📊 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 reveals that in AI-assisted software development, the model itself is only about 10% of system behavior. The majority depends on how developers design the harness and context, shifting focus from models to configuration and verification.
A new whitepaper from Google, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, states that the AI model accounts for only about 10% of the behavior in AI-driven software systems. The more significant factors are the harness and context engineering, which determine how the model is used and how its outputs are verified. This shifts the traditional focus on models to the importance of configuration, scaffolding, and verification processes.
The whitepaper emphasizes that the dominant part of an AI agent’s behavior lies outside the core model, which is often mistakenly viewed as the primary driver. Instead, the harness — including prompts, tools, rules, and observability — makes up roughly 90% of the effective system performance. Evidence from benchmarks shows that changing the harness can dramatically improve AI agent outcomes without switching models, highlighting the importance of configuration over raw model power.
The authors introduce the concept of context engineering as a core discipline, involving the strategic loading of instructions, knowledge, memory, and tools. They argue that the key to scalable and cost-effective AI development is designing dynamic, on-demand context loading rather than static, always-loaded information, which impacts both performance and costs.
Economically, the paper notes that while vibe coding appears cheap initially, it incurs high long-term costs due to token wastage, maintenance, and security vulnerabilities. Conversely, disciplined engineering with well-structured schemas and verification can reduce total costs, despite higher upfront investments.
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 the 10% Model Is Not the Main Focus
This shift in understanding has significant implications for AI development strategies. It suggests that investing in better harness design, configuration, and verification processes can yield more substantial and cost-effective improvements than simply adopting newer or larger models. For organizations, this means rethinking resource allocation, emphasizing configuration and testing to maximize AI system reliability and efficiency.
Moreover, it challenges the common narrative that the latest model innovations are the primary source of AI progress, redirecting attention to the surrounding infrastructure and engineering practices. This could influence how companies plan their AI roadmaps and allocate budgets for model development versus system engineering.
AI prompt engineering tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Evolution of AI Development Practices and the SDLC Shift
The new whitepaper builds on the ongoing evolution of AI development, where the focus has shifted from solely improving models to optimizing how they are integrated into systems. As of early 2026, AI adoption is widespread, with 85% of professional developers using AI coding agents regularly, and 41% generating code via AI daily, according to industry reports.
Previous trends emphasized model improvements, but recent experiments and benchmarks demonstrate that configuration and scaffolding can dramatically enhance outcomes. This aligns with broader industry observations that system design and verification are now central to AI success, especially as models become more commoditized.
The paper also contextualizes this within the broader SDLC (Software Development Life Cycle), advocating for a disciplined, structured approach to AI system engineering that prioritizes verification, guardrails, and context management over raw model capabilities.
“The model accounts for only about 10% of the behavior; the rest is the harness and context engineering.”
— Addy Osmani
AI system verification software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Aspects of the Harness and Context Are Still Unclear
While the whitepaper provides strong evidence that harness and context engineering dominate system behavior, it does not specify precise best practices or standardized frameworks for implementation. The exact methods for optimal context loading, the impact across different AI tasks, and how to measure success in harness design remain areas for further research and experimentation.
Additionally, the long-term effects of this shift on AI model development, industry standards, and organizational workflows are still evolving and not yet fully understood.
AI harness configuration tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for AI Engineering and Industry Adoption
Organizations are likely to prioritize investing in harness and context engineering, developing better tools, and establishing standards for verification and configuration. Further research and case studies will clarify best practices and quantify cost savings. Industry leaders may also explore new training programs focused on system configuration and verification skills, shifting the AI development paradigm further toward disciplined engineering.
Monitoring how these insights influence AI product development and organizational workflows in the coming months will be critical to understanding their full impact.
context engineering software for AI
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 behavior?
The whitepaper argues that most of an AI agent’s behavior depends on how the system is configured, including prompts, tools, rules, and verification processes, which collectively shape the output more than the model itself.
How does this shift affect AI development costs?
While initial investments in harness design and context engineering may be higher, they lead to lower long-term costs through improved efficiency, reduced token waste, and better security, making AI development more cost-effective overall.
What practical steps can organizations take now?
Organizations should focus on designing flexible, dynamic context loading, establishing verification protocols, and building scaffolding that can be owned and improved over time, rather than relying solely on model upgrades.
Does this mean models are becoming less important?
Models remain essential as the core engine, but their role is now understood as a component within a larger system where configuration and verification play a more significant role in determining outcomes.
Will this change the future of AI research?
Yes, future research is likely to focus more on system engineering, verification, and context management, rather than solely on developing larger or more sophisticated models.
Source: ThorstenMeyerAI.com
