Vibe Coding vs Agentic Engineering: The Blur Lines Are Getting Uncomfortable

Simon Willison, the creator of Datasette and long-time Python developer, recently joined Joseph Ruscio on Heavybit's High Leverage podcast (Ep. #9) to discuss AI coding tools. In a follow-up blog post on 6th May 2026, he shares a disturbing realization: the lines between vibe coding and agentic engineering are blurring in his own daily work.

Vibe coding vs agentic engineering: original definitions

Willison previously drew a sharp distinction between the two:

• Vibe coding: Non-programmers or programmers who don't review the code. You ask for something, get something, and if it breaks you cross your fingers. Fine for personal tools where bugs only hurt you — grossly irresponsible for production software used by others.
• Agentic engineering: Professional software engineers using AI to build higher-quality production systems faster, while still applying their expertise in security, maintainability, and operations. Every line is reviewed.

Where the convergence happens

The problem is that coding agents like Claude Code have become reliable enough that Willison now finds himself skipping code review for routine production tasks. His example:

Ask Claude Code to build a JSON API endpoint that runs a SQL query and outputs the results as JSON.
It's just going to do it right. It's not going to mess that up.
You have it add automated tests, you have it add documentation, you know it's going to be good.

He admits he feels guilty about this: “If I haven't reviewed the code, is it really responsible for me to use this in production?”

The mental model: trusting an agent like a team

Willison compares this to working at a large organization where another team provides an image resize service. You don't read every line of their code — you read the docs, test the service, and treat it as a semi-black box until you hit problems. He's now treating AI agents the same way.

But he notes an uncomfortable difference: “Claude Code does not have a professional reputation! It can't take accountability for what it's done. But it's been proving itself anyway — time and time again it's churning out straightforward things and doing them right in the style that I like.”

He calls this a form of “normalization of deviance” — the gradual acceptance of lower scrutiny as the tool keeps proving itself.

Takeaway for developers using AI agents

Willison's post is a practical caution: as agents become more reliable, the temptation to skip review grows. His advice is not to stop using agents, but to remain conscious of where you're cutting corners. For production systems that affect others, you still need a human who understands the code and can be held accountable.