📊 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
The latest research highlights that AI models are just a small part of software systems. Success depends more on how teams configure, verify, and manage AI agents, shifting focus from models to system design.
A new Google whitepaper, “The New SDLC With Vibe Coding,” emphasizes that AI models constitute only about 10% of the overall system behavior. The paper argues that the real challenge in AI-driven software development lies in system configuration, verification, and management, shifting the focus from model improvements to system design and control.
The paper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, reports that 85% of developers use AI coding agents, with over half employing them daily and around 41% of new code being AI-generated. Despite the focus on models, the authors stress that model performance is only a small part of system success. They highlight that most failures in AI agents are due to misconfigurations, missing tools, or poor context management, not the models themselves.
The authors introduce the concept of harnessing—the prompts, tools, rules, and observability layers surrounding the model—as being responsible for 90% of system behavior. They cite experiments where tweaking the harness significantly improved system performance, even with the same underlying model. This underscores the importance of system configuration and context engineering in AI development.
The paper also discusses the economics of AI development, noting that ad-hoc prompting is costlier over time due to token inefficiencies, maintenance, and security issues. Conversely, a disciplined approach—what they call agentic engineering—requires higher initial investment but offers lower marginal costs and greater reliability.
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 System Configuration Outweighs Model Improvements
This research shifts the AI development paradigm, emphasizing that system design, configuration, and verification are more critical than the choice of model. For organizations, this means investing in harness development, context management, and testing frameworks can deliver better results and cost savings, rather than chasing the latest model upgrades. It also suggests that competitive advantage lies in system stability and configurability, not just in access to the newest AI models.
MUCAR 892BT AI-Assisted Bidirectional Scan Tool, Full System OBD2 Scanner, Bi-Directional OBD2 Scanner Diagnostic Tool,ECU Coding, 35 Services, FCA Autoauth, CANFD and DOIP, Free Lifetime Upgrade
【Powerful Performance】: OBD2 scanner, featuring an 8-inch ultra-large display, the MUCAR 892BT runs on Android 10 with a…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Evolution of AI Development Strategies
Prior to this publication, the industry largely viewed AI models as the primary driver of system performance. The release of the Google whitepaper in early 2026 consolidates a shift that began in late 2024, when organizations started recognizing that configuration and system management influence AI effectiveness more than raw model capabilities. The paper builds on previous findings that most AI failures are due to misconfiguration and poor context management.
Recent experiments, such as those by LangChain and other AI toolkits, reinforce the idea that tuning prompts, tools, and context loading can dramatically improve outcomes with the same AI models. This evolution suggests a new focus for AI teams: system architecture and configuration, rather than model procurement or training.
“Model performance is only a small part of system success; most failures are configuration issues.”
— Addy Osmani
Verification and Validation of AI Software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Aspects of System Configuration Remain Unclear
While the paper convincingly shows that harness and configuration are critical, it does not specify precise best practices or standardized frameworks for system setup. The extent to which these findings apply across different AI models and use cases remains to be fully validated. Additionally, the optimal balance between upfront configuration effort and ongoing maintenance costs is still under discussion.
AI-Powered Observability: From Noise to Insight: Transforming How We Monitor, Detect, and Respond
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 expected to prioritize system configuration, testing, and guardrails in their AI workflows. Future research may focus on developing standardized frameworks and tools for harness management. Additionally, industry leaders will likely invest more in training and best practices for context engineering to maximize AI system reliability and cost efficiency. Monitoring and evaluating the impact of these strategies over the coming months will be critical to validate the paper’s findings.
The Human-Agent Orchestrator: Leading and Scaling AI-Driven Organizations
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 paper argues that the surrounding system—prompts, tools, rules, and observability—shapes AI output far more than the model itself. Proper configuration and management are the main determinants of success.
What is meant by ‘harness’ in AI systems?
Harness refers to the setup around the AI model, including prompts, rules, tools, and monitoring systems that guide and control the model’s behavior.
How does this shift affect AI development costs?
While initial investment in system configuration is higher, disciplined engineering reduces long-term costs related to token inefficiency, maintenance, and security vulnerabilities.
Will this change how AI models are selected?
Yes, organizations may place less emphasis on choosing the latest models and more on developing robust, configurable systems around those models.
What are the risks of focusing on system configuration?
The main risk is underestimating the importance of ongoing maintenance and the need for skilled system engineering to keep configurations effective and secure.
Source: ThorstenMeyerAI.com
