📊 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 new Google whitepaper emphasizes that the key to effective AI development is not the AI model’s size but the surrounding harness and context engineering. This shift impacts how companies should allocate resources and develop AI systems.

A new Google whitepaper asserts that the dominant factor in AI development is not the size of the model but the harness and context engineering surrounding it. This revelation challenges common assumptions and suggests a strategic shift for developers and organizations investing in AI, emphasizing the importance of configuration, verification, and contextual scaffolding over raw model improvements.

The paper, authored by Addy Osmani, Shubham Saboo, and Sokratis Kartakis, states that only 10% of AI behavior is determined by the model itself, with the remaining 90% shaped by the harness — including prompts, tools, rules, and context policies. Evidence from benchmarks shows that changing the harness can significantly improve performance without upgrading the model, exemplified by a team moving from outside the Top 30 to the Top 5 on a benchmark by only reconfiguring their setup.

The authors distinguish between ‘vibe coding,’ characterized by minimal structure and rapid iteration, and ‘agentic engineering,’ which involves formal specifications, automated tests, and oversight. They argue that the shift towards structured, verified workflows is crucial for cost efficiency and security, as ad-hoc prompting can lead to higher operational costs and vulnerabilities over time.

The paper emphasizes that organizations should focus on building durable, configurable scaffolds—what it calls the ‘harness’—and on developing skills in context engineering, which involves loading relevant knowledge, tools, and guardrails dynamically, rather than relying solely on the latest model advancements.

At a glance
reportWhen: published early 2026
The developmentThe whitepaper by Addy Osmani, Shubham Saboo, and Sokratis Kartakis highlights that the most significant change in SDLC is focusing on verification, judgment, and context rather than model size.
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 Model Size Is Not the Key in AI Development

This shift redefines how organizations should approach AI investments. Instead of chasing the latest, largest models, companies will benefit more from investing in the development of robust harnesses and context management. This approach can reduce costs, improve security, and enable more reliable AI systems, fundamentally changing the traditional SDLC in AI projects.

Amazon

AI model configuration tools

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Background of the AI Development Paradigm Shift

Historically, AI progress has been driven by larger and more powerful models. However, recent industry experiments and benchmark results indicate that configuration, scaffolding, and context engineering can dramatically influence performance. The whitepaper builds on these insights, framing the current era as one where the ‘model’ is only a small part of the overall system—an idea that challenges the conventional focus on model size and complexity.

This perspective aligns with ongoing industry trends towards more disciplined, verified AI workflows, and the increasing importance of cost management and security in AI deployment.

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

— Addy Osmani

Amazon

AI prompt engineering software

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Unclear Aspects of Implementing the New SDLC

While the paper presents compelling evidence and a clear framework, it remains unclear how organizations will effectively transition from traditional model-centric approaches to harness-centric workflows at scale. Specific strategies for retraining teams, reconfiguring existing systems, and measuring success are still emerging.

Additionally, the long-term impact on model development cycles and the competitive landscape is not yet fully understood, as many organizations are still adapting to rapid AI advancements.

Amazon

AI verification and testing tools

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Next Steps for Organizations Adapting to the Shift

Organizations should start by auditing their current AI workflows, focusing on configuration, verification, and context management. Developing expertise in harness and context engineering will be critical. Industry leaders may also begin experimenting with benchmark-driven reconfiguration to improve performance without model upgrades. Monitoring emerging best practices and tools for scalable context management will be essential as this paradigm gains traction.

Amazon

AI development workflow software

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 AI behavior?

The whitepaper cites evidence that most of an AI system’s behavior depends on how it is configured, scaffolded, and the context it is given, rather than the raw size or complexity of the model itself.

How does this shift affect AI development costs?

Focusing on harness and context engineering can lower long-term costs by reducing token consumption, improving security, and decreasing maintenance efforts, despite higher initial investment in configuration and testing.

What skills should AI teams develop now?

Teams should focus on skills in context engineering, system configuration, automated testing, and security practices related to AI workflows.

Will this change the competitive landscape?

Yes, organizations that excel at harness and context engineering may gain a significant advantage over those relying solely on larger models, shifting the industry focus from model size to system configuration.

Source: ThorstenMeyerAI.com

You May Also Like

Best Code Review Tools For Developers Compared

Compare popular code review tools based on features, usability, integration, and cost to find the best fit for your development team.

RMarkdown and Knitr: Simplify Your Statistical Reports

Gaining mastery with RMarkdown and Knitr streamlines your reports, making complex analysis presentation effortless—you’ll want to learn more to unlock their full potential.

How Much Backup Storage Do Researchers Really Need?

Just how much backup storage do researchers truly require to safeguard their vital data? Discover the surprising factors that determine your needs.

Briefro: A Document That Tells The Truth

Briefro introduces an AI-powered document platform that guarantees data integrity, privacy, and brand consistency, running entirely on user hardware.