates vary by survey: a Blind survey reports 41%, while supportive structures push the figure into the 46–60% range. Related: The PIP Trap: Why PIPs Don’t Mean What You Think in Japan — TokyoDev (Article 16 of Japan’s Labor Contract Act constraint). ↩︎ ↩︎2 ↩︎3

  12. How Many Direct Reports? — Harvard Business Review (April 2012). Related: Span of Control and Span of Attention — Harvard Business School Working Paper. ↩︎ ↩︎2 ↩︎3

  13. “I Don’t Want to Study Even a Little Bit on My Personal Time”: A Story About an Employee — Ayumu Yonemura, Axia Inc. (September 7, 2018). ↩︎

  14. The Half-Life of Skills Is Shrinking — What It Means for L&D — Skillable / IBM research review (2024). General-skill half-life ~5 years, IT technical-skill half-life under 3 years. Related: An Engineering Career: Only a Young Person’s Game? — IEEE Spectrum. ↩︎ ↩︎2

  15. Future of Jobs Report 2025 — World Economic Forum (January 2025). 39% of skills obsolete or transformed between 2025–2030. ↩︎ ↩︎2

  16. Reiwa 6 Basic Survey on Human Resources Development — Ministry of Health, Labour and Welfare, Japan (published November 2024). Self-development participation 36.8% (regular employees 45.3%, non-regular 15.8%). Corporate education spend / GDP figure cited from MHLW analysis and related reporting. ↩︎ ↩︎2

  17. Global Career Survey 2024 — Recruit Works Institute (2024). Seven-country comparison: Japan self-development 27.2%, OJT 44.3%. ↩︎

  18. Survey of Adult Skills (PIAAC) 2023 — Japan Country Note — OECD (December 2024). Japan high-skilled adults’ learning participation 52% (OECD average 70%). ↩︎ ↩︎2

  19. Stack Overflow Developer Survey 2024 — Stack Overflow (2024). 68% code as a hobby; 40% study outside work for career advancement. ↩︎ ↩︎2

  20. Peers at Work — Mas & Moretti, American Economic Review 99(1), 112–145 (2009, DOI 10.1257/aer.99.1.112). A 10% rise in coworker permanent productivity yields about a 1.5% rise in focal-worker productivity (elasticity ≈ 0.15). ↩︎ ↩︎2

  21. Situated Learning: Legitimate Peripheral Participation — Lave & Wenger, Cambridge University Press (1991). ↩︎

  22. The Four-Phase Model of Interest Development — Suzanne Hidi & K. Ann Renninger, Educational Psychologist 41(2), 111–127 (2006, DOI 10.1207/s15326985ep4102_4). Four-stage interest development (situational → individual). ↩︎ ↩︎2

  23. Flow: The Psychology of Optimal Experience — Mihaly Csikszentmihalyi, Harper & Row (1990). ↩︎ ↩︎2 ↩︎3

  24. Consequences of individuals’ fit at work: A meta-analysis of person-job, person-organization, person-group, and person-supervisor fit — Kristof-Brown, Zimmerman & Johnson, Personnel Psychology 58, 281–342 (2005, DOI 10.1111/j.1744-6570.2005.00672.x). 172-study meta-analysis. Person-Job Fit with job satisfaction ρ = .56, organizational commitment ρ = .47, task performance ρ = .20. ↩︎ ↩︎2 ↩︎3 ↩︎4 ↩︎5

  25. Now, Discover Your Strengths — Marcus Buckingham & Donald O. Clifton, Free Press (2001). ↩︎

  26. Crafting a Job: Revisioning Employees as Active Crafters of Their Work — Amy Wrzesniewski & Jane E. Dutton, Academy of Management Review 26(2), 179–201 (2001, DOI 10.5465/amr.2001.4378011). ↩︎

  27. Bloomberg L.P. Case — Tokyo High Court, April 24, 2013. Capability-based dismissal after three rounds of PIPs ruled invalid. ↩︎

  28. Toa Paint Case — Supreme Court, July 14, 1986. Three-element test for the right to order reassignment. ↩︎

  29. Shiga Prefecture Social Welfare Council Case — Supreme Court, April 26, 2024. First-of-its-kind ruling that reassignment without consent is unlawful for employees with a contractually-limited job category. Commentary: TMI Associates. ↩︎ ↩︎2

  30. On the dimensionality of organizational justice: A construct validation of a measure — Jason A. Colquitt, Journal of Applied Psychology 86(3), 386–400 (2001, DOI 10.1037/0021-9010.86.3.386). Related: Justice at the Millennium: A Meta-Analytic Review of 25 Years of Organizational Justice Research — Colquitt et al., Journal of Applied Psychology 86(3) (2001). Meta-analysis of the four dimensions (distributive, procedural, interpersonal, informational). ↩︎

  31. Career Goal Beyond Department Head: The Choice Not to Pursue Management and Specialist Tracks — off.company (2020). Case study of Omron’s specialist track (introduced 2005, four tiers, capped by Fellow). ↩︎ ↩︎2

  32. Dual Ladders in Research: A Paradoxical Organizational Fix — Allen & Katz et al. ↩︎ ↩︎2

  33. Japanese Labor Market Institutional Reform: Problem-Awareness and Prescription Perspectives — Kotaro Tsuru, RIETI Discussion Paper 08-J-015 (2008). Related: Does Employment Protection Lower Productivity? — RIETI Discussion Paper 08-J-017. And On the OECD Employment Protection Indicators 2013 — JILPT. A one-unit change in EPL is associated with about a 0.02 percentage-point decline in annual labor-productivity growth and 0.04 in TFP (OECD 2007). Japanese employees with under one year of tenure: 7.3%. ↩︎ ↩︎2 ↩︎3 ↩︎4 ↩︎5

  34. What is Job-Type Employment Society? Contradictions and Turning Points of the Regular-Employee System — Keiichiro Hamaguchi, Iwanami Shinsho (2021). Related: The Truth About Job-Type Employment — Policy Research Institute lecture material (2022). Concepts of membership-type vs. job-type and the cyclical structure of redundancy-dismissal doctrine. ↩︎ ↩︎2

  35. The Peter Principle: Why Things Always Go Wrong — Laurence J. Peter & Raymond Hull, William Morrow (1969). Structural observation of promotion to the level of incompetence in hierarchies. ↩︎ ↩︎2

  36. Promotions and the Peter Principle — Alan Benson, Danielle Li & Kelly Shue, The Quarterly Journal of Economics 134(4), 2085–2134 (2019, DOI 10.1093/qje/qjz022). Empirical confirmation across 38,843 sales workers in 131 US firms. ↩︎ ↩︎2

  37. Japan Management and Organizational Practices Survey (JP MOPS) — Cabinet Office Economic and Social Research Institute. Japanese counterpart of Bloom & Van Reenen’s World Management Survey. Japanese firms’ management-practice scores: above-average on monitoring/target-setting, below-average on incentives. ↩︎

  38. The Happenstance Learning Theory — John D. Krumboltz, Journal of Career Assessment 17(2), 135–154 (2009, DOI 10.1177/1069072708328861). Planned-happenstance theory: five traits — curiosity, persistence, flexibility, optimism, risk-taking. ↩︎ ↩︎2

  39. The Strength of Weak Ties — Mark S. Granovetter, American Journal of Sociology 78(6), 1360–1380 (1973, DOI 10.1086/225469). ↩︎ ↩︎2

Career & Organization
Engineering Management Career Organization Playing Manager Japan AI Self-Determination Theory
This post is licensed under CC BY 4.0 by the author.