Choosing the right AI approach

Don't use a fully autonomous agent to write a one-line regex.

The idea

Using AI effectively means matching the tool to the task's complexity. We have a spectrum of tools: Autocomplete (Copilot), Chat (ChatGPT/Claude), Agents (Devin/AutoGPT), and Do-It-Yourself (Human).

Your choice depends on three factors: Ambiguity (how well-defined is the task?), Blast Radius (how dangerous is a mistake?), and Novelty (has this been done 1,000 times on the internet?).

Ambiguity Clear Spec

Blast Radius Safe (UI)

Novelty Boilerplate

Autocomplete
(Copilot)

Chat / Spec
(Claude)

Autonomous
Agent

Human
(DIY)

Low complexity across the board. Use Autocomplete to write it faster.

How it works (The Decision Matrix)

# 1. Autocomplete (Inline IDE tools)
# Best for: Boilerplate, standard algorithms, typing out tests.
# Worst for: Architecture design, complex multi-file refactors.

# 2. Chat / Code Gen (ChatGPT, Claude)
# Best for: "Write a React component for X", "Explain this bug".
# Requires: A human to paste the code, verify it, and wire it up.

# 3. Autonomous Agents (Devin, etc)
# Best for: "Migrate these 50 files from Python 2 to 3".
# Requires: Low ambiguity (clear tests), but high manual labor.

# 4. Human (Do it yourself)
# Best for: Highly novel, proprietary business logic, or extreme
# blast radius (e.g., rewriting the payments processing core).