The Model Is Only 10%: The Real Lesson of the New SDLC
AIThis post was created with the assistance of artificial intelligence (AI).

📊 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-driven development, the model itself is only 10% of system behavior. The real focus should be on harness, context, and verification, which constitute 90%. This shifts strategic priorities for developers and organizations.

A new whitepaper from Google, titled “The New SDLC With Vibe Coding”, asserts that in AI-assisted software development, the model accounts for only about 10% of system behavior. The paper emphasizes that the harness, context, and verification are where most of the development effort and strategic value lie, marking a shift in how organizations should approach AI integration.

The whitepaper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, highlights that the dominant part of AI system performance depends on configuration, scaffolding, and context engineering rather than the underlying model. Concrete evidence includes experiments where changing only the harness or context improved performance significantly, despite using the same model. The authors argue that cost and effectiveness in AI development are driven more by these factors than by the latest model versions.

They introduce the concept of agentic engineering, where AI systems are built with structured prompts, tools, guardrails, and context management, rather than relying solely on prompt engineering or model upgrades. The whitepaper stresses that organizations should invest more in building and owning their harnesses to gain a durable competitive advantage, as opposed to chasing the latest model releases.

At a glance
reportWhen: published March 2026
The developmentThe Google whitepaper argues that the most significant shift in SDLC is moving from focusing on models to emphasizing harness and context engineering, with the model representing only 10% of system behavior.
Crypto market snapshot
Fear & Greed Index
19/100 — Extreme Fear
Bitcoin BTC$60,416▲ 2.9%
Ethereum ETH$1,624▲ 2.9%
Tether USDT$0.9987▲ 0.0%
BNB BNB$550.75▲ 0.6%
USDC USDC$0.9997▲ 0.0%
XRP XRP$1.06▲ 1.3%
Solana SOL$78.21▲ 4.8%
TRON TRX$0.3155▼ 0.3%
Live data · CoinGecko · alternative.me (24h change)
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 Focus on Harness and Context Matters

This shift in understanding impacts how companies allocate resources for AI development. By recognizing that most of the system’s effectiveness depends on configuration and context management, organizations can improve performance, reduce costs, and enhance security. It challenges the common narrative that upgrading models is the primary lever for better AI outcomes, emphasizing instead the strategic importance of customized harnesses and disciplined verification.

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

  • Large 8-inch Display: Clear, responsive diagnostics interface
  • Powered by Android 10: Fast, reliable performance
  • Quad-core Processor: Smooth operation and quick response

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Background of the Model-Centric Narrative

Prior to this whitepaper, the industry largely focused on model advancements as the primary driver of AI improvements, with many companies investing heavily in acquiring or developing the latest models. The narrative was that better models automatically lead to better results. However, recent experiments and practical experiences have shown that configuration and context engineering often have a greater impact. The paper builds on ongoing discussions about the total cost of ownership and the importance of system design in AI deployment, especially as AI becomes more embedded in software development processes.

“The model is only 10% of what determines behavior; the harness is 90%. Focus on configuration, tools, and verification to truly leverage AI.”

— Addy Osmani

Beyond the Prompt: Spec-Driven Development and Context Engineering for AI Coding Agents

Beyond the Prompt: Spec-Driven Development and Context Engineering for AI Coding Agents

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Unclear Aspects of Implementation and Impact

While the whitepaper presents compelling evidence and a clear conceptual shift, it remains unclear how quickly organizations will adopt this perspective at scale. The precise methods for systematically building and managing harnesses, as well as the long-term impact on AI development costs and security, are still being explored. Additionally, the industry consensus on this paradigm shift has yet to fully solidify, and some may continue to prioritize model upgrades due to existing investment and infrastructure.

AI Use and Verification Basics: Level One General Business Training Manual

AI Use and Verification Basics: Level One General Business Training Manual

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Next Steps for AI Development Strategies

Organizations are expected to reevaluate their AI development approaches, investing more in harness and context engineering. Future research and practical experiments will likely focus on establishing best practices for building scalable, secure, and maintainable harnesses. Industry leaders may also develop new tools and frameworks to facilitate this shift, with the goal of reducing costs and improving AI reliability over time.

The Authentic Pocket Engineer: Small 3 Inch Metal Engineering Ruler Tool

The Authentic Pocket Engineer: Small 3 Inch Metal Engineering Ruler Tool

  • Compact Multitool for Engineers: Includes ruler, protractor, compass
  • Built-in Engineering Reference: Quick access to equations and constants
  • Precise Measurement Tools: Accurate for drawings and fieldwork

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?

According to the whitepaper, the model’s behavior is heavily influenced by the surrounding harness, prompts, tools, and verification processes. These elements shape how the model is used and how its outputs are interpreted and applied.

What is the main implication for AI teams?

AI teams should focus more on designing and owning their harnesses and context frameworks rather than solely chasing the latest models. This approach offers more durable performance improvements and cost savings.

Does this mean models are no longer important?

Models remain a critical component, but their role is now understood as part of a larger system where configuration, context, and verification play a bigger role in determining outcomes.

How will this affect AI development costs?

Investing in harness and context engineering can reduce long-term costs by minimizing token waste, improving security, and decreasing reliance on frequent model upgrades.

Is this shift already happening in industry?

Some organizations are beginning to adopt this perspective, but widespread change will depend on further validation, tooling development, and industry consensus.

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

15 Best Graphics Cards for Gaming, AI, and Creative Work in 2026

Explore the 15 best graphics cards of 2026 for gaming, AI, and creative tasks. Find out which models suit your needs and what remains uncertain.

Superior Shooting With 6 AI-Enhanced Camera Lenses In 2026

In 2026, a new lineup of six AI-enhanced camera lenses offers improved image quality and versatility, transforming professional and amateur photography.

Decoding GTA 6: Pricing, Launch & Trends Based On Signal Monitoring

Recent signal monitoring suggests GTA 6’s pricing, release date, and pre-order details. Here’s what is confirmed and what remains uncertain.

Apple Silicon’s Quiet Memory Advantage

Apple Silicon’s unified memory architecture allows running large AI models beyond traditional GPU limits, offering capacity at lower cost and power.