📊 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 coding, the core value lies in how systems are configured and guided, not in the AI model itself. The model is just 10% of the equation; the rest is in harness and context management.

A new Google whitepaper argues that in AI-assisted software engineering, the AI model accounts for only about 10% of system behavior. Instead, the harness and context engineering—the configuration, prompts, tools, and guardrails—are where most control and value reside. This shift has significant implications for how organizations should invest in AI development and deployment.

The whitepaper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, emphasizes that the dominant factor in AI-generated code is not the underlying model but the surrounding framework. Experiments cited show that changing only the harness—such as prompts, tools, and configuration—can dramatically improve AI performance, often more than upgrading the model itself. For example, a team improved a coding agent’s ranking from outside the top 30 to the top 5 by adjusting only the harness, with no change to the model.

The authors introduce the concept of agentic engineering, where AI is embedded within a structured environment of verification, testing, and guardrails, contrasting with the more casual vibe coding approach. They argue that the primary skill in AI development is now context engineering: designing the information and rules that guide the AI’s output. They also highlight the economic aspect, stating that while vibe coding appears cheap, it incurs higher long-term costs due to inefficiency, security vulnerabilities, and maintenance challenges. Conversely, investing in structured, disciplined approaches leads to lower marginal costs over time.

At a glance
reportWhen: published March 2026
The developmentGoogle’s new whitepaper highlights that in AI-driven software development, the majority of system behavior depends on configuration, not the AI model itself.
The Model Is Only 10% — The New SDLC With Vibe Coding
AI Dispatch · Field Notes
Google · Osmani, Saboo & Kartakis · May 2026

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.

A spectrum, not a binary — the differentiator is how outputs get verified
Vibe Coding
Casual prompts · “does it seem to work?” · disposable code · high risk
Structured AI-Assisted
Detailed prompts + constraints · manual testing · features in real codebases
Agentic Engineering
Formal specs · automated tests + evals + CI gates · production scale · low risk
Tests verify the deterministic; evals verify the rest. Without both, it’s vibe coding — however clever the prompt.
The idea worth building your strategy around
Agent = Model + Harness
~10%
HARNESS — prompts · tools · context · hooks · sandboxes · observability
MODEL~90% IS YOUR SURFACE AREA, NOT THE PROVIDER’S
Outside Top 30 → Top 5 on Terminal Bench 2.0 by changing only the harness — same model.
“Most agent failures, examined honestly, are configuration failures” — a missing tool, a vague rule, a noisy context.
The economics: it’s a token-cost problem (CapEx vs OpEx)
Vibe Coding
Low CapEx · High OpEx
Looks free, hides debt: token burn (fix-it loops), maintenance tax (AI spaghetti), security remediation. Crosses over to 3–10× more per feature.
Agentic Engineering
High CapEx · Low OpEx
Pay upfront (specs, evals, context), then ship cheaply. Levers: context engineering for first-pass success + intelligent model routing — cheap models for the easy work.
85%
of devs use AI coding agents (51% daily)
41%
of all new code is AI-generated
~90%
of agent behavior is the harness, not the model
+19%
longer on some tasks (METR) — verification is the cost
The read

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.

Source: Osmani, Saboo & Kartakis, “The New SDLC With Vibe Coding,” Google (May 2026). Figures are the paper’s own, incl. METR & LangChain. Analysis is the author’s.
thorstenmeyerai.com

Impact of Configuration Focus on AI Development Strategies

This new understanding shifts the strategic focus for AI teams from chasing the latest models to optimizing how they configure and control AI systems. Organizations that master harness and context engineering can achieve better performance at lower costs, gaining a competitive advantage. It also implies that AI’s true value lies in the human-led design of systems, not just in access to powerful models, which are now considered a commodity.

YAML Made Simple: A Beginner’s Guide to Configuration and Data Structuring

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.

Evolution of AI-Assisted Coding and SDLC Practices

The whitepaper builds on recent trends where AI is now integral to software development, with over 85% of developers using AI coding agents as of early 2026. Previously, the focus was on acquiring more advanced models; now, the emphasis is on how these models are integrated and guided. The concept of vibe coding, popularized by Andrej Karpathy, is contrasted with agentic engineering, which involves formal specifications, automated tests, and oversight. The shift represents a maturation in AI practices, emphasizing reliability and cost-efficiency over raw model power.

This development aligns with broader industry observations that AI system failures often stem from poor configuration rather than model limitations. The whitepaper underscores that the most durable advantage comes from controlling the environment around the model, not the model itself.

“The model you’re paying so much attention to is only 10% of what determines behavior; the harness is 90%.”

— Addy Osmani

AI Engineering: Building Applications with Foundation Models

AI Engineering: Building Applications with Foundation Models

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Unclear Aspects of Practical Implementation

While the whitepaper presents compelling evidence that harness and context are critical, it remains unclear how organizations will effectively scale these practices across diverse teams and projects. Specific methodologies, tools, and best practices for mastering context engineering are still emerging, and the long-term impact on AI development costs and security is under ongoing investigation.

AI Integrated Software Automation Testing JAVA with Selenium: Selenium WebDriver with JAVA | Software Automation Testing with AI Tools | TestNG ... 2025 | Code with AI | Auto Coding with AI

AI Integrated Software Automation Testing JAVA with Selenium: Selenium WebDriver with JAVA | Software Automation Testing with AI Tools | TestNG … 2025 | Code with AI | Auto Coding with AI

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Next Steps for AI-Driven Software Engineering

Organizations are likely to focus on developing internal capabilities in harness and context engineering, including investing in tooling, training, and best practices. Further research and case studies are expected to clarify how best to implement these principles at scale, and industry standards may evolve around configuration management for AI systems. Monitoring how these practices influence software quality and costs will be critical in the coming months.

Serious Managers Guide to AI Guardrails

Serious Managers Guide to AI Guardrails

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 system behavior?

The whitepaper shows that the majority of an AI system’s behavior depends on how it is configured, guided, and integrated—the harness and context—rather than the underlying model itself.

What is agentic engineering?

Agentic engineering involves embedding AI within a structured environment of specifications, tests, and guardrails, emphasizing configuration and verification over casual prompting.

How does this shift affect AI development costs?

While initial investment in harness and context setup may be higher, it leads to lower long-term costs through improved efficiency, security, and maintainability, reducing the need for frequent model upgrades.

What skills are now most important for AI teams?

Mastering context engineering—designing prompts, rules, and configurations—is now more valuable than solely focusing on model selection or tuning.

Does this mean models are becoming less important?

Models are still essential, but their role is now seen as a commodity; the real value lies in how they are integrated and controlled through configuration and environment design.

Source: ThorstenMeyerAI.com

You May Also Like

The Deployment Company, Back to the 70s, Apple and Intel

OpenAI launches The Deployment Company to accelerate enterprise AI deployment, signaling a shift toward AI replacing human roles, akin to 1970s computing.

AI data centers require 36 times more fiber than designs with standard servers — severe glass shortages push cable lead times out to a full year

AI data centers require significantly more fiber optic cable—36 times more than standard server designs—due to surging demand, straining supply chains and expanding infrastructure needs.

AI‑Powered Design: How Artificial Intelligence Inspires Creativity

Jump into the world of AI-powered design and discover how artificial intelligence can unlock your creative potential like never before.

Évian and the Fallout: What Europe Actually Wants From Amodei, Hassabis, and Altman

European leaders at the G7 summit in Évian pressed US AI CEOs for access, sovereignty, and safety guarantees amid US export controls.