📊 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 emphasizes that in AI-assisted software development, the model itself is only 10% of the system. The focus should be on harness design and context engineering, which drive performance and cost-efficiency.
A new whitepaper from Google, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, states that the AI model represents only about 10% of what determines the behavior of AI coding agents. This shifts the focus from models to the surrounding infrastructure, which has significant implications for software development strategies and costs.
The whitepaper argues that the dominant factor in AI system performance is the harness: the prompts, tools, rules, and observability layers built around the model. Experiments cited show that changing only the harness—such as prompts or middleware—can move an agent from mediocre to top-tier performance, even with the same underlying model.
Furthermore, the paper emphasizes context engineering as a critical skill, involving the design of instructions, knowledge bases, memory, and guardrails. These elements significantly influence code quality and system reliability, often more than the prompt itself.
The authors warn that the misconception that the model is the main driver leads to misallocated investments, as most failures stem from configuration issues rather than the AI core. They also highlight the economic impact, noting that a disciplined approach—focused on harness and context—can reduce total ownership costs by avoiding the high token burn and maintenance expenses associated with vibe coding or ad-hoc prompting.
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 Design and Context Engineering Are Game-Changers
This shift in understanding impacts how organizations should invest in AI development. By recognizing that 90% of behavior is driven by harness and context, companies can optimize performance, reduce costs, and improve system reliability. It challenges the industry’s focus on model upgrades, urging a strategic pivot towards configuration, tooling, and knowledge management.
For CTOs and developers, this means prioritizing the development of robust scaffolding, reusable context modules, and verification processes. The approach promises more durable competitive advantages and better control over AI system behavior, especially as models continue to evolve rapidly.
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.
Background on AI Development and the Shift in Focus
As of early 2026, AI coding agents are widely adopted, with 85% of professional developers using them regularly and 41% generating most of their code through AI, according to industry surveys. Historically, the industry has concentrated on improving models—such as GPT and Claude—believing that better models would automatically lead to better results.
The recent whitepaper challenges this notion, illustrating through experiments that configuration and scaffolding—collectively called harness—are more influential than the model itself. This perspective aligns with ongoing industry observations that failures often stem from poor setup or inadequate context rather than the AI’s core capabilities.
It also reflects a broader trend towards treating AI development as a total-cost-of-ownership problem, emphasizing long-term costs and system robustness over short-term model performance improvements.
“The model is only 10% of what determines behavior; the harness and context are the real drivers.”
— 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 How Organizations Will Shift Investment Strategies
It is not yet clear how quickly and widely organizations will adopt the recommended focus on harness and context engineering, or how this will reshape industry standards and tooling. Further studies are needed to quantify long-term cost savings and performance gains across different sectors.
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.
Next Steps for AI Development and Industry Adoption
Organizations are likely to reevaluate their AI development priorities, investing more in configuration, tooling, and knowledge management. Industry leaders may develop new standards and best practices for harness design and context engineering. Future research will aim to quantify the impact of these strategies on performance, cost, and reliability, while tooling providers might introduce new solutions tailored to these insights.
AI system testing and validation tools
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 scaffolding—prompts, tools, rules, and context—has a much larger influence on AI behavior than the core model itself.
How can companies improve their AI systems based on this insight?
By focusing on harness design and context engineering—building robust prompts, knowledge bases, and guardrails—organizations can significantly enhance system reliability and reduce costs.
Does this mean model improvements are no longer important?
Model improvements remain valuable, but the whitepaper emphasizes that most performance gains come from better configuration and management of the surrounding infrastructure.
What are the economic implications of this shift?
Prioritizing harness and context can lower total ownership costs by reducing token burn, maintenance, and security vulnerabilities, leading to more sustainable AI operations.
Will this change how AI development teams operate?
Yes, teams will need to develop expertise in configuration, context management, and verification, moving beyond just model selection and training.
Source: ThorstenMeyerAI.com
