📊 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 argues that AI models constitute only about 10% of the software development process. The real focus should be on harnessing and verifying AI outputs through structured context and configuration, which has significant implications for development costs and strategies.
A new Google whitepaper, The New SDLC With Vibe Coding, emphasizes that the AI model accounts for only about 10% of the software development process. Instead, the focus should be on the harness—the prompts, tools, and configurations surrounding the model—and on verification. This shift has major implications for how organizations deploy and manage AI in development workflows.
The paper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, states that the dominant part of AI-driven development is not the model itself but the configuration, tools, and policies that guide its behavior. Experiments cited show that adjusting the harness—such as prompts, rules, and middleware—can dramatically improve performance, even with the same underlying model. For example, moving a coding agent from outside the Top 30 to the Top 5 on a benchmark involved only harness modifications.
Furthermore, the paper distinguishes between vibe coding—quick, minimal prompts—and agentic engineering, which involves structured verification, testing, and oversight. The authors argue that verification, judgment, and context engineering are the new craft, shifting the focus from model innovation to configuration and process management.
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 insight shifts the strategic focus for organizations adopting AI. Instead of investing heavily in the latest models, companies should prioritize building robust harnesses—the configurations, tools, and verification processes that shape AI behavior. This approach can lead to significant cost savings and more reliable outcomes, as the paper notes that ad-hoc prompting can cost 3–10 times more per feature than disciplined, structured development.
By understanding that the model is only 10% of the equation, organizations can better allocate resources, reduce vulnerabilities, and improve the quality of AI-generated code and decisions. This paradigm encourages a shift from model chasing to process engineering, which has long-term benefits for scaling AI in enterprise settings.
Mastering Codex for Parallel AI Agents: Run multiple AI agents at once and verify their work — a non-engineer's guide to supervising Codex (Codex Mastery Series Book 2)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Evolution of AI Coding and Development Practices
The whitepaper builds on recent trends where AI tools are now used by over 85% of professional developers, with roughly 41% generating new code via AI. Prior to 2026, the focus was on model improvements; now, the emphasis has shifted to how models are integrated and controlled within workflows. The concept of vibe coding, popularized in early 2025, is contrasted with the emerging discipline of agentic engineering, which emphasizes structure, testing, and verification.
This development reflects a broader industry understanding that AI’s value depends heavily on how it is used and controlled, not just the underlying technology. The experiments cited in the paper demonstrate that small changes in configuration can produce outsized improvements in AI performance, reinforcing the importance of process over raw model power.
“The biggest shift in software engineering isn’t a new language or framework; it’s moving from writing code to expressing intent and trusting machines to realize that intent.”
— 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.
Unexplored Aspects of Implementation and Cost
While the paper presents compelling evidence that harness and verification are critical, it does not specify exact best practices for different types of projects or organizational sizes. The long-term impact on cost savings and security remains to be validated across diverse real-world scenarios. Additionally, the optimal balance between upfront configuration costs and ongoing operational expenses is still under discussion.
lweiyupeixx Press Model Separator Press Type Automatic Model Parts Detacher Part Separation Tool Hobby Assembling Model Ergonomic
Effortlessly separate model components with our Press Type Model Separator, enhances efficiency and minimize damage risk.
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Organizations Adopting AI-Driven Development
Organizations should begin assessing their current AI workflows to identify configuration and verification bottlenecks. Developing structured harnesses, investing in testing frameworks, and training teams in context engineering will be key. Further research and case studies are expected to clarify best practices and quantify long-term benefits, especially around cost and security.
AI Model Validation & Testing: Ensuring Reliable AI Systems — Bias Testing, Robustness Evaluation & Regulatory Compliance (AI Compliance Toolkit)
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 AI development process?
The whitepaper argues that the model’s behavior is heavily influenced by how it is configured, guided, and verified. The surrounding harness—prompts, tools, rules—accounts for roughly 90%, making it the primary driver of performance and reliability.
How can organizations improve their AI outputs without changing models?
By focusing on harness design—adjusting prompts, adding tools, setting rules—and implementing rigorous verification and testing processes, organizations can significantly enhance AI performance using the same underlying models.
What are the risks of neglecting harness and verification?
Neglecting these aspects can lead to higher costs, security vulnerabilities, and unreliable outputs, as most failures are configuration-related rather than model-related, according to the whitepaper.
Does this mean AI models are becoming less important?
Not less important, but the whitepaper highlights that their impact is limited without proper harnessing, configuration, and verification. The focus shifts from model innovation to process engineering.
What should development teams prioritize next?
Teams should prioritize building and refining their harnesses—prompts, tools, rules—and establishing robust verification processes to maximize AI effectiveness and control costs.
Source: ThorstenMeyerAI.com
