How to roll out Claude Code to your team
Rolling out Claude Code to a team takes six steps, in this order: make the five admin decisions, deliver and verify your managed settings, run a pilot, launch with a real first task, recruit champions, and measure adoption. Most rollouts that stall skip the first two steps and go straight to an announcement, so the first week turns into a support queue about logins and blocked hosts.
I call this the CCC Rollout Ladder. Each rung depends on the one below it. You cannot measure adoption you never enabled, and you cannot launch cleanly on settings you never checked. The rest of this post walks the ladder from the bottom.
Step 1: Make the five admin decisions first
Every Claude Code deployment answers the same five questions. Answer them in writing before you announce anything, because each one changes what developers see on their first day.
The five decisions behind a Claude Code rollout
| Decision | What you are choosing | Where to go deeper |
|---|---|---|
| API provider | Where developers sign in and how you are billed: Claude for Teams or Enterprise, the Claude Console, Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry | [Authentication](/blog/claude-code-authentication), [Bedrock](/blog/claude-code-bedrock) |
| How settings reach devices | Server-managed settings from the claude.ai admin console, an MDM plist or registry policy, or a managed settings file on disk | [Managed settings](/blog/claude-code-managed-settings) |
| What to enforce | Permission rules, sandboxing, which MCP servers and plugin marketplaces are allowed, which models people can pick, a minimum version | [Permissions](/blog/claude-code-permissions-explained), [sandbox](/blog/claude-code-sandbox-explained), [managed MCP](/blog/claude-code-managed-mcp) |
| Usage visibility | The analytics dashboard, OpenTelemetry export, and spend limits | [Analytics dashboard](/blog/claude-code-analytics-dashboard), [OpenTelemetry](/blog/claude-code-opentelemetry) |
| Data handling | What you tell people about where their code goes, plus zero data retention or HIPAA if you need them | [Zero data retention](/blog/claude-code-zero-data-retention) |
My default pick for most companies is Claude for Teams or Enterprise with server-managed settings. One subscription covers Claude Code and claude.ai, there is no infrastructure to run, and policy reaches every computer without a device management project. Pick a cloud provider when you need to stay inside existing AWS, Google Cloud or Azure billing and compliance controls.
Step 2: Deliver managed settings and verify them
Managed settings are your policy, and they win over anything a developer sets locally, apart from a few security exceptions. Claude Code checks four sources, from highest priority to lowest: server-managed settings from the admin console, an MDM plist or HKLM registry policy, a managed settings file on disk, and the Windows user registry. Server-managed settings are fetched at startup and refreshed hourly during a session.
- Start small. A first policy usually sets a minimum version, blocks bypass permissions mode, and restricts which MCP servers and plugin marketplaces people can add.
- Remember that list settings merge. Entries in permissions.allow and permissions.deny combine across sources, so developers can add to your lists but cannot remove from them.
- Do not rely on the Windows user registry (HKCU) for enforcement. Anyone can write to it without admin rights. Use HKLM, the plist, or the file.
- If you mix providers, use server-managed settings for claude.ai users plus a file or MDM fallback so everyone else still gets policy.
Then verify on a real computer. Have one developer open Claude Code and run /status. On the Status tab, the Setting sources line should read Enterprise managed settings, followed by the source that won. If it does not, fix it now. A policy you think is deployed but is not is worse than no policy, because you will plan around it.
Step 3: Run a pilot with people who will tell you the truth
Pick a small group who will use Claude Code on real work and report back honestly. Give them one ask: use it on at least one real task this week, then post what worked, what was annoying and what surprised them in a #claude-code-pilot channel. That feedback shapes the company-wide launch.
Tell pilots to start their first multi-file change in plan mode. Claude lays out what it intends to do without editing anything, which is the fastest way for a skeptical engineer to calibrate how much to trust it. In the terminal, press Shift+Tab until plan shows. In the desktop app, pick it from the mode selector.
The pilot is also where you find your install problems. A proxy that blocks a host, a certificate your TLS inspection breaks, a login method your managed settings reject: you want five people to hit those, not five hundred. The corporate proxy guide covers the network side.
Step 4: Launch with the six-item checklist
Do not send the announcement until each of these is true. Every missing item becomes a launch-day support thread.
Pre-launch checklist
| Item | Why it matters |
|---|---|
| A #claude-code channel exists and is linked in the message | Questions land in one place, and every answer helps everyone watching |
| The install path is tested on at least one computer in your environment | Catches proxy and firewall problems before everyone hits them at once |
| A data handling link is ready | "Where does my code go?" is the first reply you will get |
| One concrete first task from your own codebase | "Fix the flaky test in this file" gets people started. A generic example does not |
| A named owner watches the channel for the first 48 hours | Unanswered launch-day questions kill momentum |
| An executive sponsor sends or co-signs the announcement | It reads as a company priority instead of an optional experiment |
Keep the announcement short. Say what Claude Code is in one line: an agent that reads your actual codebase, edits files and runs commands, not autocomplete. Give the install path, the first task, the data handling answer and the channel. That is the whole message.
On the install path, lead with the desktop app. In the Claude desktop app, Claude Code is the Code tab: open a project folder and start working, no terminal needed. Mention that the terminal, the VS Code extension and the JetBrains plugin run the same agent, for the engineers who prefer them. The desktop app guide is a good link to include.
Answer "where does my code go?" before anyone asks
Put the answer in the announcement itself. On Team, Enterprise, Claude API and cloud provider plans, Anthropic does not train models on your code or prompts. Your provider choice decides retention and compliance posture, and Claude Code sends the context it needs to the API for each request. If you need nothing stored after a request completes, qualified Enterprise accounts can get zero data retention.
Decide who answers the follow-up questions. Champions should not improvise data handling answers. They point to your internal policy page or to the admin, and the admin owns the answer. If you need a per-request audit log, plan a gateway between developers and the provider, which the LLM gateway post covers.
Fix the three errors you will see on launch day
Common launch-day problems
| What the developer sees | Fix |
|---|---|
| "You haven't been added to your organization yet" | Their seat does not include Claude Code access. Update the seat in the admin console |
| Signed in to the wrong account or organization | Run /logout, then /login, and pick the company account |
| The enterprise sign-in option is missing | Run claude update, then restart. In the desktop app, update the app |
Pin this table in the channel before launch. Most day-one messages are one of these three, and a pinned answer saves the channel owner from typing the same answer over and over.
Step 5: Recruit champions and run four weeks
Adoption rarely comes from the announcement. It comes from a few engineers using Claude Code well in public. After launch, message the two or three people posting the most in #claude-code and ask them to make it semi-official: keep posting what works, answer questions with the actual prompt they used, and host a weekly thread.
A four-week champion rhythm
| Week | What happens | Sign it is working |
|---|---|---|
| 1 | Seed the channel: pin the quickstart and post two or three real examples with the prompts included | A few replies and at least one question |
| 2 | Start a Friday thread asking what Claude helped with this week, and share one skill or CLAUDE.md snippet | Someone other than the champion posts an example |
| 3 | Offer a few 15-minute pairing sessions and pin an FAQ of the common questions | The same people come back, not just first-time tries |
| 4 | Name a second champion and send the admin a short summary of what works and what does not | Other people answer questions in the channel |
One shortcut worth sharing with champions: run /team-onboarding in a project they have used a lot. Claude scans their recent sessions, commands and MCP servers and writes a setup guide a new teammate can paste in as their first message. Pin it in the channel. A shared skill helps the same way, and these are the skills worth installing first.
Step 6: Measure adoption instead of guessing
On Team and Enterprise plans, the analytics dashboard at claude.ai/analytics/claude-code shows adoption and contribution metrics with a leaderboard. Per-user usage and spend come from the spend report in your organization's analytics settings, not the dashboard. On the Console, usage and spend per user live at platform.claude.com/claude-code. For anything custom, OpenTelemetry export works on every provider.
- Check who is active at the end of week one and week four. A big gap means people tried it once and stopped. Point champions at them.
- Read the spend report next to the adoption numbers, so you have an answer before finance asks. Set spend limits before you need them, which the team costs post covers.
- Share the numbers with the executive sponsor. It keeps the rollout a priority after the launch email is forgotten.
Short, practical drops on skills, MCP, agents, prompts, and more. No spam, unsubscribe anytime.
Frequently asked questions
How do I roll out Claude Code to my team?
Make five admin decisions first (provider, how settings reach devices, what to enforce, usage visibility and data handling), deploy managed settings and verify them with /status, run a pilot, launch with a tested install path, a real first task and a channel with an owner, then recruit champions and measure adoption.
Should we roll out Claude Code to everyone at once or start with a pilot?
Start with a pilot. A small group finds the proxy, certificate and login problems before the whole company does, and their honest feedback shapes the launch message.
Does Anthropic train on our code if we roll out Claude Code?
On Team, Enterprise, Claude API and cloud provider plans, Anthropic does not train models on your code or prompts. Your provider choice decides retention and compliance posture.
Do developers need the terminal to use Claude Code?
No. In the Claude desktop app, Claude Code is the Code tab. The terminal, VS Code extension and JetBrains plugin run the same agent for people who prefer them.
What does "You haven't been added to your organization yet" mean?
The developer's seat does not include Claude Code access. An admin needs to update that seat in the admin console.
How do I check that managed settings reached a developer's computer?
Have them run /status in Claude Code. The Setting sources line should show Enterprise managed settings, followed by the source that won.
Last reviewed by David Iya on November 12, 2026
