Vibe coding and agentic engineering are getting closer than I'd like

2026-05-06 · Source: Simon Willison's Weblog · Field: Technology & Digital — Software Development & Engineering, Artificial Intelligence & Machine Learning · Depth: Intermediate, medium

Summary

Simon Willison discusses the blurring lines between "vibe coding" and "agentic engineering" in AI-assisted programming, a realization stemming from a podcast discussion on May 6, 2026. Initially, vibe coding was seen as irresponsible for production due to lack of code review and accountability, while agentic engineering involved professional engineers using AI to build higher-quality systems faster. However, as AI coding agents like Claude Code become more reliable for routine tasks, Willison finds himself reviewing less code, leading to a "normalization of deviance." This shift challenges traditional software evaluation, where a GitHub repository's apparent quality can be AI-generated quickly. The increased code generation speed also shifts development bottlenecks, potentially reducing the need for extensive upfront design processes. Despite these changes, Willison believes software engineering careers are safe, as AI tools amplify existing expertise for a still "ferociously difficult" task, and proven, professionally managed software remains highly valued.

Key takeaway

For Machine Learning Engineers building production systems, you should critically assess your reliance on AI coding agents. While these tools boost efficiency, your accountability for code quality remains paramount. Consider treating AI-generated code as a "semi-black box" from a trusted internal team, focusing your review efforts on documentation, tests, and observed behavior rather than every line, and only deep-diving into the code when issues arise.

Key insights

The increasing reliability of AI coding agents blurs the distinction between casual "vibe coding" and professional "agentic engineering."

Principles

• Accountability is crucial for production software.
• AI tools amplify existing engineering expertise.
• Proven solutions are more valuable than quickly generated ones.

Method

Treat AI coding agents like trusted internal teams: review documentation and outcomes, digging into code only when problems arise, rather than line-by-line review.

In practice

• Re-evaluate design processes for reduced cost of error.
• Prioritize software usage history over superficial repository metrics.
• Focus on higher-level problem-solving with AI assistance.

Topics

• Vibe Coding
• Agentic Engineering
• AI Coding Tools
• Software Development Lifecycle
• Code Quality Evaluation

Best for: Machine Learning Engineer, Software Engineer, AI Engineer, Director of AI/ML

Related on AIssential

• Vibe Coding: The Complete Story (2015–2026) — The coding paradigm has shifted from human-driven to AI-agent-dominated, requiring new human oversight and expertise.
• Vibe Coding Is Great. Vibe Reviewing Is Terrifying. — AI-assisted code generation creates significant challenges for traditional code review processes and senior engineers.
• Agentic Engineering Is Not Vibe Coding: The Patterns That Actually Work — Distinguishing "vibe coding" from "agentic engineering" is crucial for effective AI-driven software development.
• Breaking the Spell of Vibe Coding — Vibe coding can induce a "dark flow" state, mimicking productive engagement while hindering genuine skill development and accurate self-assessment.
• SE Radio 711: Scott Hanselman on AI-Assisted Development Tools — Specificity and foundational knowledge are crucial for effectively steering AI-assisted development tools and managing inherent ambiguity.
• Vibe Coding Ate My Homework: An evaluation of AI approaches to greenfield software engineering and programming — Vibe coding's hands-off approach is unreliable for complex greenfield software, despite initial efficiency gains.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.