Complexity Wars 03: AI Won't Replace Programmers, but Organizational Structures Will

Forum topic · 小凯 · 2026-05-18

Summary

This analysis argues that AI will not replace programmers—instead, it will render large software organizations obsolete. Building on Fred Brooks' The Mythical Man-Month (1975) and Brooks' Law, the article explains why adding people to late projects makes them later: training costs, exponentially growing communication overhead (n(n-1)/2 interfaces), and sequential task constraints. Citing Balatro's solo developer (30 months, 5 million copies sold) and Base44's 3-person team acquired by Wix for $80M, it contends that AI's real revolution is not faster coding but enabling 3-person teams to match the output of 10-person teams, cutting communication interfaces from 45 to 3. The piece proposes a three-layer future organizational model—autonomous ~3-person Product Squads, 3-8 person internal Module Teams, and shared Infrastructure Teams—framed as a reverse application of Conway's Law. It concludes that organizational architecture, not code, becomes the competitive moat in the AI era.

One-line thesis: **The curse of Brooks' *The Mythical Man-Month* still holds in the AI era, but AI has changed one critical variable—it gives small teams the firepower that once only large teams had, fundamentally shaking the software industry's organizational logic.

Key points

Core claim: AI won't replace programmers, but organizational structures will. The bottleneck in software has never been writing code—it is communication.

Brooks' Law (1975): "Adding manpower to a late software project makes it later." IBM's OS/360 project showed how a thousand engineers got stuck in the tar pit: one bug fix bred more bugs.

Communication cost grows exponentially: a 10-person team has 45 communication interfaces (15x a 3-person team); a 50-person team has 1,225.

Conceptual integrity cannot be divided: Balatro was built by one solo developer in 30 months and sold 5 million copies. Ten top programmers could not rebuild it in 3 months, because its core asset is not code but the coherence of thousands of micro-design decisions.

AI's real impact: not "everyone codes 2x faster," but "3 people do the work of 10"—cutting interfaces from 45 to 3 and compressing alignment time (roughly: coding 1h + communication 7h = 8h, versus coding 1h + communication 2h = 3h).

Why small teams gain an unfair advantage

1. Feedback loop speed: decision → code → user feedback in hours, not sprints. 2. Controlled cognitive load: one person + AI tools (Cursor, Claude Code, v0, Lovable) replicates Brooks' "surgical team" (chief surgeon + assistants + toolsmith + documenter). 3. Conceptual integrity preserved: 3 people can share one unified mental model; at >5 people it decays sharply.

The three-layer future organization

1. Product Squad (~3 people): the minimal combat unit with end-to-end ownership, from idea to revenue. Examples: Base44 (3 founders, acquired by Wix for $80M in 6 months); solo AI developers. 2. Module Team (3–8 people): internal suppliers of shared capabilities (payments, auth, recommendations). Unlike traditional Chinese-style "middle platforms," they are API-first, multi-tenant, self-operating, and respond in days—not quarterly sprints. 3. Infrastructure Team:** servers, databases, compliance, security, deployment—the "air and water" every product needs but none builds itself.

Conway's Law in reverse

Melvin Conway (1967): *organizations design systems that mirror their communication structures.* Tree-shaped hierarchies produce monolithic systems with thick department walls. The three-layer model reverses the law: design the desired system architecture first (small autonomous products + shared modules + unified infrastructure), then build the organization to match. Product squads stay loosely coupled; module teams provide standardized API contracts; infrastructure teams underpin everyone.

Is AI the silver bullet? No—but something better

Brooks said in 1975 there is "no silver bullet." AI is not one either, because the bottleneck was never doing the work—it was coordination. AI didn't repeal Brooks' Law; it changed its scope of applicability. You no longer need a large team to build large products, so the curse dissolves on its own. This is not a technical advance but an organizational paradigm shift.

The deeper question: organization as moat

When everyone codes with AI, code itself stops being the differentiator. What remains:

Decision speed: 2 hours of alignment vs. 2 weeks of scheduling
Conceptual integrity: a unified mental model vs. committee design
User proximity: direct feedback vs. filtered requirement documents
Experiment density: three iterations a day vs. one release a week

Organizational architecture is shifting from a cost center to a competitive barrier. Future software companies will be measured not by headcount, but by how many independent, end-to-end product squads they can sustain.

References

Fred Brooks, *The Mythical Man-Month* (1975)
Melvin Conway, "How Do Committees Invent?" (1967, Conway's Law)
Balatro: developed solo by LocalThunk, 5M+ copies sold in 2024
Base44: 3-person team acquired by Wix for $80M in 6 months