📊 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 highlights that AI models constitute only about 10% of the system, with the majority of performance depending on harness and context engineering. This shifts focus from model selection to configuration, verification, and design. The development community needs to adapt accordingly.
A new whitepaper from Google, titled The New SDLC With Vibe Coding, states that the AI model itself accounts for only about 10% of the behavior in AI-driven software systems. This challenges the common focus on upgrading models and shifts attention to harness and context engineering, which comprise the remaining 90%. This insight has significant implications for how development teams should allocate resources and design AI systems.
The whitepaper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, emphasizes that the core of effective AI-assisted development lies in the harness—the prompts, tools, rules, and observability layers surrounding the model. Concrete experiments cited in the paper show that changing only the harness can dramatically improve agent performance, even with the same underlying model. For example, a team improved their coding agent’s ranking from outside the top 30 to within the top 5 by tweaking only the harness, not the model itself.
The paper also introduces the concept of context engineering, which involves carefully selecting and structuring the information fed into the AI. It highlights six types of context—instructions, knowledge, memory, examples, tools, and guardrails—and stresses that the quality of this context often outweighs prompt cleverness. An architectural choice—loading static versus dynamic context—further influences efficiency and scalability.
Furthermore, the whitepaper discusses economic considerations, noting that ad-hoc prompting can be 3–10 times more expensive per feature than disciplined, structured approaches. While vibe coding may seem cheaper initially, it incurs higher long-term costs due to token inefficiency, maintenance, and security vulnerabilities. Conversely, investing in harness design and context structuring offers lower marginal costs over time.
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.
Implications for AI Development Strategies
This shift in understanding fundamentally alters how organizations should approach AI integration. Instead of prioritizing the latest models, development teams should focus on building robust harnesses and refining context management. This approach enhances system reliability, reduces costs, and offers a durable competitive advantage. The insight underscores that trusting and controlling the environment around the model is more impactful than chasing the newest model versions, which may only represent 10% of the system’s success.
AI prompt engineering tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Evolution of AI-Assisted Software Engineering
The whitepaper builds on the rapid adoption of AI coding agents, with early 2026 reports indicating that 85% of professional developers use AI tools, and about 41% generate most of their code with AI assistance. Previously, the focus was on improving models and training data. Now, the emphasis is shifting towards system design, configuration, and verification.
This perspective aligns with ongoing industry discussions about the limitations of model upgrades alone and the importance of system architecture, testing, and security. The paper’s findings challenge the notion that better models alone will lead to better AI systems, emphasizing instead the importance of the surrounding infrastructure.
“The behavior you experience in AI tools is dominated by scaffolding you can build, own, and improve — not the model itself.”
— Addy Osmani
AI observability and monitoring software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Uncertainties in Implementing the New SDLC
It is not yet clear how widely organizations will adopt this paradigm shift or how quickly they will reallocate resources from model upgrades to harness and context engineering. The long-term impact on AI development costs and security practices remains to be fully understood. Additionally, the specific techniques for optimal harness design are still evolving, and industry standards have yet to emerge.
AI context management tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for AI Development Teams
Organizations should evaluate their current AI workflows and invest in developing robust harnesses and context management strategies. Further research and case studies are expected to clarify best practices and cost-benefit trade-offs. Industry standards and tooling for harness and context engineering are likely to develop, guiding best practices. Monitoring these developments will be crucial as the AI landscape continues to evolve.
AI development harness frameworks
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 explains that most of the AI system’s performance depends on how the model is integrated, configured, and guided—collectively called the harness—and the quality of the context provided. These factors shape the model’s outputs much more than the model itself.
How does this change the way companies should invest in AI?
Instead of focusing primarily on acquiring or upgrading models, companies should invest in designing better harnesses—prompts, tools, rules—and in managing context effectively. This approach offers more control, efficiency, and long-term value.
What are the economic implications of this shift?
While vibe coding appears cheaper initially, it often results in higher costs due to token inefficiency, maintenance, and security risks. A disciplined approach with structured harnesses and context management can reduce long-term expenses and improve system robustness.
Are there risks or challenges in adopting this new approach?
Implementing effective harness and context engineering requires expertise and initial investment. There is also a learning curve in designing scalable, secure, and adaptable systems, and industry standards are still emerging.
Source: ThorstenMeyerAI.com
