DoneThat
Back to AI Adoption Guide

Agentic Engineering

Agentic engineering lets AI draft, refactor, and explain code for you, with the strongest results on boilerplate, repetitive implementation, and well scoped changes that you can still review and maintain.

Overview

Agentic engineering lets AI draft, refactor, and explain code for you, with the strongest results on boilerplate, repetitive implementation, and well-scoped changes that you can still review and maintain.

What it is, how it works

  • The most successful use case of AI so far.
  • Let AI write code for you, often better than you could.
  • Works best for boilerplate things.

When not to use it

  • When you can't read code and don't understand what's happening.
  • When you aren't prepared to test and maintain all the extra code.
  • When there is an existing SaaS you could use instead.

How to get started

  • Build trust slowly, review all code to understand risks
  • Compare models, best performing is always changing
  • Right now, Codex has the best model, Cursor the best ID, and Claude the best vibes.

Best practices / further reading

  • Watch token usage, pricing models are changing
  • Use an AGENTS.md file to give context on your repos
  • Related: Vibe Coding