📊 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 argues that the core of AI system success lies in harness and context engineering, not the AI model itself. This shift impacts how companies should allocate resources and design AI workflows.
A new Google whitepaper titled The New SDLC With Vibe Coding asserts that the most significant shift in software engineering is moving from writing code to expressing intent and trusting machines to interpret that intent. The key takeaway is that the AI model itself accounts for only about 10% of system behavior, with the remaining 90% determined by the harness and context engineering, a revelation that challenges common assumptions about AI development.
The whitepaper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, highlights that 85% of professional developers use AI coding agents regularly, with over half using them daily. It emphasizes that the real value lies not in the latest models but in how they are integrated and controlled through harnesses—tools, prompts, rules, and observability—comprising roughly 90% of the system’s effectiveness. Experiments cited in the paper demonstrate that changing only the harness or context can significantly improve AI performance, often more than upgrading the model itself.
The authors distinguish between ‘vibe coding’—quick, minimal prompts suitable for prototypes—and ‘agentic engineering,’ which involves structured, verified workflows with formal specs, tests, and oversight. They argue that the cost and complexity of AI systems are primarily driven by configuration, not the underlying models, and that organizations should focus on building durable, scalable harnesses and context strategies. Cost analysis indicates that ad-hoc prompting can be 3–10 times more expensive per feature than disciplined engineering approaches, especially over the long term.
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 and Investment
This shift means that organizations should prioritize developing robust harnesses and context management rather than solely chasing the latest AI models. Since the harness accounts for the majority of system behavior, investing in tooling, configuration, and context engineering offers a more sustainable competitive advantage. It also highlights that total cost of ownership in AI projects is heavily influenced by how well the system is structured and controlled, not just by the model’s capabilities.
AI harnessing tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background on the Evolving AI Development Paradigm
Prior to this whitepaper, the industry largely focused on acquiring and deploying advanced AI models, assuming that performance improvements came from model upgrades. However, as AI adoption increased—85% of developers using AI agents—the importance of how these models are integrated and managed became more apparent. The paper builds on recent experiments showing that tweaks to prompts and scaffolding can outperform model upgrades, emphasizing a shift from model-centric to system-centric thinking in AI engineering.
“The behavior you experience in AI systems is dominated by scaffolding you can build and control, not the model itself.”
— 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.
What Aspects of the New SDLC Are Still Unclear?
While the paper convincingly argues that harness and context are paramount, it does not specify precise best practices or standardized frameworks for building these components across different AI applications. Additionally, the long-term effects of this shift on AI model development and industry standards are still emerging, and more empirical data is needed to confirm the scalability of these insights across diverse use cases.
AI observability and monitoring tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Organizations and Developers
Organizations should reevaluate their AI development strategies, emphasizing investment in harness and context engineering. Developing standardized tools, frameworks, and best practices for managing AI system configuration will be critical. Further research and case studies are expected to validate and refine these approaches, potentially leading to industry-wide shifts in AI engineering practices.
structured AI development platforms
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 through experiments that most of an AI system’s effectiveness depends on how the model is integrated, controlled, and guided through harnesses and context—these components shape the output more than the model itself.
How should companies change their AI development approach?
Companies should focus on building robust harnesses, managing context effectively, and verifying outputs through structured testing, rather than just upgrading to the latest models.
Does this mean AI models are no longer important?
Models remain important, but their role is now seen as part of a larger system. The emphasis shifts toward how models are used, controlled, and integrated into workflows.
What are the economic implications of this shift?
Cost analysis suggests that investing in configuration and control (harness and context) can reduce long-term expenses, as ad-hoc prompting and model upgrades are more costly over time.
What remains uncertain about this new approach?
It is still unclear how best to standardize harness and context engineering across industries and what the long-term industry impacts will be as this approach becomes more widespread.
Source: ThorstenMeyerAI.com
