📊 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 emphasizes that in AI-assisted coding, the model itself is just 10% of the system. The focus should be on harness design and context engineering, which drive most behavior and costs.

A Google whitepaper released in March 2026 states that the AI model used in software development accounts for only 10% of the overall system behavior. The primary influence lies in the harness—the prompts, tools, rules, and observability surrounding the model—comprising 90%. This challenges common assumptions that upgrading models alone significantly improves AI systems and shifts focus toward configuration and context engineering.

The whitepaper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, argues that the common industry practice of blaming models for failures is misguided. Instead, most issues stem from how the AI is integrated—through prompts, rules, and tooling. Concrete experiments cited include a team improving their coding agent’s performance by only adjusting the harness, not the model itself, raising the importance of system design. The authors emphasize that the behavior of AI agents depends heavily on configuration failures, such as missing tools or vague rules, rather than the underlying model’s capabilities.

The report also introduces the concept of context engineering, which involves loading the right information, instructions, and tools dynamically based on the task. This approach allows a generalist AI to adapt to specialized roles without carrying all capabilities at once, optimizing cost and performance. The whitepaper underscores that the economics of AI development are shifting, with disciplined engineering—focused on harness and context—being more cost-effective over time than ad-hoc vibe coding, which incurs high token and maintenance costs.

At a glance
reportWhen: published March 2026
The developmentThe new Google whitepaper highlights that in AI-driven software development, the model is only 10% of system behavior, shifting focus to 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

Implications for AI Development Strategies

This shift in understanding has major implications for how organizations approach AI integration. Instead of chasing the latest model upgrades, companies should invest in system design, tooling, and context management. Focusing on harness and context engineering can lead to more reliable, cost-effective AI systems, reducing token waste, security risks, and maintenance burdens. This perspective redefines the core skills needed for AI teams, emphasizing configuration and system architecture over model selection alone.

The AI Prompt Playbook: Master AI Prompt Engineering with 140 Ready-to-Use Templates for ChatGPT, Claude, Gemini & Copilot

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.

Background on AI System Design and Industry Trends

Prior to this whitepaper, industry focus largely centered on acquiring or developing the most advanced AI models, assuming that model improvements directly translated into better system performance. However, as AI adoption accelerates, practitioners have observed that system failures often trace back to how models are integrated, rather than the models themselves. The paper builds on recent experiments demonstrating that small tweaks to prompts and tooling can outperform major model upgrades, highlighting a paradigm shift in AI engineering.

Since early 2026, the proliferation of AI coding agents has led to a recognition that 85% of developers use AI tools regularly, with 41% generating most code via AI. Yet, the industry is now realizing that the real challenge is system configuration—creating robust, secure, and cost-efficient harnesses—rather than simply deploying larger models.

“The behavior you experience in AI systems is dominated by how you build and tune the harness, not just the model itself.”

— Addy Osmani

Observability in the AI-Native Era: Leveraging AIOps to build, observe, and operate resilient systems

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.

Unclear Impact on Future AI Model Development

It remains unclear whether this shift will influence the development priorities of major AI providers, or if model improvements will still be prioritized despite the evidence that harness and context are more impactful. The long-term consequences for AI innovation and competitive advantage are still being evaluated, and some industry leaders may continue to focus on model size and capabilities.

MUCAR 892BT AI-Assisted Bidirectional Scan Tool, Full System OBD2 Scanner, Bi-Directional OBD2 Scanner Diagnostic Tool,ECU Coding, 35 Services, FCA Autoauth, CANFD and DOIP, Free Lifetime Upgrade

MUCAR 892BT AI-Assisted Bidirectional Scan Tool, Full System OBD2 Scanner, Bi-Directional OBD2 Scanner Diagnostic Tool,ECU Coding, 35 Services, FCA Autoauth, CANFD and DOIP, Free Lifetime Upgrade

【Powerful Performance】: OBD2 scanner, featuring an 8-inch ultra-large display, the MUCAR 892BT runs on Android 10 with a…

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Next Steps for AI Engineering Teams and Industry

Organizations should reassess their AI development strategies, investing more in system architecture, tooling, and dynamic context management. Future research and best practices are likely to focus on scalable harness design and context loading techniques. Additionally, industry standards may evolve to emphasize configuration and verification processes as core competencies in AI deployment.

Building AI-Powered Products: The Essential Guide to AI and GenAI Product Management

Building AI-Powered Products: The Essential Guide to AI and GenAI Product Management

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?

Because most AI failures and behaviors are determined by how the model is integrated, configured, and guided through prompts, tools, and rules—collectively called the harness.

Should companies stop upgrading AI models?

Not necessarily, but the whitepaper suggests that investing in harness and context engineering may deliver better value and reliability than solely focusing on model upgrades.

What is meant by ‘harness’ in AI systems?

The harness includes prompts, rules, tools, observability, and other systems that wrap around the AI model to control its behavior and performance.

How does this shift affect AI development costs?

Focusing on harness and context engineering can reduce token consumption, security risks, and maintenance costs, making AI development more economically sustainable over time.

What skills will be most important for AI engineers moving forward?

Skills in system architecture, tooling, context management, and verification will become more critical than just model training or fine-tuning.

Source: ThorstenMeyerAI.com

You May Also Like

The European Union: Rules First, Cushion Always

The European Union emphasizes regulation, worker voice, and social protections over ownership in its strategic response to AI and labor shifts.

Japan ad agency fights click fraud with biometric tech app

Hakuhodo’s new biometric verification app aims to reduce AI-driven ad fraud by ensuring human interaction in digital advertising.

Productivity Paradox: Is AI Really Making Us More Productive at Work?

Unlock the truth behind AI’s impact on productivity and discover why embracing change is crucial for real results.

Forward-Deployed Engineer Economics 2.0: The Unit Economics Math, Six Months Later

Six months after initial analysis, FDE economics reveal high profitability at scale but risks of losses at lower tiers, impacting enterprise AI deployment strategies.