Austin Wallace — Data Engineer
Minimal Version 🔌 AWS
← Back to blog

Proposing a Taxonomy of Agentic Engineering

Chart showing grokking in modular addition - train loss drops immediately while test loss stays high for thousands of epochs before suddenly collapsing

The Problem

The Thesis


The Three Knobs

Stakes

Verifiability

Grokking


The Taxonomy

Vibe Coding

Vibe Engineering

Agentic Engineering

Human-Driven Development

Hand Coding


How to Choose


Grokking: The Real Bottleneck


Closing


Easy

High | Human-driven | Agentic Eng | Stakes | (or invest in | (review | | verification) | critical | | | paths) | |---------------|-------------| Low | Vibe Eng | Vibe Coding | | (guardrails) | (YOLO) |


Grokking overlays as a gradient or badge: “Grokking required” increases toward top-left

Could render this as an actual component with hover states.

OPTIONAL: Decision Quiz Widget

Three sliders:

  • Stakes: Low / Medium / High
  • Verifiability: Easy / Medium / Hard
  • Grokking: Strong / Moderate / Weak

Output: Recommended mode + caveats

This is more work but very on-brand for the “tools for deciding” goal. —>

Base text written by Austin, enhanced by various AI tooling providers, and deployed with SST.