The Model Is Only 10%: The Real Lesson of the New SDLC

📊 Full opportunity report: The Model Is Only 10%: The Real Lesson of the New SDLC on ThorstenMeyerAI.com — validation score, market gap, and execution plan.

TL;DR

The latest whitepaper from Google emphasizes that in AI-driven software development, the model accounts for only 10% of system behavior. The real expertise lies in designing the harness and context engineering, which determine performance and cost-efficiency.

A new whitepaper from Google emphasizes that the most significant shift in software engineering is moving from focusing on AI models to designing the surrounding system — the harness, context, and verification processes. This insight challenges the common perception that advances in AI models alone drive progress, highlighting instead that system engineering now plays the dominant role in AI development and deployment.

The paper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, reports that 85% of professional developers use AI coding agents regularly, with 51% using them daily. It states that approximately 41% of all new code is generated by AI. Despite this, the authors argue that the model itself accounts for only about 10% of the final system’s behavior, with the remaining 90% determined by the harness — the prompts, tools, rules, and observability layers surrounding the model.

The whitepaper emphasizes that failures in AI agents are often due to configuration issues such as missing tools, vague rules, or poor context management, rather than the model’s capabilities. It advocates for a shift in focus toward system design and context engineering as the key to building reliable, cost-effective AI systems.

At a glance
reportWhen: published early 2026
The developmentGoogle’s new whitepaper highlights that the core of modern AI software development is the system design around the model, not the model itself, marking a shift in SDLC practices.
The Model Is Only 10% — The New SDLC With Vibe Coding
AI Dispatch · Field Notes
Google · Osmani, Saboo & Kartakis · May 2026

The model is only 10%

A Google whitepaper argues software’s biggest shift is from writing code to expressing intent. Its sharpest claim: the model you obsess over is the smallest part of the system — the scaffolding around it does the real work.

A spectrum, not a binary — the differentiator is how outputs get verified
Vibe Coding
Casual prompts · “does it seem to work?” · disposable code · high risk
Structured AI-Assisted
Detailed prompts + constraints · manual testing · features in real codebases
Agentic Engineering
Formal specs · automated tests + evals + CI gates · production scale · low risk
Tests verify the deterministic; evals verify the rest. Without both, it’s vibe coding — however clever the prompt.
The idea worth building your strategy around
Agent = Model + Harness
~10%
HARNESS — prompts · tools · context · hooks · sandboxes · observability
MODEL~90% IS YOUR SURFACE AREA, NOT THE PROVIDER’S
Outside Top 30 → Top 5 on Terminal Bench 2.0 by changing only the harness — same model.
“Most agent failures, examined honestly, are configuration failures” — a missing tool, a vague rule, a noisy context.
The economics: it’s a token-cost problem (CapEx vs OpEx)
Vibe Coding
Low CapEx · High OpEx
Looks free, hides debt: token burn (fix-it loops), maintenance tax (AI spaghetti), security remediation. Crosses over to 3–10× more per feature.
Agentic Engineering
High CapEx · Low OpEx
Pay upfront (specs, evals, context), then ship cheaply. Levers: context engineering for first-pass success + intelligent model routing — cheap models for the easy work.
85%
of devs use AI coding agents (51% daily)
41%
of all new code is AI-generated
~90%
of agent behavior is the harness, not the model
+19%
longer on some tasks (METR) — verification is the cost
The read

The clearest map yet of how serious AI development works — and mostly tool-agnostic. But it’s a Google funnel: the concepts are neutral, the on-ramps point to Gemini, Jules & the ADK. If the harness is 90% and it’s yours, your moat and your costs both live there — so own your scaffolding, route across models, and remember: AI amplifies whatever engineering culture it lands in.

Source: Osmani, Saboo & Kartakis, “The New SDLC With Vibe Coding,” Google (May 2026). Figures are the paper’s own, incl. METR & LangChain. Analysis is the author’s.
thorstenmeyerai.com

Why System Design Outweighs Model Advancements

This shift matters because it redefines where organizations should invest resources in AI development. Instead of chasing the latest model, companies should prioritize building robust harnesses and context frameworks. This approach can significantly reduce costs, improve reliability, and create durable competitive advantages, as evidenced by experiments showing that tweaks to the harness can dramatically improve performance without changing the model itself.

LEAN PROGRAMMING FOR FORMAL SOFTWARE VERIFICATION: Mathematical proof systems and logical frameworks for verified computation

LEAN PROGRAMMING FOR FORMAL SOFTWARE VERIFICATION: Mathematical proof systems and logical frameworks for verified computation

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Evolution of AI Development Practices

Historically, AI progress was driven by improvements in model architectures and training data. However, recent developments indicate a growing recognition that system-level engineering—including prompt design, tool integration, and verification—has a larger impact on system behavior and cost management. The whitepaper builds on prior industry observations that configuration and context are critical for effective AI deployment, now framing it as the core skill in the AI era.

“The behavior you experience in AI tools is dominated by the scaffolding you build around the model, not the model itself.”

— Addy Osmani

Requirements Engineering for Software and Systems (Applied Software Engineering Series)

Requirements Engineering for Software and Systems (Applied Software Engineering Series)

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Unanswered Questions on Implementation and Costs

While the whitepaper presents compelling evidence that system design dominates model choice, it remains unclear how organizations will effectively scale this approach across diverse projects. Specific strategies for optimizing harness design, managing dynamic contexts, and balancing upfront costs with long-term savings are still developing. Additionally, the precise impact on security and maintenance costs requires further investigation.

Hands-On AI Engineering: Code First Guide to Building Production Grade LLM Systems with Python | Accompanied with GitHub Tutorials | Learn about Transformers Foundation Models & ML Pipelines

Hands-On AI Engineering: Code First Guide to Building Production Grade LLM Systems with Python | Accompanied with GitHub Tutorials | Learn about Transformers Foundation Models & ML Pipelines

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Next Steps for AI System Engineering

Organizations are expected to begin prioritizing system architecture, context management, and verification frameworks in their AI workflows. Future developments may include tools and standards for harness design, best practices for context engineering, and metrics to evaluate system robustness. Industry leaders will likely experiment with different configurations to optimize performance and cost-efficiency at scale.

The Complete Guide to AI Observability: Monitoring, Evaluating & Governing AI Systems in Production

The Complete Guide to AI Observability: Monitoring, Evaluating & Governing AI Systems in Production

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Key Questions

Why is the model only 10% of the system’s behavior?

The whitepaper shows that the surrounding system — prompts, tools, rules, and observability — largely determines how the model’s outputs are used and interpreted, making it the dominant factor in system behavior.

How does this shift affect AI development budgets?

Focusing on system design and verification may require higher upfront investments but can lead to lower ongoing costs, better reliability, and fewer security issues, ultimately reducing total cost of ownership.

What skills should teams develop to adapt to this new SDLC?

Teams should strengthen skills in system architecture, prompt engineering, context management, and verification processes, moving beyond model selection to system-level optimization.

Does this mean model improvements are no longer important?

Model improvements remain valuable, but the whitepaper suggests they are only part of the equation. System engineering and configuration now have a larger impact on performance and cost.

What are the risks of focusing on system design over models?

Overemphasizing system design without keeping pace with model advancements could limit potential performance gains. Balance is needed, but the whitepaper advocates for system-centric strategies as the primary focus.

Source: ThorstenMeyerAI.com

You May Also Like

Capital: The Lever Beneath the Levers

Analysis of how the flow and funding of capital shape AI development, highlighting recent public listings and systemic risks in the AI industry.

Évian and the Fallout: What Europe Actually Wants From Amodei, Hassabis, and Altman

European leaders press U.S. AI firms for reliable access, sovereignty, and safety measures amid U.S. export controls and geopolitical tensions.

The Door: Why the Interface Is Worth More Than the Model

SpaceX’s $60B purchase of a coding interface highlights the growing importance of interface ownership over AI models in distribution and control.

The Model Is Only 10%: The Real Lesson of the New SDLC

A new Google whitepaper emphasizes that in AI-driven software development, the model is just 10% of the system; the harness and context engineering are the key factors.