feat: AI-powered Changesets workflow for monorepo releases

[activadee]

Problem

Monorepo release management has two main approaches, each with trade-offs:

Approach	Pros	Cons
semantic-release	Fully automated, zero friction	Version spam, all packages bump together
Changesets	Explicit intent, independent versions, better changelogs	Manual step = friction = adoption barrier
Release Please	PR-based, good monorepo support	Changelog conflicts, complex config

Proposal: AI-Enhanced Changesets

Combine Changesets' explicit release model with AI automation to eliminate friction.

Workflow

PR opened/updated
       ↓
AI analyzes: diff, commit messages, PR description
       ↓
Generates .changeset/<random>.md:
  ---
  "cli": patch
  ---
  Fixed download retry logic for rate-limited requests
       ↓
Commits to PR branch (or suggests via comment)
       ↓
Human reviews/approves changeset
       ↓
Merge → Changesets bot handles release PR

AI Responsibilities

1. Scope detection: Determine which packages changed (cli, server, core, etc.)
2. Version inference: Suggest patch/minor/major based on conventional commits
3. Changelog generation: Write user-facing summaries (not dev-speak)
4. Smart batching: Group related changes into single changeset

Implementation Options

1. GitHub Action - Runs on PR events, commits changeset file
2. GitHub App/Bot - Comments with suggested changeset, author approves
3. CLI wrapper - npx ai-changeset generates from recent commits

Prior Art

• Changesets - Core library
• semantic-release - Commit-based automation
• Release Please - Google's PR-based approach

Context

This idea emerged from experimenting with Release Please for kanban-ai monorepo. Release Please had merge conflicts and complex configuration. Changesets has the right model but manual friction is a barrier.

AI can bridge this gap.

---

Labels: enhancement, ai, release-automation