📊 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 Google whitepaper reveals that in AI-enabled software development, the model itself is only about 10% of the system’s behavior. The focus should be on harness design and context engineering, which constitute the majority of system effectiveness.
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 system behavior. The paper emphasizes that the harness and context engineering are where most of the control and value lie in AI-assisted development, marking a shift in how organizations should approach AI integration.
The whitepaper, titled The New SDLC With Vibe Coding, highlights that the dominant factor in AI system performance is not the underlying model but the configuration, prompts, tools, and rules surrounding it. Evidence from public benchmarks shows that changing only the harness can significantly improve an agent’s performance, even with the same model. For example, a team moved a coding agent from outside the Top 30 to the Top 5 by tweaking only the harness components, such as prompts and middleware.
The authors argue that most agent failures are configuration issues—missing tools, vague rules, or noise in the context—rather than model flaws. This shifts the strategic focus for developers and organizations toward building and owning their scaffolding, rather than relying solely on the latest models.
The paper also stresses the importance of context engineering, which involves providing the right instructions, knowledge, memory, examples, tools, and guardrails to the AI. A key architectural decision is whether to load static or dynamic context, impacting cost and flexibility. The authors introduce the concept of Agent Skills, which load procedural knowledge only when needed, enabling more scalable and efficient AI systems.
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 Harness and Context Are the Key to AI Success
This shift in focus from models to scaffolding and context management has major implications for cost, security, and performance. While vibe coding may seem cheap initially, it leads to higher long-term costs due to token inefficiency, maintenance, and vulnerabilities. Conversely, disciplined agentic engineering involves upfront investment in design and testing but yields lower marginal costs and more reliable outcomes.
For organizations, this means that competitive advantage will come from how well they design and control their AI environment, not just from adopting the latest model. The emphasis on configuration and context engineering redefines best practices in AI development and deployment.
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.
The Evolution of AI Development Strategies
Prior to this whitepaper, the common narrative centered on acquiring and deploying advanced AI models as the primary driver of performance. However, recent experiments and benchmarks demonstrate that tuning the harness and context can produce dramatic improvements even with the same model. This aligns with broader trends in software engineering, where configuration and scaffolding often determine success more than core algorithms.
The paper builds on the growing recognition that AI development is shifting from a focus on model innovation to system design and operational control. This evolution reflects the increasing maturity of AI tools and the need for disciplined practices to manage complexity and costs effectively.
“The behavior you experience in AI tools is dominated by scaffolding you can build, own, and improve, not by whichever frontier model is ahead this quarter.”
— Addy Osmani
UGREEN NAS DH2300 2-Bay Desktop NASync, Support Capacity 64TB (Diskless), Remote Access, AI Photo Album, Beginner Friendly System, 4GB RAM on Board,1GbE, 4K HDMI, Network Attached Storage(Diskless)
Entry-level NAS Personal Storage：UGREEN NAS DH2300 is your first and best NAS made easy. It is designed for…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Aspects of the New SDLC Are Still Unclear
It is not yet clear how widely organizations will adopt the recommended focus on harness and context engineering over the coming months. The specific methodologies for effectively scaling and managing large, complex AI systems remain under development. Additionally, the long-term impact on AI costs and security practices requires further observation as more organizations implement these principles.
Context Engineering for Claude: A Practitioner's Guide to CLAUDE.md, Memory Tools, and Three-Layer Workflows for Solopreneurs, Freelancers, and Product Managers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Developers and Organizations
Organizations should evaluate their current AI workflows, emphasizing building and owning their harnesses and context management systems. Future developments may include standardized frameworks and tools for scalable context engineering. Monitoring how these practices influence operational costs, security, and AI performance will be critical in the coming months.
Further research and case studies are expected to validate and refine these strategies, helping organizations transition from model-centric to system-centric AI development.
AI middleware development kit
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 infrastructure—prompts, tools, rules, and context—primarily determines how an AI system performs, making the model itself only about 10% of the overall behavior.
How can organizations improve their AI systems based on this insight?
By focusing on designing better harness components, such as prompts, middleware, and guardrails, and managing context effectively, organizations can significantly enhance AI performance without always upgrading the model.
Does this mean models are no longer important?
Models remain a foundational component, but the whitepaper emphasizes that their influence is limited compared to how they are integrated and controlled within the system. The focus shifts toward configuration and system design for better results.
What are the risks of focusing too much on harness and context?
Over-reliance on configuration without proper discipline could lead to security vulnerabilities, maintenance challenges, or inconsistent performance if not managed carefully. Proper standards and testing are essential.
Source: ThorstenMeyerAI.com
