Velocity Framework
AI Delivery Stack

AI-Native IDEs

Capabilities we expect from AI-powered coding environments (with Cursor as a reference example).

The Velocity Framework assumes engineers work in an AI-native IDE: an environment where the model is integrated into navigation, planning, refactoring, and review—not just autocomplete.

Cursor is our current reference implementation of an AI-native IDE, but the principles on this page are tool agnostic. If another editor can provide the same capabilities, the framework still applies.

Required Capabilities

An IDE is considered "AI-native" in the context of Velocity if it can:

  • Understand the workspace
    Index the codebase so the model can navigate types, modules, and dependencies, and answer questions grounded in real files.
  • Generate and execute plans
    Turn a spec or description into a step-by-step change plan tied to concrete files and functions, then help execute that plan incrementally.
  • Work with diffs, not just text
    Propose edits as structured diffs that engineers can review and modify before applying.
  • Integrate external context
    Connect to repos, schemas, docs, and trackers via something MCP-like, so engineers don’t have to copy-paste large amounts of context.

If your IDE cannot do these things, you will struggle to apply the Velocity workflow effectively.

How Engineers Use an AI-Native IDE

Within the 6-step workflow, the IDE supports:

  • Define & Clarify: Asking questions about existing code and surfacing related files.
  • Design & Spec: Drafting specs from tickets and code, then refining them with the engineer.
  • Plan: Generating and iterating on an executable plan for a change.
  • Implement & Review: Applying small, reviewable diffs with clear context.
  • Test & Finalize: Helping generate tests, explain failures, and draft PR descriptions.

The key pattern is human-led orchestration:

  1. The engineer provides intent, constraints, and relevant files/specs.
  2. The IDE proposes plans and code changes.
  3. The engineer inspects diffs, adjusts prompts, and edits the result.

Cursor as a Reference Example

Cursor today provides these capabilities out of the box:

  • Workspace-aware chat bound to the current project.
  • A dedicated planning surface that can generate multi-file change plans.
  • AI-assisted diffs for new code, refactors, and reviews.
  • Support for MCP-style context integrations.

The Framework does not require Cursor specifically, but it is the tool we currently standardize on internally. Other IDEs may be adopted in the future as long as they can support the same patterns and constraints.