📊 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 the core of AI-based software development isn’t the model itself but the surrounding harness and context engineering. The model accounts for only 10% of behavior, shifting focus to configuration, verification, and judgment.
A new Google whitepaper from Addy Osmani, Shubham Saboo, and Sokratis Kartakis states that the most significant shift in software engineering is moving from writing code to expressing intent and trusting AI to generate software, with the model itself only representing about 10% of the system’s behavior.
The paper reports that as of early 2026, 85% of professional developers use AI coding agents regularly, and 41% of all new code is AI-generated. However, the key insight is that the model’s size and raw capability are less important than the harness and configuration surrounding it. The authors argue that 90% of an agent’s behavior is determined by the harness — prompts, tools, rules, and observability — not the model itself.
This shift means that effective AI development depends heavily on verification, judgment, and context engineering. The paper highlights that failures in AI agents are often due to configuration issues rather than model limitations, emphasizing the importance of scaffolding and operational design.
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 Focus on Harness and Context Matters
This finding challenges the common perception that investing in larger, more powerful models is the best strategy. Instead, it suggests that building durable, configurable scaffolding around AI models offers a more cost-effective and reliable path. For organizations, this means shifting resources from chasing the latest model to developing robust harnesses, verification methods, and context management. The emphasis on cost efficiency and security makes this a crucial insight for AI teams aiming to scale responsibly.
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.
Background on AI-Driven Software Development Shifts
The whitepaper builds on the ongoing evolution in AI-assisted coding, where the focus has shifted from simple prompt-based interactions to agentic engineering — structured, verified, and controlled AI workflows. Since early 2026, the adoption of AI coding agents has increased dramatically, with more teams integrating AI into their SDLC. Previous discussions centered on model capabilities; now, the emphasis is on how models are integrated, configured, and governed.
This development aligns with broader trends toward verification, automation, and cost management in AI workflows, marking a departure from the hype around model size and raw AI power.
“The model is only 10% of what determines behavior; the harness is 90%. Focus on configuration and verification.”
— Addy Osmani
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.
Unclear Aspects of Model Versus Harness Impact
It remains uncertain how widely organizations will adopt this perspective and whether the emphasis on harness over models will lead to measurable improvements in AI reliability and cost-efficiency at scale. The exact proportion of behavior influenced by harness in different domains and models is still being studied, and the long-term implications for AI innovation are not yet fully understood.
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.
Next Steps for AI Development and Strategy
Organizations are expected to reevaluate their AI strategies, investing more in harness development, verification frameworks, and context management. Future research will likely focus on quantifying the influence of harness components across different AI applications and models. Additionally, industry leaders may prioritize cost-benefit analyses of model size versus configuration quality, shaping the next phase of AI engineering practices.
YAML Made Simple: A Beginner’s Guide to Configuration and Data Structuring
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 explains that most of an AI agent’s behavior is determined by its harness — the prompts, tools, rules, and environment around the model — not the model’s raw capabilities.
How does this shift affect AI development costs?
Focusing on harness and configuration can reduce costs by improving reliability and decreasing the need for frequent model upgrades, shifting investment toward better scaffolding and verification.
Will larger models become less important?
According to the whitepaper, model size alone is less critical; effective harnesses and context management are more impactful for system behavior and cost-efficiency.
What does this mean for AI security?
Enhanced harnessing and verification can improve security by reducing vulnerabilities caused by unpredictable model outputs and better controlling AI behavior.
Is this approach applicable to all AI systems?
The principles are broadly applicable, but the effectiveness depends on the specific application and how well organizations implement harness and context engineering.
Source: ThorstenMeyerAI.com
