📊 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 emphasizes that in AI-assisted software development, the model’s size and quality are less critical than the surrounding harness and context engineering. This shift impacts how companies should invest in AI tools and infrastructure.
According to a new Google whitepaper, the model used in AI coding agents accounts for only about 10% of the system’s behavior, with the remaining 90% determined by the harness and context engineering. This insight challenges the common focus on model size and underscores the importance of configuration, scaffolding, and strategic design in AI development. The findings have significant implications for how companies allocate resources in AI projects and develop their SDLC approaches.
The whitepaper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, states that the most impactful factor in AI system performance is the harness — the prompts, rules, tools, and observability layers that surround the model. Evidence from benchmarks shows that a single change in harness configuration can dramatically improve performance, even when using the same underlying model. For example, moving an agent into the top tier in a public benchmark was achieved solely through harness adjustments.
The paper emphasizes that verification, judgment, and direction are now the new craft in AI development, shifting the focus from model selection to system design. It advocates for a disciplined approach called agentic engineering, where AI is embedded within a structured framework of tests, evals, and guardrails, rather than vibe coding, which relies on minimal prompts and oversight.
Furthermore, the authors highlight that the cost dynamics favor investing in harness and context engineering, as ad-hoc prompting can lead to higher token consumption, maintenance, and security risks over time. The strategic takeaway is that organizations should prioritize building robust scaffolding and context management to achieve sustainable AI performance.
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 and Resource Allocation
This shift means that organizations can achieve better AI outcomes by focusing on the design of the surrounding infrastructure rather than solely on acquiring the latest models. Investing in harness development, context engineering, and verification processes can provide a durable competitive advantage, reducing costs and increasing reliability over time. It also suggests that AI teams should reframe their skills toward configuration, system architecture, and strategic planning, rather than just prompt engineering or model experimentation.
AI Engineering: Building Applications with Foundation Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background of the SDLC Shift and Prior AI Practices
The new whitepaper builds on ongoing debates in AI development about the importance of model size versus system design. Historically, the focus was on acquiring larger, more powerful models, but recent developments and benchmarks indicate that system configuration and context management are more impactful. The paper references prior trends like vibe coding, which prioritized quick prompts with minimal oversight, and contrasts it with a disciplined approach called agentic engineering, emphasizing structured, verified, and goal-oriented AI workflows.
This evolution reflects broader changes in the AI industry, where the emphasis is shifting from raw model capabilities to the surrounding infrastructure that ensures reliable, cost-effective, and secure AI deployment. The findings align with recent experiments demonstrating performance gains through harness optimization alone, underscoring the importance of system design over model size.
“The behavior you experience in AI tools is dominated by scaffolding you can build, own, and improve — not the frontier models themselves.”
— Addy Osmani
Ai Automation Kit PLC Programming Software, Logic Function HMI, Run Simulator
1 PLC Controller
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unclear Aspects of Model Versus Harness Impact
While the evidence suggests that harness and context engineering are more influential than model size, the precise thresholds and best practices for different applications remain to be fully established. It is still unclear how these principles scale across varied domains or with future model improvements, and whether certain models might still outperform in specific scenarios despite system configuration.
Context Engineering for Claude: A Practitioner's Guide to CLAUDE.md, Memory Tools, and Three-Layer Workflows for Solopreneurs, Freelancers, and Product Managers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Future Focus on System Design and Best Practices
Organizations are expected to shift their AI development strategies toward investing in harness and context engineering, including building reusable schemas, testing frameworks, and security guardrails. Further research and industry collaboration will likely define standardized best practices, while AI vendors may offer more configurable tools to support this approach. Monitoring the evolution of benchmarks and case studies will be essential to refine these strategies.
Observability with Grafana: Monitor, control, and visualize your Kubernetes and cloud platforms using the LGTM stack
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 according to the new whitepaper?
The whitepaper argues that the model itself accounts for only about 10% of the system’s behavior; most of the performance depends on how the model is integrated, configured, and guided through prompts, rules, and tools — collectively called the harness.
How does this shift affect AI project costs?
Focusing on harness and context engineering can lower long-term costs by reducing token waste, improving reliability, and decreasing maintenance and security expenses. It shifts investment from model acquisition to system design and configuration.
What skills should AI teams prioritize now?
Teams should develop expertise in system architecture, prompt and context engineering, verification, and security guardrails, rather than solely focusing on prompt engineering or model experimentation.
Does this mean model quality is less important?
Not necessarily; while the model remains a critical component, the whitepaper emphasizes that its impact is limited compared to how it is integrated and guided within the system.
Source: ThorstenMeyerAI.com
