Society's Regeneration

A Generational Shift in Defense — From Large Platforms to Drones + AI

The trigger — extreme wartime necessity

What set off the generational shift in defense was not fossil resource prices, environmental policy, or shifting consumer preferences. It was the extreme necessity of war.

After the full-scale invasion in 2022, Ukraine faced a Russian military that overwhelmed it head-on in both quantity and quality. Buying F-35s and Patriots from the West to match numbers was impossible — the money was not there, and even if it had been, it would not have pushed the front line back. So Ukraine found a different answer: $50 AI autonomous-navigation modules × tens-of-thousands-of-dollars drones, $30,000 interceptor drones vs. $4 million Patriots, a domestic drone industry producing hundreds of thousands of units per month (Brave1), and battlefield feedback that turned into weapon updates within weeks (see Chapter 10).

While peacetime Lockheed Martin runs one cycle in 20 years, Brave1 runs hundreds of cycles via battlefield feedback. The "lose and you die" environment delivers an iteration speed that peacetime cannot.

A second proof — the 2026 Iran War

Ukraine's case might seem unique. But the 2026 Iran War repeats the same structure in a different context.

In March 2026, the US launched Operation Epic Fury and struck 12,300+ targets in five weeks — the largest such campaign in modern history. Yet by US intelligence assessment (April 2026):

• Missile launchers: ~50% intact
• One-way attack drones: thousands remaining
• Coastal-defense cruise missiles: mostly intact
• IRGC navy: hundreds to thousands of small boats and unmanned surface vessels remaining
• Strait of Hormuz: cannot be reopened

Iran's structural answer: decades-long underground tunnel networks, mass-produced Shahed/Geran-2 drones, distributed coastal missiles, small-boat fleets. Ukraine invented in a short time; Iran prepared over decades. Different means, same structure — expensive, concentrated, large platforms are denied by cheap, distributed alternatives.

Survivability of legacy weapons

"All legacy weapons disappear" oversimplifies. The decisive variables are mobility and concealability vs. dependence on fixed infrastructure. The first targets are fixed military bases (airfields, ports, supply depots).

Weapon category	Survivability	Why
Mobile ground-launched missiles	High	Can be dispersed, placed underground, and moved. Iran demonstrated this — about 50% remained after 5 weeks and 12,300 strikes
Coastal-defense missiles	High	Distributed deployment; sustains the ability to close a strait (Iranian case)
Fighter aircraft	Unclear	The aircraft itself is useful in combat, but dedicated airfields are fixed targets. If they cannot take off and land, they do not function. Underground hangars and dispersal airfields mitigate, but cannot fully escape, this exposure
Tanks and armored vehicles	Low	Large and expensive; cheap drone strikes break the economics. Demonstrated in Ukraine
Aircraft carriers	Very low	Huge and concentrated. With USVs combined with long-range missiles, multibillion-dollar carriers can be sunk by tens-of-millions-of-dollars assets — a brutal asymmetry
Surface combatants (destroyers, frigates, etc.)	Low to medium	Concentrated, vulnerable to anti-ship missiles and USVs (Black Sea Fleet)
Heavy transport / patrol aircraft	Low	Depend on fixed bases, with long airborne dwell times that expose them
Drones / AI-autonomous weapons	High	Dispersed, mass-produced cheaply, evolve on the battlefield
Electronic warfare systems	Medium	Mobile / distributed systems survive; large fixed systems are targets

In short, mobile / dispersed / underground / mass-produced weapons survive; concentrated / fixed / scarce / expensive weapons are culled. Within the legacy primes, mobile missile, mobile EW, and underground-hardening construction lines have room to survive. Carrier shipbuilders, 5th-generation fighter prime contractors, heavy transport / patrol aircraft, and large fixed-system makers enter a structurally hard era.

Rise of the new defense industry — "who can have a defense industry" is being rewritten

In parallel with the legacy contraction, a new defense industry built around drones + AI rises. Ukraine's Brave1: 300+ AI projects, 70+ in active battlefield use, 100+ companies, 2026 production target 7 million drones/year, AI modules ~$50 per unit.

Legacy defense industry	New defense industry
Lockheed Martin (F-35, missiles)	Anduril (autonomous drones, AI defense SaaS)
BAE Systems, Northrop Grumman	Helsing (European AI defense — Germany, UK, France)
Raytheon (Patriot, missile defense)	Quantum-Systems, Skydio, Shield AI
Japanese heavy industry (MHI, Kawasaki, IHI)	Saker, the 100+ companies in Ukraine's Brave1
Expensive single platforms (billions per unit)	Cheap distributed drones (tens of thousands to a few million yen)
Decade-long development cycles	Improved every few weeks via battlefield feedback

Legacy primes were a game only great powers — with multi-trillion-yen fighter-development budgets — could play. The new industry's basic units are $50 AI modules and tens-of-thousands-of-dollars drones, so middle powers can sustain it as a domestic industry. This is the industrial foundation of the "security provider" pivot (Chapter 10).

The most important variable — two-stage adaptation

The technical and industrial change does not implement itself. What actually happens depends on two stages of adaptation:

1. Can politics and the military adapt to technological change? — revise procurement rules, fund AI-native defense startups, rewrite officer education and doctrine, decouple military identity from legacy platforms
2. Can domestic industry adapt to political/military change? — legacy primes are slow due to organizational inertia, new entrants need talent / capital / regulatory environment / clearances / procurement precedents, investment networks (American Dynamism, European defense-tech VCs, Rakuten × Brave1) act as the seedbed

Ukraine's existential war pressure moved all of this at once. Brave1 is a state program that synchronizes politics, military, and industry. In the US and Europe, Anduril and Helsing went through long political battles before breaking into procurement. Synchronization not done in peacetime cannot be done in time during war. Brave1 cannot stand up overnight.

Failure of the great powers — the US and Russia

The US and Russia are the canonical examples where the two-stage adaptation is not working. Political systems are opposite, but both fail to adapt precisely because they are great powers.

US — political gravity of legacy defense plus an unbound executive: AI defense startups (Anduril, Shield AI, Skydio), American Dynamism, Replicator Initiative — the individual ingredients are in place. But the world's largest legacy primes (Lockheed Martin, Raytheon, Northrop Grumman, Boeing, General Dynamics) act as continuous political gravity that slows implementation: F-35 lifecycle cost over $1.7 trillion, employment scattered across congressional districts, revolving doors of retired generals, lobbying — the classical entrenched-interest failure. On top of that, the Trump administration since 2025 has exercised presidential power ad hoc, beyond the checks of Congress, the courts, and the federal bureaucracy, so the self-correcting mechanism of democracy no longer functions — tariffs, immigration, science budgets, and federal personnel get manipulated on short cycles, and long-term procurement reform and industrial-policy continuity are lost. The fact that 12,300 strikes in the 2026 Iran War left half of Iran's capability intact reads as the political gravity of entrenched interests × ad hoc executive decision-making — a double failure.

Russia — no bottom-up innovation under authoritarianism: lost most of its Black Sea Fleet to Magura USVs, lost tanks and armor in droves to FPV drones, became dependent on imported Iranian Shaheds. Tactical insights from soldiers don't reach leadership, state-owned defense primes leave no room for startups, failures cannot be acknowledged (fear of purges), and the Brave1 / Army+ / E-Points-style battlefield → manufacturer feedback loop cannot exist in principle. Mass production is possible but evolution is slow.

When faced head-on by smaller adversaries (Iran, Ukraine), neither can convert quantitative or qualitative advantage into combat power.

Being a great power is not a sufficient condition to win the generational shift. If anything, being a great power prevents adaptation — the most striking fact this chapter surfaces about the late 2020s. Middle powers that can run the two-stage adaptation quickly become the protagonists of the new security era.

The same structure repeats in IT and desk work. But AI is much more powerful, wider in scope, and more complex than defense.

IT — Pressure on Legacy IT, and the Rise of AI-Native Companies

The IT industry is the next to be re-formed. As with defense, this is not decline but generational shift. Look at it as two distinct things: the reality already in motion — pressure on legacy IT — and beyond that, the structurally predictable rise of AI-native companies.

Reality — pressure starts to build on legacy IT

As of 2024–2026, legacy IT incumbents still look strong on the surface. AWS / Azure / GCP cloud revenue, Oracle / Microsoft license revenue, the SIer business — none of it has entered an outright decline.

But structural pressure is building from several directions:

• AI CapEx economics — whether multi-hundred-billion-dollar data-center investments will earn their return is undecided. This series treats it as a structural forecast in Chapter 6.
• GPU market — NVIDIA's monopoly is being eroded gradually by AMD, custom ASICs, and OSS inference (Chapter 7).
• OSS DB migration — moves from Oracle to PostgreSQL and similar databases are growing in new projects.
• Pushback against "license tax" — rising Office / Azure pricing is producing real interest in alternatives (Chapter 8).
• Pressure on SIer rates — productivity gains from Cursor + Claude are starting to affect the unit-rate negotiations of the SIer industry.

Structural prediction — AI-native rises, but splits into two layers

While pressure builds on legacy IT, a new generation — designed with AI at the core from line one — is rising. Lumping them together misreads the structure: they split into two layers.

Layer 1 — model providers (Anthropic, OpenAI, etc.) The companies building the LLMs themselves. They deliver coding, research, and conversational capability directly (Claude Code, ChatGPT).

Layer 2 — cloud-based wrappers / hosting (Cursor, Vercel, Hugging Face, Perplexity, etc.) Companies that lay an IDE, hosting, model serving, or search UI on top of an LLM. These are likely transitional. As Anthropic's Claude Code delivers coding directly and the model providers absorb hosting, serving, and search UI in-house, the wrapper layer's reason to exist gets eaten away.

Legacy IT	Layer 1 (model providers)	Layer 2 (transitional wrappers)
Microsoft / Google / Apple	Anthropic (Claude)	OpenAI's wrapper products
Atlassian, Salesforce	(Claude Code, directly)	Cursor (AI IDE)
AWS / GCP legacy hosting	(provider, in-house)	Vercel (AI SDK, v0)
Oracle, SAP	(provider, in-house)	Hugging Face, Replicate
Bloomberg, Reuters	(provider, in-house)	Perplexity, Glean

What survives long term is model providers and individuals / small organizations using the model directly. The cloud-wrapper layer between them gets eaten away. The structural foundation of the "one person + AI" business model from Chapter 9 is here.

AI in Desk Work — Reasoning Structurally From the Current Reality

The generational shifts in defense and IT play out at the level of companies and nations. The broader impact happens inside every company — the AI transformation of desk work overall.

Current state — the AI agent race and large data-center investment

In 2024–2026 the AI industry is running two trends in parallel.

• The AI agent product race — vendors are releasing, one after another, products that minimize human involvement and have AI execute tasks (Copilot Workspace, Devin, ChatGPT Operator, Claude Computer Use, AutoGen / CrewAI, etc.)
• Large-scale data-center investment — OpenAI / Microsoft Stargate, Anthropic / AWS Project Rainier, xAI Colossus, Google / Meta AI cluster expansion, UAE Stargate, Saudi HUMAIN. All sized assuming autonomous agents are operated at scale

Reasoning structurally — autonomous agents cannot do the work

Autonomous AI agents grew out of the wish to "remove the human work" — and they carry structural problems.

Two recognitions the industry encountered in implementation:

1. AI doing only pure execution still leaves instructions, verification, environmental adaptation, and final judgment with humans — hallucinations, environmental change beyond training, and silent failure on edge cases all propagate exponentially under AI's mass processing
2. Supply on the human side is limited — domain expertise, critical reading, and judgment-experience accumulate over years to a decade, and not everyone can acquire them (cannot); plus a substantial number do not want to carry always-on accountability (does not want to)

The industry's answer — "let agents do verification and judgment autonomously" — is recognizable as wish, but it fails structurally:

• Verifier and executor are the same AI from the same training data — hallucinations, blindness to environmental change, and silent edge-case failure appear on both sides
• And it adds new problems — accountability dissolves, API call counts inflate 100×–1,000× (cost collapse, same structure as AGI unprofitability in Chapter 6), and prompt injection turns authorized credentials into a weapon (Chapter 5 Mythos)
• Anthropic itself warns "do not run in autonomous mode" (/en/ai-native-ways/ai-delegation/)

Autonomous agents are not "a technology that solves the labor shortage." They are a different set of problems made to look like a solution to the labor shortage.

Natural selection

The structural failure of autonomous agents does not bankrupt the whole industry at once. Natural selection proceeds.

• Companies that pushed in — cannot sustain themselves under the cumulative weight of cost collapse, surveillance hell, cyber-attack vulnerability, and dissolved accountability (AI-industry companies are no exception)
• Companies that stop and adapt — many existing companies choose this. Mid-sized, regional, and family-run companies, where the top decides directly, often adapt faster than large corporations
• Companies that keep pushing — autocratic management, weak governance, cannot stop (a structural risk)
• The Anthropic pattern — model providers that survive and grow — human-in-the-loop by design (Constitutional AI, Responsible Scaling Policy, Claude Code). Coding, research, and conversation delivered directly
• Transitional wrapper-type AI-native companies — Cursor, Vercel, Hugging Face, Perplexity. The wrapper layer on top of LLMs gets eaten away as model providers absorb its function in-house
• The long-surviving one-person + AI — individuals and small organizations connecting directly to model providers, no wrapper required (Chapter 9)

This is not limited to the AI industry. Existing non-AI industries trying to adopt AI in accounting, legal, sales, and customer support follow the same structure. The generational shifts in defense and IT in the first half of this chapter are the leading examples.

Not "everything goes AI" and not "everything fails." Change progresses as the process of redrawing the line between autonomy and humans. Same structure as Christensen's Innovator's Dilemma — not every existing company collapses; what is culled is companies that could not adapt to change.

So migration to land-based work becomes necessary

At the level of the whole society, demand for physical work rises rapidly: bio-material manufacturing (Chapter 2), soil regeneration and microbial management (regenerative agriculture), food production, forestry, regional infrastructure maintenance. None of this can be replaced by AI; it is the real receiving capacity for people freed from pure execution.

But migration does not happen automatically. Constraints of age, health, and geography; skill acquisition, relocation, and securing land take years; livelihood guarantees, vocational training, and farmland-forestland redistribution all require policy support (next section, "the great population migration").

UBI is not a structural solution (Chapter 6). Work does not disappear; it changes — those who can change will change, those who cannot need a different path (land-based work).

From Megacities to Land — The Great Population Migration

As the structural changes above unfold, the relationship between people and place is rewritten. The new industries do not require concentration in megacities: the new defense industry runs on regionally distributed manufacturing, AI-native companies on remote work and small head offices, and as OSS AI spreads AI itself becomes infrastructure accessible from anywhere.

And the work that grows in the post-fossil society — bio-material manufacturing, soil regeneration, food production, forestry, regional infrastructure — needs land. None of it can be done inside a Tokyo office building.

	Present	After transformation
Greater Tokyo	36M commuting to offices	No reason to concentrate
Regional towns	Depopulation accelerating	Function as the receiving capacity
Primary employment	Desk work, IT, service industry	Bio-material manufacturing, food production, forestry, soil regeneration
Required resources	Offices, data centers, electricity	Farmland, forestland, water, sunlight, soil

Rural life itself does not change much. Shops, schools, clinics, surrounding farmland and forestland — the receiving capacity is already there. What changes is that Tokyo's extreme concentration dissolves.

But having the receiving capacity does not mean people move on their own. Policy support is needed: farmland and forestland acquisition, vocational training, vacant-house utilization for migrants, fiscal support for rural clinics / schools / shops, practice fields for hands-on cultivation.

The most critical policy is rethinking free trade. Cheap oil → cheap transportation → produce abroad and consume in Japan was a model that rested on fossil-resource dependency. Even if Japan builds a domestic system for bio-materials, food, and timber, free trade as is will crush it under cheap imports — and land-based work cannot survive. As fossil resources deplete, transportation costs eventually rise, but if the domestic production base is destroyed before then, it is too late.

Rural depopulation is not "a problem to be solved" — it becomes a result solved by industrial transformation. Policies are needed to make that transition smooth.

Healthcare and Pensions Do Not Fit the New Society

When population disperses and work becomes land-based, current healthcare and pension systems are fundamentally mismatched. They were designed for an era of fossil resources, urban concentration, and salaried desk work. That era is ending — only the systems remain.

Shifting toward preventive medicine — Cuba's achievement and its limits

Re-organize medicine from "treatment-centered" to "prevention-centered" and the resources required drop sharply. Cuba — at GDP per capita around $10,000 (roughly a quarter of Japan's) — has sustained life expectancy 78.1 years (close to the US's 79.25), under-5 mortality 0.8%, and stunting (under 5) 7.1% (well below the LAC average of 11.3%).

The mechanism: the Family Doctor Program (founded 1983, 40 years old in 2024), where a doctor-and-nurse pair continuously cares for ~600–700 residents, sitting on a three-layer structure — 11,548 family-doctor offices → 451 polyclinics → hospitals — that resolves most problems at the lowest layer. Doctor density is 9 per 1,000 (more than twice the LAC average). 100% of births attended by skilled medical personnel. In 2026, former President Trump publicly said the US "should learn from Cuba's system" for its primary-care provider shortage.

But Cuba's system is now collapsing. Under the US fuel blockade and financial sanctions, blackouts have become routine, neonatal incubators and ventilators stop, and 70% of the 651 essential medicines have disappeared from pharmacies. Infant mortality rose from 4.0 in 2018 to 9.9 in 2025 — a 148% increase (CEPR estimates ~1,800 newborn lives could have been saved 2019–2025 had the 2018 rate held). In a single year (2021–2022), 46,000 healthcare workers, including 12,000+ doctors, left the country.

The lesson is two-sided:

A prevention-centered design delivers large outcomes from limited resources. But unless the electricity, clean water, basic medicines, and the livelihoods of medical workers that make it run are secured, no design — however excellent — can function.

Social insurance at 30% — the working generation's limit

Across all social security: health insurance ~10% + pension ~18.3% + long-term care ~1.8% + employment insurance ~0.9% = ~31% total disappears from working-age wages. The ratio rises every year; with continued demographic decline it heads toward 40–50%.

The pension system — a relic of the desk-work era

The "work until 65 → live on pension" model rested on two premises: retirement is short (1960s: life expectancy 65–67, pension start 55, retirement ~10 years) and population keeps growing (working-age is always the majority). Both have collapsed — life expectancy 84, pension start 65, retirement 20 years; advanced medicine drives elderly healthcare costs higher. Fewer workers, more retirees → mathematical collapse.

A Very Different World Must Be Designed

Almost every premise of current society collapses — fossil resources (Chapter 2), chemical-fertilizer dependency (Chapter 3), the fusion-materials problem (Chapter 5), the generational shift in defense, IT, and desk work (the first half of this chapter), the premise of extreme Tokyo concentration, the premise of free trade, urban- salaried-worker healthcare and pensions. Incremental fixes will not do. A world quite different from today must be designed.

Where work happens changes. Where people live changes. Trade premises change. The design premises of healthcare and pensions change. These are not isolated phenomena — they are the reorganization of a social structure whose two supporting pillars (fossil resources and urban concentration) are both giving way.

The next chapter, "Subtraction Design," organizes structurally which premises of the current society are being subtracted.