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

A recent whitepaper from Google emphasizes that in AI-assisted development, the model accounts for only 10% of system behavior. The main challenge and opportunity lie in harness design and context management, shifting focus from models to configuration and verification.

A Google whitepaper released in early 2026 states that the AI model used in coding agents accounts for only about 10% of the system’s behavior. The main focus for developers should be on harnesses and context engineering, which constitute the remaining 90%, marking a paradigm shift in AI-assisted software development.

The whitepaper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, argues that the common perception of models as the core of AI systems is misleading. Instead, the behavior of AI agents is predominantly shaped by their harnesses—the prompts, tools, rules, and observability layers built around the models.

Concrete examples from experiments show that tweaking the harness can significantly improve performance, sometimes by over 13 points, even when using the same underlying model. This indicates that the configuration and scaffolding are the primary levers for optimizing AI systems.

The whitepaper also emphasizes the importance of context engineering, which involves managing the information fed to the AI, including instructions, knowledge, examples, and guardrails. Proper context management can dramatically influence output quality and system robustness.

From an economic perspective, the paper warns that vibe coding—minimal prompts and quick fixes—may seem cheap initially but incurs higher long-term costs due to inefficiencies, security vulnerabilities, and maintenance burdens. In contrast, disciplined, agentic engineering involves higher upfront investment but lower marginal costs over time.

At a glance
reportWhen: published early 2026
The developmentThe Google whitepaper introduces a new framework for software development that prioritizes harnesses and context engineering over the AI model itself, marking a significant shift in AI-driven coding strategies.
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

Implications for AI Development Strategies

This shift means that development teams should focus more on designing and refining harnesses and context management rather than solely relying on the latest model versions. It highlights that competitive advantage in AI systems lies in configuration, verification, and process control, not just in accessing cutting-edge models.

For organizations, this redefines how to allocate resources, emphasizing scaffolding, testing, and security to ensure reliable, cost-effective AI deployment. It also suggests that future innovations may come more from system architecture improvements than from model improvements alone.

Ultimately, the whitepaper urges a strategic reevaluation, where the focus is on building durable, configurable AI systems that can adapt and improve over time without constantly chasing newer models.

AI-Powered Web Design Mastery: Harness the Power of Framer AI to Build, Customize, and Launch Stunning Websites—A Step-by-Step Guide for Beginners and Professionals

AI-Powered Web Design Mastery: Harness the Power of Framer AI to Build, Customize, and Launch Stunning Websites—A Step-by-Step Guide for Beginners and Professionals

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Background and Evolution of AI Coding Practices

Since the rise of AI coding agents, the industry has often equated model quality with system performance. Early assumptions suggested that the latest, most powerful models would deliver the best results. However, recent experiments and industry insights have shown that configuration, prompts, and scaffolding play a more critical role than previously thought.

The whitepaper builds on the ongoing shift from vibe coding—quick, unstructured prompts—to agentic engineering, which involves structured, verified, and maintainable AI systems. This evolution reflects a broader understanding that cost, security, and reliability depend heavily on how systems are built and managed, not just on the models themselves.

Earlier developments, such as the adoption of tools like LangChain and the emphasis on context management, laid the groundwork for this new perspective. The current focus is on making AI systems more controllable and cost-efficient.

“The biggest shift in software engineering isn’t a new language or framework; it’s moving from writing code to expressing intent and trusting machines to interpret it.”

— Addy Osmani

Beyond the Prompt: Spec-Driven Development and Context Engineering for AI Coding Agents

Beyond the Prompt: Spec-Driven Development and Context Engineering for AI Coding Agents

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Unanswered Questions About Long-Term Impact

While the whitepaper presents strong evidence that harnesses and context are dominant factors, it remains unclear how this approach will scale with future model advancements or how organizations will best implement these strategies at scale. The precise methods for standardizing and automating harness design are still evolving, and industry adoption timelines are uncertain.

Additionally, the long-term effects on AI development costs, security, and innovation pace are still being studied, making some aspects of this paradigm shift still uncertain.

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.

Next Steps for Developers and Organizations

Organizations should start reevaluating their AI development processes, prioritizing the design of robust harnesses and effective context management. Investing in tooling, testing frameworks, and security measures aligned with this new framework will be critical.

Further research and industry collaboration are expected to refine best practices for harness design and configuration. Monitoring how these strategies impact cost, security, and system reliability over time will be essential for adapting to this evolving landscape.

In the near term, expect more case studies and tools emerging to support structured AI system engineering, as the industry shifts focus from model chasing to system engineering excellence.

The AI Prompt Playbook: Master AI Prompt Engineering with 140 Ready-to-Use Templates for ChatGPT, Claude, Gemini & Copilot

The AI Prompt Playbook: Master AI Prompt Engineering with 140 Ready-to-Use Templates for ChatGPT, Claude, Gemini & Copilot

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?

According to the whitepaper, the model itself contributes only about 10% to the overall system behavior. The remaining 90% is shaped by the harnesses, including prompts, tools, rules, and observability layers that control how the model is used and how it responds.

What is the main takeaway for AI development teams?

The key insight is that configuration, scaffolding, and context management are more impactful than constantly upgrading to newer models. Focusing on harness design can lead to better performance and lower costs over time.

How does this shift affect AI project costs?

While vibe coding appears cheap initially, it often results in higher long-term costs due to inefficiencies, security vulnerabilities, and maintenance. Disciplined engineering with well-designed harnesses can reduce marginal costs and improve system stability.

Will this change how AI models are developed?

This perspective suggests that future improvements will focus more on system architecture, harnesses, and context management rather than solely on developing larger or more powerful models.

What should organizations do now?

Start reevaluating AI workflows to emphasize harness design, context management, and verification. Building expertise in these areas will be crucial for staying competitive and controlling costs.

Source: ThorstenMeyerAI.com

Nothing in this article is financial or investment advice. Cryptocurrency and precious-metal investments carry significant risk — do your own research and consider a licensed advisor.
You May Also Like

Building DApps With MEV‑Resistant Design

Many strategies exist for building MEV-resistant DApps, but understanding them is essential for ensuring fairness and security.

HBM Ate The Fab

High Bandwidth Memory (HBM) now accounts for a significant share of global memory supply, driving shortages in RAM and GPUs amid rising demand and manufacturing challenges.

The Twelve Real Complaints About AI Tools in 2026 — A Reddit, Twitter, and GitHub Synthesis

A comprehensive report on the top twelve user complaints about AI tools in 2026, based on Reddit, Twitter, and GitHub discussions, highlighting real-world friction points.