The New SDLC With Vibe Coding

Google published «The New SDLC With Vibe Coding» in May 2026, the opening piece of its "Day 1" series, credited to «Addy Osmani», Shubham Saboo and Sokratis Kartakis. Its core assertion concerns the interface rather than the technology: developers move from writing code to expressing intent, and intelligent systems translate that intent into working software. The framing figures are 41% of new code generated by AI, 85% of professional developers using coding agents regularly, 51% of them daily.

The authors place «vibe coding» and agentic engineering on one spectrum, graded by how much structure, verification and human judgment surrounds the output. Tests cover deterministic behaviour; evals (labelled datasets, rubrics, LM judges) cover the non-deterministic part. «Context Engineering» becomes the central skill, split between static context that is always loaded and costly (AGENTS.md, CLAUDE.md, GEMINI.md) and dynamic context fetched on demand, with Agent Skills as its main pattern. The paper's central equation is Agent = model + harness, the model weighing roughly 10% and the harness roughly 90%. On Terminal Bench 2.0, changing only the harness moved an agent from outside the Top 30 into the Top 5. "Most agent failures are configuration failures."

Compression across the lifecycle is uneven: implementation drops from weeks to hours while requirements, architecture and verification stay at human pace. SFEIR's eleven-phase cycle, written on 16 June 2026, cites the whitepaper as evidence of industry convergence. Its closing line states the remaining problem: "Generation is solved. Verification, judgment, and direction are the new craft."