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

Taxonomy Components Demo

This is a demo to compare the two visualization approaches.


Option 1: Decision Matrix (2x2)

The matrix gives a quick visual for “where should I be?” based on stakes and verifiability:

Stakes
High Low
Human-Driven or invest in verification Grokking: Required
Agentic Engineering review critical paths Grokking: Recommended
Vibe Engineering invest in guardrails Grokking: Helpful
Vibe Coding ship it Grokking: Optional
Hard Easy
Verifiability

Reading the matrix:

  • High stakes + hard to verify = you need tight control (Human-Driven)
  • High stakes + easy to verify = let AI draft, review what matters (Agentic)
  • Low stakes + hard to verify = invest in guardrails (Vibe Engineering)
  • Low stakes + easy to verify = let it rip (Vibe Coding)

Grokking requirement increases toward the top-left. If you can't verify, the price is understanding.


Option 2: Taxonomy Cards

Cards give more detail per mode. Here’s the full set:

Vibe Coding

No-look shipping

You do
  • Prompt → run → accept
  • Trust the model did the obvious thing
You don't
  • Read the code
  • Review architecture

Visual check, quick run, 'does it work?'

  • Throwaway prototypes
  • One-off scripts
  • Zero-stakes experiments

Vibe Engineering

No-look, but with guardrails

You do
  • Invest in prompts, tests, and tooling
  • Define crisp acceptance criteria
  • Let the system verify for you
You don't
  • Read most of the code
  • Manually review every change

Tests, linters, type checks, self-checks

  • Things you care about but not enough to read every line
  • Unfamiliar stacks with verifiable behavior

Agentic Engineering

Co-authoring with selective review

You do
  • Write some code yourself
  • Review critical paths and architecture
  • Verify with tests AND reasoning
You don't
  • Let AI make unsupervised architecture decisions
  • Ship without understanding failure modes

Tests + mental model of invariants

  • User data involved
  • Correctness matters
  • Will be extended by future-you or teammates

Human-Driven Development

AI as autocomplete

You do
  • Drive design and implementation
  • Use AI for boilerplate and lookup
You don't
  • Prompt for large code blocks
  • Let AI make structural decisions

You are the verification

  • Delicate work requiring tight control
  • Learning and building muscles
  • Domains you don't trust AI in

Hand Coding

Pre-2022 mode

You do
  • Write everything yourself
You don't
  • Use AI beyond maybe search

You, exhaustively

  • Perfection required
  • Proving competency (interviews)
  • When time doesn't matter

Which to use?

Matrix is good for:

  • Quick orientation
  • The core decision logic at a glance
  • People who want the TL;DR

Cards are good for:

  • Full context on each mode
  • Reference material
  • People who want to understand before deciding

You could use both: matrix first for the decision framework, then cards below as the “full taxonomy reference.”

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