How to Migrate Off Claude Code: A Practical Vendor-Neutral Guide
Microsoft is ending Claude Code access for 200,000 engineers by June 30. Here's exactly how to move your unattended coding workflow to a vendor-neutral setup — in one evening, with no lock-in to any single AI provider.
Codeberg-first
Ralph Workflow is free and open source. Inspect the primary repo on Codeberg before you install — or jump to the GitHub mirror.
Microsoft's decision to end internal Claude Code access by June 30, 2026 puts a hard deadline on 200,000 engineers. But the lesson extends beyond Microsoft — any team that ties its development workflow to one AI provider is one policy change, price hike, or deprecation away from disruption.
The fix isn't picking a different single vendor. It's making your workflow agent-agnostic — so the AI model underneath can change without changing how you work.
Here's how to do it in one evening.
What you're replacing
If you currently use Claude Code for overnight or unattended coding work, you probably have something like:
# Your current overnight script
claude --project ./my-repo \
"Implement the feature described in spec.md. \
Run tests, fix failures, commit passing code."
This works until Claude Code isn't available. After June 30 — at Microsoft and at any org that follows suit — it stops working entirely.
The migration: 3 steps
1. Install Ralph Workflow (one command)
pipx install ralph-workflow
Ralph Workflow is a free, open-source loop framework. It doesn't replace your AI agent — it orchestrates whichever agent is available, running it in a quality-gate loop (spec → implement → test → review → repeat) until the work passes.
2. Write your spec (5 minutes)
Move your task description from a CLI flag to a spec file. This is the same content — just saved as a durable artifact:
# my-repo/specs/feature-x.md
## Goal
Add user-visible dark mode toggle to settings
## Acceptance Criteria
- Toggle appears in Settings → Appearance
- Changes `data-theme` attribute on `<html>` element
- Preference persists across page reloads (localStorage)
- Existing tests pass
- New test for the toggle component
3. Run it overnight
ralph run specs/feature-x.md
Ralph picks up the spec, runs your configured AI agent, and loops until the acceptance criteria are met — or surfaces exactly what's blocking.
The default model is whatever you have configured. When APIs change, pricing shifts, or your org mandates a new provider:
# Switch the underlying model in one line
ralph config set model anthropic/claude-sonnet-4-5
# or
ralph config set model google/gemini-3-flash
# or
ralph config set model openai/gpt-5-codex
Your specs and workflow stay intact. Only the agent underneath changes.
What you get
| Before (single-vendor Claude Code) | After (vendor-neutral Ralph Workflow) |
|---|---|
| One model | Any model — switch in one line |
| CLI flag task description | Durable spec file (version-controlled) |
| Pass/fail output | Quality-gate loop: implement → test → review → retry |
| No history of what changed | Every commit attributed to a spec + agent |
| Stops if Claude Code is unavailable | Keeps working with whatever agent is available |
The real win: agency independence
The same spec that runs on Claude Code today will run on GPT-5 Codex tomorrow, Gemini 3 next month, or an open-weight model you self-host. This isn't theoretical — the Microsoft news confirms it's the difference between "ten days to figure out a migration" and "change one config line."
The Loop Engineering pattern — now independently converging across 23+ projects from individual hackers to JPMorganChase — emerged precisely because single-shot claude "do X" breaks the moment the agent underneath changes. A spec-driven quality-gate loop is the durable abstraction: the agent underneath is an implementation detail.
Start tonight
pipx install ralph-workflow
ralph init
# Pick one real backlog task. Write a short spec.
# Run it overnight.
ralph run specs/my-first-task.md
If you're at Microsoft, this is your alternative to rewriting every Claude Code script by June 30. If you're anywhere else, this is insurance against the next vendor pivot.
The repo is at codeberg.org/RalphWorkflow/Ralph-Workflow — free, open-source, no registration, no vendor lock-in, no telemetry. If it saves you an evening of migration work, star it so the next engineer finds it too.
Related: Microsoft Is Ending Claude Code Access — Why Vendor-Neutral AI Coding Matters Now · 23 Projects Reinvented the Same AI Coding Loop · Your First Overnight Task — Start Here
Related Posts
Microsoft Is Ending Claude Code Access — Here's Why Vendor-Neutral AI Coding Matters Now
Microsoft is directing its engineers away from Claude Code and toward GitHub Copilot CLI by June 30, 2026. If a 200,000-person org can have its AI coding tool changed by policy, your startup can have its stack disrupted by pricing, deprecation, or a vendor pivot. Ralph Workflow is a free, open-source loop framework that keeps your development workflow intact — no matter which AI agent runs underneath.
23 Projects Reinvented the Same AI Coding Loop — Here's What They All Got Right
Independent developers across GitHub and Codeberg built the same plan→build→verify architecture for AI coding agents. From ralphex (1,296★) to nightshift (14★), the loop pattern is converging into a standard. Here's every project, the architecture they share, and why AI agents perform better inside a structured loop.
Good vs Bad Unattended AI Coding Tasks: How to Know Before You Start
Not every backlog item works as an overnight coding run. Here is how to tell the good fits from the bad ones before you spend an evening setting it up.
Best evaluator path
Turn the idea into a real overnight test, not another saved tab.
Codeberg-first: open the primary repo, star it to track releases, choose one bounded backlog task, run it tonight, and ask one question tomorrow morning — would I merge this? GitHub stays available as the mirror.
Open the primary Codeberg repo
Read the public source before you install anything.
Pick a first task
Use the guide to choose a bounded backlog item that is honest to review.
Install and run Ralph Workflow
Keep the machine awake, then decide in the morning whether the diff is good enough to merge.