← Back

The AI Development Tools Ladder

February 19, 2026

Over the past year, I've made the following observations about AI tools in software development. I've identified eight steps that engineers typically go through, in this order:

1. External AI Chat
   The engineer uses an AI chat tool (like ChatGPT) to help with problem-solving.
2. IDE AI-Driven Code Completion
   The engineer uses AI-driven code completion to write boilerplate quickly.
3. IDE AI Chat
   The engineer uses an IDE with a built-in AI chat feature, enabling faster queries and providing the AI with more context about the codebase.
4. IDE AI Agent
   The engineer uses an AI agent in the IDE to handle more complicated tasks (e.g., "Implement change X in module A"), while still supervising closely.
5. Local AI Agent
   The engineer uses a CLI to spin up agents locally to perform various tasks.
6. Cloud AI Agent
   The engineer uses cloud agents to avoid running out of local resources and to work in a consistent environment every time. This also reduces the need to duplicate repositories and setups locally.
7. Automated Cloud AI Agent
   The engineer uses cloud agents to solve tasks like reviewing code, picking up errors from an error tracker, managing upgrades, doing translations, and more. These agents can run even when no one is working and create pull requests for review.
8. AI Agent Orchestration
   The engineer builds an agent that can orchestrate other agents.

The Biggest Leap

When an engineer reaches step 4 and wants to jump to step 5, it's the biggest leap. That's because it involves transitioning away from the IDE and into a more terminal-like workflow (which can vary).

That's why I see this as two separate groups: the first group is heavily IDE-driven, while the second group (steps 5–8) is more abstract.

Using this ladder, engineers can take one step at a time and learn the tools at each level. Once a step is mastered, the new knowledge can be applied whenever it makes sense.