📊 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’s size accounts for only 10% of system behavior. The critical factors are the harness and context engineering, which determine performance and cost efficiency.
A new Google whitepaper, ‘The New SDLC With Vibe Coding’, reveals that the model size accounts for only about 10% of AI system behavior. The report emphasizes that the harness and context engineering are the primary drivers of performance, cost, and reliability in AI-assisted software development. This shifts the traditional focus from developing larger models to optimizing system configuration and contextual inputs, a change that could redefine best practices across the industry.
The whitepaper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, states that over 85% of professional developers now use AI coding agents regularly, with 51% doing so daily. It highlights that roughly 41% of all new code is AI-generated, underscoring the growing reliance on AI tools in software engineering.
The core insight is that the model itself is only a small part of the system. The majority of system behavior—about 90%—is determined by the harness—the prompts, rules, tools, and observability layers surrounding the model. Concrete experiments cited in the paper show that changing only the harness or the context setup can dramatically improve an AI agent’s performance, often more than upgrading the model itself.
The authors argue that this realization shifts the strategic focus for organizations: investing in system configuration, context management, and verification offers a more durable competitive advantage than chasing the latest model release. They also warn that ad-hoc prompting and vibe coding—quick, minimal review workflows—are less sustainable and more costly over time compared to disciplined, structured approaches called agentic engineering.
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-Driven Software Development Strategies
This shift in understanding has significant implications for how companies allocate resources. By recognizing that system configuration and context engineering drive most behavior, organizations can focus on building durable, scalable AI systems that are easier to maintain, more secure, and cost-effective. It challenges the industry to rethink the value of large models and prioritize system design, which could lead to more predictable outcomes and better long-term ROI in AI projects.
Learn Model Context Protocol with TypeScript: Build agentic systems in TypeScript with the new standard for AI capabilities
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Industry Practices Are Evolving with AI Integration
The whitepaper builds on the rapid adoption of AI coding tools, which as of early 2026, have become integral to many development workflows. It references recent experiments showing that performance improvements often come from better harness design rather than model upgrades. The paper also situates its insights within broader trends: increasing AI use, rising costs associated with token economy, and the need for disciplined engineering practices, such as verification, testing, and guardrails, to ensure quality and security.
This perspective aligns with ongoing industry debates about balancing AI innovation with operational reliability and cost management, emphasizing that AI development is now more about system engineering than model innovation alone.
“The model is only 10% of what determines system behavior; the harness and context are the other 90%.”
— 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 presents compelling evidence that harness and context are dominant factors, it does not specify how organizations can best implement these practices at scale or quantify cost savings precisely. The long-term impact on model development priorities remains to be seen, and industry adoption of these insights is still emerging.
GenAI Observability and Evaluation: The Complete Engineer's Guide to Monitoring, Measuring, and Trusting AI Systems
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Organizations Embracing the New SDLC Paradigm
Organizations should focus on developing expertise in system configuration, context engineering, and verification processes. Future research and case studies will likely explore best practices for harness design, cost optimization, and security in AI-driven SDLC. Companies that adapt quickly may gain competitive advantages by building more reliable, scalable AI systems based on these principles.
Intelligent Test Automation: Mastering Playwright with Agentic AI for Self-Healing, Adaptive Web Testing: A Complete Guide to Building Autonomous Testing … Playwright & AI Automation Suite Book 2)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Why is the model size considered only 10% of the system?
The whitepaper demonstrates through experiments that most of the AI system’s behavior is determined by the harness—prompts, rules, tools, and context—rather than the underlying model size.
How does this shift affect AI development investments?
It suggests that investing in system configuration, verification, and context management offers more long-term value than focusing solely on acquiring larger or more advanced models.
What is ‘agentic engineering’?
Agentic engineering involves designing AI systems with structured harnesses, verification, and context loading, moving away from vibe coding towards disciplined, reliable workflows.
Does this mean smaller models are better?
Not necessarily better, but the whitepaper emphasizes that system design and harness optimization are more critical for performance, cost, and security than model size alone.
What are the main challenges in adopting this approach?
Implementing effective harnesses, developing expertise in context engineering, and establishing verification processes require investment and organizational change, which may be challenging for some teams.
Source: ThorstenMeyerAI.com
