MaGong

How History Teaches AI Scaling: Lessons from Zhu Rui

MaGong··AI Coding

How History Teaches AI Scaling: Lessons from Zhu Rui

Every AI Coding pilot ends the same way: "What's next?" The answer is usually something like "we need more budget," "we don't have executive support," or "we need more time to explore."

Sound familiar? These are fake problems.

Eighty years ago, Zhu Rui faced a real one. In September 1945, he arrived in Manchuria with nothing: no weapons, no budget, no infrastructure. He was appointed commander of artillery for the Fourth Field Army of the Chinese Communist forces. His order was simple: build a professional artillery corps.

He did something remarkable. He told his superior, Lin Biao: "Give me 500 experienced officers, budget, and supplies. In three months, I'll deploy four artillery regiments to the front lines."

This wasn't "let's explore." It was specific numbers (four regiments), a hard deadline (three months), and verifiable results (combat operations).

He did it. By December 1947, at the Battle of Changwu, he concentrated fire from 591-900 artillery pieces and killed over 6,000 enemy soldiers. By March 1948 at Four-Ping, 538 guns fired 500 rounds per minute for 23 hours and took the city. By end of 1948, the Fourth Field Army had 9,000 artillery pieces.

Three years later, from Manchuria to the southern islands, artillery was the decisive force in every major battle.

Here's what Zhu Rui teaches us: the problem was never resources. It was always courage.

Pseudo-Problem #1: "We Need More Budget"

The typical pitch: "Give us budget for licenses and training. Maybe we can improve productivity."

Zhu Rui's pitch: "Give me 500 officers, a command structure, supplies, and three months. I'll deploy four combat-ready regiments. If I fail, I resign."

See the difference? He didn't ask for permission to "explore." He committed to results. He staked his position on delivery.

In today's terms: Don't ask for budget to "try" AI coding. Commit to shipping deliverables.

What does that look like? Instead of "we'd like to try Claude Code," it's: "I'll take our 10 best engineers, full-time access to AI tools, and three months. Milestone 1: deliver two features with 3-5x faster velocity. Milestone 2: train ten more engineers. Milestone 3: deliver five more features. If I miss these, I step down."

The magic is specificity + accountability. With those two things, leadership can justify the investment up the chain.

Pseudo-Problem #2: "Train Everyone"

Common approach: Buy licenses for 1,000 developers. Run training sessions. Expect adoption.

Six months later: 15% actually using it. Conclusion: "They need more training."

Wrong approach entirely.

Zhu Rui didn't train all infantry officers to use artillery. Instead, he built a professional artillery corps. He placed 500 specialists throughout the army. They weren't teaching infantry officers—they were building artillery units with their own command structure, tactics, and culture.

Why? Infantry officers and artillery officers need completely different skills. Forcing infantry officers to learn artillery is inefficient. Better to specialize.

Apply this to AI coding: Not 1,000 developers, each dabbling. Instead, 10-20 full-time AI coding specialists. These aren't "people who use AI tools when they have time." This is their job. They do intensive bootcamp training (two weeks), become genuinely expert, then embed in product teams.

When do you concentrate them? When it matters. Zhu Rui had a principle: "broad general development, concentrated use at critical moments."

During the siege of Jinzhou, he concentrated 591-900 guns and broke the siege in 31 hours, killing 100,000+ enemy soldiers. During Tianjin, 538 guns in 29 hours killed 130,000+. Concentrated firepower at the decisive point.

Commander Zhu De later wrote: "Artillery was the main instrument of destroying enemy forces." Chairman Mao added in his 1949 New Year address: "Since the People's Liberation Army possessed more artillery and engineering troops than the Nationalists, the enemy's defense system, along with their aircraft and tanks, became insignificant."

This wasn't Zhu Rui's imagination. It was proven in practice.

For AI coding: Don't spread specialists thin. During a major product launch, concentrate all ten specialists on one team for six weeks. Not "AI everywhere," but "AI where it decides the outcome."

Pseudo-Problem #3: "We Need More Time to Explore"

Common refrain: "We need a year to explore AI coding best practices and build culture."

Six months later: still exploring.

Zhu Rui's timeline: September 1945 arrival → May 1946 (700+ guns collected, 500,000 rounds) → October 1946 (command structure, 100+ artillery companies) → December 1948 (9,000 guns).

Every stage had specific deliverables. Not "under development," not "still figuring it out"—delivered.

He worked in three-month cycles. Each cycle: set goal → execute → deliver → use results to justify next phase.

For AI adoption: Month 1-3 is proof-of-concept. Deliver concrete results: 3-5x velocity, measurable features shipped. Month 4-6 is center of excellence: training materials, documented best practices, methodology. Month 7-12 is scale: specific teams, specific numbers, expanded capacity.

Use each deliverable to fund the next phase. Not "we're still exploring," but "we proved 3x productivity last quarter, so we can now train five more teams."

Zhu Rui said: "Historical opportunity is fleeting. Seize it or lose it."

What was his window? 1945-1946. The Nationalist government hadn't yet moved military equipment, and Japanese supplies were still available in Manchuria. Miss that window, you miss the biggest equipment opportunity ever. Grab it, and you have an unstoppable force.

Is there an AI coding window in 2026? Yes.

Early adopters (starting in 2024-2025) are already 12-18 months ahead. By 2027, AI coding becomes standard practice—every developer uses it like GitHub. At that point, early adopter advantage evaporates because it's the baseline.

If you start now, you have 6-12 months of real advantage remaining. After that, it's table stakes, not differentiation.

Five Principles from Zhu Rui

1. Commit to numbers, not exploration. Not "we'll try," but "three months, four teams, X features shipped, personal accountability." Frame it as a deal: resources in exchange for delivered results.

2. Ask for what you need. Zhu Rui asked for 500 officers, command structure, supply budget—not because he was greedy, but because those were the inputs that produced the outputs he promised. You ask for specialists, tool licenses, time—whatever the math says you need.

3. Create mutual accountability. This is a two-way deal, not a one-way favor. Leadership provides resources and protection. You deliver results. Zhu Rui could ask Lin Biao for a lot because Lin Biao knew Zhu Rui would deliver. Build that mutual confidence, and everything else follows.

4. Build specialists, concentrate them at critical moments. Not mass training, but expert roles. Boring daily work, extraordinary impact when focused. Reserve the specialists for battles that matter.

5. Operate in three-month cycles with clear deliverables. Each cycle proves viability and funds the next. No vague "continuous exploration." Every phase answers: "What have we delivered? What do we need next?"

None of these are technical principles. They're all about organisation, leadership, and commitment.

2026: The Window Is Closing

There are three tiers:

Tier 1: Early movers. Started in 2024-2025. Already trained specialists. Already validated 5-10x productivity. Advantage locked in.

Tier 2: Fast followers. Starting now. Maybe six months of meaningful advantage left before it becomes standard. But you have to move fast.

Tier 3: Late arrivals. Starting 2027 onwards. AI coding is now table stakes. No competitive advantage.

Zhu Rui said: "The resources are scattered everywhere. First occupant wins. History's moment is fleeting. Not seizing it is wrong."

What's scattered everywhere in 2026? Claude Code, Cursor, GitHub Copilot—all ready. The tools are here. The window is closing.

If you don't move, what will your competitors look like in two years? They'll have trained 500 specialists. They'll have validated stable 10x productivity. They'll have their own best practices playbook. They'll know exactly which projects amplify AI impact. They'll have built seamless human-AI team dynamics.

By then, catching up isn't about "first-mover advantage." You're in a different competitive league entirely.

Maybe that's too stark. But Zhu Rui's case shows: some windows, once missed, don't reopen. His September 1945 decision shaped three years of war. Your 2026 decision might shape three years of your company's competitiveness.

The difference might just come down to whether you dare make the commitment.

References

Originally written in Chinese. Translated by the author.

← Back to all articles