📊 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 in AI-assisted development, the actual AI model is only about 10% of what determines system behavior. The focus shifts to harness design and context engineering, which are more critical for success and cost-efficiency.
A new Google whitepaper titled The New SDLC With Vibe Coding emphasizes that the core of AI-driven software development is shifting from focusing on the AI model itself to the design of the harness and context. The paper states that the model accounts for only about 10% of system behavior, with the remaining 90% determined by how the AI is configured, guided, and integrated. This insight challenges prevalent assumptions and has significant implications for how organizations invest in and manage AI systems.
The whitepaper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, underscores that the dominant factor in AI system performance is the harness—which includes prompts, tools, rules, and observability—rather than the underlying model. Experiments cited show that changing only the harness or prompts can lead to dramatic improvements, such as moving an agent into the top ranks on benchmarks without changing the model itself. The authors argue that most failures in AI agents stem from configuration issues, not model limitations.
Furthermore, the paper introduces the concept of context engineering—carefully structuring the information provided to AI agents. Six types of context are identified: instructions, knowledge, memory, examples, tools, and guardrails. The authors highlight that effective context management, especially via dynamic loading, is vital for scalable, cost-effective AI deployment. They also emphasize that the total cost of AI systems is driven more by token economy and maintenance than by the initial model choice.
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.
Impact of Harness and Context on AI Development
This shift in focus from models to harness and context has profound implications for organizations deploying AI. It suggests that investments should prioritize configuration, tooling, and knowledge management over chasing the latest model versions. Cost efficiency and system reliability depend more on how AI is integrated and guided, which can be controlled and improved over time. This redefines the strategic approach to AI development, emphasizing long-term control and customization rather than model selection alone.
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.
Evolution of AI Development Strategies
The whitepaper builds on early 2026 data indicating that over 85% of professional developers use AI coding agents, with 51% doing so daily. The industry has largely viewed the model as the core driver of AI performance. However, recent experiments and analyses challenge this view, revealing that the majority of AI behavior is shaped by the surrounding scaffolding and context. This represents a paradigm shift from the traditional focus on model improvements to a broader system-level approach.
Prior to this, the dominant narrative emphasized model advancement—such as larger neural networks or more sophisticated architectures—as the key to progress. The new insight suggests that system design, configuration, and context management are now more critical for achieving reliable, cost-effective AI solutions.
“The model is only 10% of what determines behavior; the harness is 90%. The behavior you experience is dominated by scaffolding you can build, own, and improve.”
— Addy Osmani
Observability in the AI-Native Era: Leveraging AIOps to build, observe, and operate resilient systems
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unclear Aspects of Practical Implementation
While the whitepaper provides compelling evidence that harness and context are critical, it remains unclear how organizations will effectively scale these practices across diverse use cases. Specific best practices for dynamic context management and long-term maintenance are still evolving, and the precise cost benefits in different industries are not yet fully quantified.
YAML Made Simple: A Beginner’s Guide to Configuration and Data Structuring
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for AI System Optimization
Organizations are likely to focus on developing robust harnesses and context management tools, investing in training for system configuration, and establishing best practices for dynamic context loading. Further research and industry case studies will clarify how to best implement these principles at scale and measure their impact on cost and performance.
AI Context Engineering: Architecting Intelligence Through Prompt Structures, Tools, and Memory
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 system behavior?
The whitepaper argues that the model’s core capabilities are shaped and directed by the harness, prompts, and context, which account for the majority of behavior and outcomes.
How should organizations change their AI development approach?
They should prioritize designing effective harnesses, managing context carefully, and focusing on configuration and tooling rather than solely chasing new models.
What are the main benefits of this shift?
Improved cost efficiency, greater system reliability, and the ability to tailor AI behavior more precisely through system design.
Is this approach applicable to all AI systems?
While most evidence points to system design being crucial, the exact applicability depends on the specific use case and implementation context.
Source: ThorstenMeyerAI.com
