📊 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-driven software development, the model accounts for only 10% of system behavior. The focus should shift to harness and context engineering, which are responsible for 90%. This insight challenges traditional views and impacts how teams should invest in AI development.
Google’s new whitepaper titled The New SDLC With Vibe Coding emphasizes that the most significant shift in software engineering is moving from focusing on the AI model itself to optimizing the surrounding harness and context. The paper states that the model accounts for only about 10% of a system’s behavior, with the remaining 90% determined by configuration, prompts, tools, and context management. This challenges common assumptions about AI development and suggests a strategic pivot for engineering teams.
The whitepaper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, reports that as of early 2026, 85% of professional developers use AI coding agents regularly, and 41% of all new code is generated by AI. Despite the focus on models like GPT or Claude, the authors argue that the ‘harness’—the prompts, rules, tools, and observability—determines system behavior far more than the model itself.
Concrete evidence is provided through experiments: one team improved an agent’s performance from outside the Top 30 to Top 5 by only changing the harness, using the same model. Similarly, a LangChain experiment increased scores by tweaking prompts and middleware, not the underlying model. The paper stresses that failures in AI agents often stem from misconfiguration, missing tools, or poor context management, making the harness the primary surface for optimization.
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 focus from model to harness and context has major implications for how organizations allocate resources and design AI systems. It suggests that investing in better configuration, tooling, and context management can yield more reliable and cost-effective AI performance than constantly chasing the latest model upgrades. The insight also redefines the competitive advantage, emphasizing durable, configurable scaffolding over fleeting model improvements.
AI Engineering: Building Applications with Foundation Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background on AI System Design and Evolving Practices
Historically, AI development has centered on acquiring and deploying the most advanced models, driven by the belief that better models produce better results. However, recent experiments and industry reports, including this whitepaper, indicate that the surrounding infrastructure—prompts, rules, tools, and context—plays a far more critical role. This perspective aligns with broader trends in software engineering, where configuration and architecture often outweigh raw algorithmic power.
Prior to this, vibe coding—giving minimal prompts and accepting AI outputs—was prevalent for quick prototypes. The whitepaper introduces a spectrum, from vibe coding to disciplined ‘agentic engineering,’ where formal specs, testing, and oversight ensure correctness. The key insight is that the model is only a small part of this larger system.
“The model you’re paying so much attention to is the smallest part of the system. The behavior is dominated by the harness and context.”
— 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 Implementation and Industry Adoption
While the whitepaper presents strong experimental evidence, it is still unclear how widely organizations will adopt this focus on harness and context engineering. The practical challenges of restructuring existing workflows and the cost of developing sophisticated scaffolding are still being evaluated. Additionally, the long-term impact of this paradigm shift on AI model development and innovation remains to be seen.
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 Teams and Industry Adoption
Organizations are likely to reevaluate their AI development strategies, investing more in configuration, tooling, and context management. Future research and case studies are expected to explore how this shift affects system performance, cost-efficiency, and security. Industry leaders may also develop new standards and best practices focused on harness design and context engineering.
Agentic Development: The Complete Guide to AI-Assisted Coding with Claude, Cursor, and Beyond
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 argues that the model’s output is heavily influenced by the surrounding harness—prompts, rules, tools, and context—making these components responsible for 90% of the system’s behavior.
How can organizations improve their AI systems based on this insight?
By focusing on designing better harnesses, including prompts, tools, guardrails, and context management, organizations can achieve more reliable and cost-effective AI performance.
Does this mean the underlying models are no longer important?
Models remain important but are now seen as just one part of a larger system. The whitepaper suggests that optimizing the surrounding infrastructure yields greater benefits.
What are the risks of focusing less on models?
The main risk is underestimating the importance of model improvements; however, the authors argue that system configuration and management are more impactful for system reliability and cost.
Will this shift change how AI development is taught or structured?
Yes, it encourages a focus on system architecture, configuration, and context engineering, potentially leading to new training and best practices in AI engineering.
Source: ThorstenMeyerAI.com
