Software Build Costs Per Project: How Costs Dropped from $7,995 to $0 Over 9 Products

Key Takeaways

Every software project is treated as if it starts from zero.
Wright's Law predicts that cumulative production experience reduces per-unit cost by a fixed percentage with each doubling.
The CEM experience curve is driven by three compounding mechanisms that traditional team-based development cannot replicate.
If your cost model assumes that each new product costs approximately the same as the last, you are leaving the experience curve on the table.

Published: February 2026 | Stealth Labz — CEM Validation Portfolio Keywords: software development learning curve, decreasing build costs per project, AI development cost trajectory

The Setup

Every software project is treated as if it starts from zero. New architecture decisions. New technology evaluations. New team onboarding. New CI/CD pipeline configuration. The result is that the second product a company builds costs approximately the same as the first — and sometimes more, because the team has grown and the process has accrued overhead.

The conventional model assumes that software development costs are linear with scope. A project with twice the features costs roughly twice as much. A new product in a different vertical requires a new requirements phase, a new design phase, and often a new team or contractor relationship. There is no structural mechanism for the cost of project N to inform or reduce the cost of project N+1.

This is not how manufacturing works. Wright's Law — first described in 1936 by Theodore Wright in the context of airplane production — establishes that for every cumulative doubling of units produced, the cost per unit falls by a consistent percentage (typically 15-25%). The Boston Consulting Group documented this as the "experience curve" in the 1960s, showing that production costs decline 20-30% with each doubling of cumulative output across industries from semiconductors to chemicals.

Software development has historically resisted this curve because each project is treated as bespoke. The question is whether AI-assisted development with systematic reuse can bring Wright's Law dynamics to software.

What the Data Shows

External: Learning Curve Economics

Wright's Law predicts that cumulative production experience reduces per-unit cost by a fixed percentage with each doubling. In manufacturing, this has been validated across decades: semiconductor costs have followed an 80% learning curve (20% cost reduction per doubling) since the 1960s. Solar panel costs follow approximately an 80-85% curve. Battery costs have dropped 97% since 1991 along a consistent experience curve.

The Boston Consulting Group's experience curve data shows that the effect extends beyond manufacturing into any domain where cumulative repetition enables process optimization, tooling improvement, and knowledge retention. The critical factor is whether the organization captures and systematizes learning — or whether each iteration starts fresh.

IEEE studies on developer productivity improvement across project iterations (published 2018-2024) find that individual developers show 15-30% improvement in velocity between their first and third projects in a given technology stack. Team-level improvement is smaller (10-20%) because knowledge is distributed across individuals, and turnover resets the curve.

The gap between individual and team improvement matters. An individual developer retains 100% of their learning. A team retains whatever survives documentation, code reviews, and personnel changes.

Internal: The Per-Project Cost Curve

The PRJ-02 portfolio shows a per-project cost trajectory that follows — and then exceeds — Wright's Law predictions. Nine products were built between October 2025 and January 2026, with costs documented from CS08:

Sequence	Month	Per-Project Build Cost	Cumulative Reduction
Product 1	Oct 2025	$7,995	Baseline
Product 2	Oct 2025	$4,080	-49%
Product 3	Oct 2025	$4,005	-50%
Product 4	Nov 2025	$1,680	-79%
Product 5	Dec 2025	$330	-96%
Product 6	Jan 2026	$90	-99%
Product 7	Jan 2026	$0	-100%
Product 8	Jan 2026	$0	-100%
Product 9	Jan 2026	$0	-100%

The cost dropped from $7,995 to $0 across 9 products. The marginal cost of building a new production system reached zero by the seventh product.

Comparing to Wright's Law Predictions

An 80% learning curve (the semiconductor benchmark) would predict the following trajectory from a $7,995 starting point:

Doubling	Wright's Law Prediction (80%)	Actual CEM Cost
Product 1	$7,995	$7,995
Product 2	$6,396	$4,080
Product 4	$5,117	$1,680
Product 8	$4,094	$0

The actual cost curve is steeper than an 80% Wright's Law prediction. By product 4, the actual cost ($1,680) was 67% below the Wright's Law estimate ($5,117). By product 8, the actual cost was $0 — a result that Wright's Law does not predict because manufacturing always has material costs that create a floor.

Software has no material cost floor. When templates, architecture patterns, deployment configurations, and integration code are fully reusable, the marginal cost of assembling a new product from existing components approaches zero. This is the structural difference between software experience curves and manufacturing experience curves.

The Compounding Mechanism

The cost curve did not flatten gradually. It collapsed in distinct phases:

Phase 1: Foundation building (Products 1-3, October 2025). The first three products established the base architecture. Authentication, admin panels, payment integrations, and deployment patterns were built for the first time. Cost range: $4,005 to $7,995.

Phase 2: Template leverage (Product 4, November 2025). By the fourth product, the operator was reusing 60-80% of the codebase. The new code required was domain-specific logic — insurance quoting rules, product-specific UX. Infrastructure code was inherited. Cost: $1,680.

Phase 3: Assembly from components (Products 5-6, December 2025-January 2026). Products 5 and 6 were assembled almost entirely from existing templates with minimal new code. The operator's time shifted from building infrastructure to configuring existing components. Cost: $330 and $90.

Phase 4: Zero marginal cost (Products 7-9, January 2026). The seventh, eighth, and ninth products required no external contractor support and no new AI tool spend. The operator assembled them from the existing foundation. Template reuse exceeded 95%. Build cost: $0.

How It Works

The CEM experience curve is driven by three compounding mechanisms that traditional team-based development cannot replicate.

Single-operator knowledge retention. The operator — Michael George Keating — carries 100% of the architectural knowledge from every prior build. There is no knowledge transfer loss, no documentation dependency, no onboarding period. When the operator starts product 7, they hold the complete context of products 1 through 6 — every design decision, every integration pattern, every deployment configuration. IEEE research confirms that individual developer improvement (15-30% per iteration) exceeds team improvement (10-20%) because of this retention effect. CEM amplifies it by concentrating all learning in one person across 9 sequential builds.

Template reuse as capital accumulation. Each product adds reusable components to the foundation layer. By the ninth product, template reuse exceeded 95% — meaning less than 5% of the code was net-new. This is not copy-paste; it is systematic architecture where authentication modules, payment processors, admin interfaces, and deployment scripts are shared infrastructure that any new product inherits automatically.

AI tools accelerate the early curve. The most expensive phase is the foundation phase (products 1-3) where patterns are being established for the first time. AI tools — Claude and ChatGPT — accelerated this phase by generating scaffolding code, reducing the cost of establishing the initial templates. Once the templates existed, AI tool usage naturally declined (from $723/month in October to $0 in January), because the operator was assembling known components rather than generating new ones.

What This Means for Technical Leaders Planning Multi-Product Roadmaps

If your cost model assumes that each new product costs approximately the same as the last, you are leaving the experience curve on the table. The PRJ-02 data shows that per-project costs can follow a trajectory steeper than Wright's Law — from $7,995 to $0 across 9 products — when the execution model is designed for compounding.

The prerequisites are specific: single-operator knowledge retention (no team turnover diluting learning), systematic template reuse (not ad hoc copy-paste), and AI-assisted foundation building (to reduce the cost of the initial template creation phase). Organizations that distribute work across rotating teams, rebuild infrastructure from scratch for each project, or treat AI tools as individual productivity boosters rather than system-level accelerators will not see this curve.

For multi-product roadmaps — portfolios of 5+ related applications — the CEM cost curve means that the first product is the most expensive and every subsequent product is cheaper. Budgeting should reflect this: invest heavily in the foundation (product 1), expect 50% reduction by product 3, and plan for near-zero marginal cost by product 7 or later. The total portfolio cost is not (per-project cost x number of projects). It is the area under a declining curve.

References

Wright, T.P. (1936). "Factors Affecting the Cost of Airplanes." Learning curve cost reduction model.
Boston Consulting Group (1968). "Experience Curve." Cumulative production cost decline across industries.
IEEE (2018–2024). "Software Engineering Productivity Studies." Developer productivity improvement across project iterations.
Keating, M.G. (2026). "Case Study: The Cost Inversion." Stealth Labz. Read case study