The Model Is Only 10%: The Real Lesson of the New SDLC

📊 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 reveals that in AI-assisted software development, the core value lies in configuring and engineering the surrounding system, not the AI model itself. This shift impacts how companies should invest in AI tools and development strategies.

A new Google whitepaper published in early 2026 states that the AI model accounts for only about 10% of a system’s behavior in AI-assisted development. The paper emphasizes that the harness and context engineering are the primary factors influencing performance, shifting focus from model upgrades to system configuration and management.

The whitepaper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, highlights that 85% of professional developers regularly use AI coding agents, with over half using them daily. It stresses that 41% of new code is AI-generated, yet the real challenge lies in system configuration and contextual setup.

The authors argue that the model itself is only a small part—roughly 10%—of what determines an AI agent’s behavior. The remaining 90% involves the harness, prompts, tools, and context. They cite experiments where tweaking only the harness or context improved performance significantly, without changing the model.

This perspective encourages organizations to invest more in system design, verification, and context engineering rather than solely chasing the latest AI models.

At a glance
reportWhen: published early 2026
The developmentThe new whitepaper by Addy Osmani, Shubham Saboo, and Sokratis Kartakis argues that in AI coding, the model is only 10% of the system’s behavior, emphasizing the importance of harness and context engineering.
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

Why the Focus on Harness and Context Matters

This shift redefines strategic priorities in AI development, suggesting that organizations can achieve better results by optimizing system configuration rather than constantly upgrading models. It also impacts cost management, security, and long-term maintenance, as the system setup becomes the primary driver of performance and reliability in AI tools.
MUCAR 892BT AI Bi-Directional OBD2 Scanner, Full System OBD2 Scanner Diagnostic Tool,35 Services,ECU Coding,FCA Autoauth,CANFD&DOIP,for Car Owners, DIYer, Technicians, Inspectors, Trainees and Others

MUCAR 892BT AI Bi-Directional OBD2 Scanner, Full System OBD2 Scanner Diagnostic Tool,35 Services,ECU Coding,FCA Autoauth,CANFD&DOIP,for Car Owners, DIYer, Technicians, Inspectors, Trainees and Others

Powerful Performance for busy days in the shop or your home garage: an 8-inch ultra-large display paired with…

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Background on the Evolving Role of AI Models

Up to early 2026, the AI community has heavily emphasized model advancements as the key to better performance. However, recent insights from the Google whitepaper challenge this view, arguing that most failures and inefficiencies stem from poor system configuration.

Previous trends focused on acquiring the latest models like GPT-4 or Claude, but experiments show that tuning the harness and context can often yield more substantial improvements than switching models. The paper situates this as a fundamental paradigm shift in software engineering for AI.

“The model is only 10% of what determines behavior; the harness and context are the real crafts.”

— Addy Osmani

AI Prompt Engineering Bible (7 Books in 1): Beginner-to-Pro System to Master ChatGPT and Generative AI for Powerful Results and Real Income (The Generative AI Creator Series)

AI Prompt Engineering Bible (7 Books in 1): Beginner-to-Pro System to Master ChatGPT and Generative AI for Powerful Results and Real Income (The Generative AI Creator Series)

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Unclear Aspects of Implementation and Impact

It is not yet clear how organizations will scale their focus on harness and context engineering across different industries or how quickly this shift will be adopted widely. The long-term effects on AI development costs and security protocols remain to be seen.
Architecting AI Software Systems: Crafting robust and scalable AI systems for modern software development

Architecting AI Software Systems: Crafting robust and scalable AI systems for modern software development

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Next Steps for Developers and Organizations

Organizations should evaluate their current AI workflows, emphasizing system configuration, prompt design, and context management. Future research and industry practices are likely to focus on developing standardized harness tools, verification frameworks, and best practices for context engineering. Monitoring how this shift influences AI performance and costs over the coming months will be crucial.

Verification of Autonomous Systems

Verification of Autonomous Systems

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 shows that most of an AI agent’s behavior depends on how it is configured, the prompts, tools, and context provided, rather than the underlying model itself.

How can organizations improve their AI systems based on this insight?

Focus on designing better harnesses, prompts, and context management tools, and invest in verification and testing frameworks rather than solely upgrading models.

Does this mean model development is less important?

Not necessarily less important, but the strategic emphasis should shift towards system configuration and management, which have a greater impact on performance and cost-efficiency.

What are the risks of over-relying on harness and context engineering?

Potential risks include increased complexity, maintenance overhead, and security vulnerabilities if system configuration is not managed carefully.

When will this shift in focus become standard practice?

It is likely to accelerate over the next 12-24 months as organizations recognize the cost and performance benefits of system-focused engineering.

Source: ThorstenMeyerAI.com

Nothing in this article is financial or investment advice. Cryptocurrency and precious-metal investments carry significant risk — do your own research and consider a licensed advisor.
You May Also Like

The SSD Squeeze: Why Storage Joined the Party

Enterprise and consumer SSD prices surge due to NAND shortages driven by AI demand and factory competition, impacting supply and pricing globally.

The Defender’s Counter-Cascade.

Google’s Threat Intelligence Group disclosed the first real-world AI-built zero-day exploit on May 11, highlighting deployment gaps in AI-driven cybersecurity defenses.

The Real Cost Of A Local-Inference Rig In 2026

An analysis of the hardware costs for running large language models locally in 2026, highlighting VRAM constraints, hardware choices, and value considerations.

Customer service + BPO. The operational-scale displacement.

Empirical evidence shows 8 million workers in India and the Philippines face AI-driven displacement, with a shift to hybrid models emerging as the new operational norm.