📊 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 by Google emphasizes that in AI-assisted software development, the core value lies in harness design and context engineering, not the AI model itself. The model accounts for only 10% of behavior, shifting focus to configuration and verification.
A new whitepaper from Google, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, states that the AI model constitutes only about 10% of the overall system behavior in AI-assisted development. This challenges the common focus on model improvements and emphasizes the importance of harness design and context engineering, which account for the remaining 90%. This shift has significant implications for how companies invest in AI infrastructure and training.
The whitepaper underscores that the dominant factor influencing AI system performance is the harness — including prompts, tools, rules, and observability — which shapes the AI’s outputs more than the model itself. Experiments cited demonstrate that changing only the harness or the context can dramatically improve system behavior, often more than upgrading the model. For example, moving a coding agent from outside the top 30 to the top 5 involved solely tweaking the harness, not the model.
Furthermore, the paper introduces the concept of agentic engineering, where AI is embedded within a structured framework of verification, testing, and context management. This approach contrasts with vibe coding, which relies on minimal prompts and quick iterations. The authors argue that the economic benefits of disciplined harness and context engineering outweigh the low upfront costs, as it reduces long-term token expenses and security risks.
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.
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.
Why Focus on Harness and Context Matters in AI Development
This shift means that organizations should prioritize building robust harnesses and investing in context engineering rather than solely chasing the latest model improvements. Since the model is only a small part of the system, durable competitive advantages will come from how effectively teams configure, verify, and control their AI workflows. This realization could reshape AI development strategies and budget allocations across the industry.
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 of the SDLC Shift in AI-Assisted Development
The whitepaper builds on the recent surge in AI coding adoption, with 85% of developers now using AI agents regularly. It follows ongoing debates about model size versus system design, emphasizing that the industry’s focus has been skewed toward larger models. The authors argue that the real innovation lies in how developers harness and structure these models, not just in the models themselves. This perspective aligns with broader trends toward automation, verification, and cost efficiency in software engineering.
“The most impactful part of AI systems is the harness, not the model. Our focus should shift from chasing bigger models to building better scaffolding.”
— Addy Osmani
AI Engineering: Building Applications with Foundation Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unclear Aspects of Harness and Model Interaction
While the whitepaper provides strong evidence that harness design outweighs model size, it does not specify how quickly organizations can reorient their development practices or how universally applicable these findings are across different AI applications. The long-term impact of this shift on model innovation and hardware investment remains to be seen.
AI observability and verification tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for AI Development Strategies
Organizations are likely to reevaluate their AI workflows, investing more in harness development, context management, and verification tools. Industry leaders may publish best practices and frameworks to standardize this approach. Additionally, further research will explore how to optimize harness components for various domains and use cases, potentially leading to new tooling and training programs.
The Agentic AI Handbook: Design Patterns, Frameworks, and Tests for Real-World LLM-Powered Agents
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Why is the model size only 10% of the system behavior?
According to the whitepaper, the model’s output is heavily influenced by how it is configured, prompted, and integrated with tools and rules — the harness. This configuration determines most of the system’s behavior.
What is meant by ‘harness’ in AI systems?
The harness includes prompts, rules, tools, context policies, and observability mechanisms that surround and control the AI model, shaping its outputs and ensuring correctness.
How does this shift affect AI development costs?
While upfront costs for designing harnesses and frameworks may be higher, they significantly reduce ongoing token expenses, maintenance, and security risks, leading to lower total cost of ownership over time.
Will this change how AI models are built or improved?
The focus is shifting from model size and raw performance to system design, verification, and context management, which could influence future model training priorities and hardware investments.
Is this approach applicable to all AI applications?
The whitepaper suggests that the principles are broadly applicable, but the specific harness design will vary depending on the domain and use case. Ongoing research will clarify best practices for different scenarios.
Source: ThorstenMeyerAI.com
