How Claude Code runs across a team
Pick a workflow on the left to watch agents hand work to each other
One setup, every developer, same rules
When each engineer configures Claude Code alone, you get inconsistent code and no control over what it runs. This map shows the shared setup we install: a CLAUDE.md with your standards, a settings.json that blocks risky commands, hooks that check work before and after each edit, and specialist agents that take a bug from report to tested PR. Start with “Bug Fix Pipeline”, or open any folder.
Departments = groups of Claude Code capabilities, not your org chart. “Management” is the main session that plans and delegates; “QA” is the testing and review agents.
Per developer
- my-prompts.txt
- notes.md
- (no rules)
3 setups, 3 rule sets
One shared repo setup
- CLAUDE.md
- settings.json
- hooks/
Same rules for Dev 1, Dev 2, Dev 3
Company map: click a capability group
Run a workflow
Claude Code for teams
Claude Code for companies: the setup we install
Inconsistent output →
A shared CLAUDE.md with your architecture, conventions and review rules.
Risky commands →
settings.json allow/deny lists (e.g. deny rm -rf, git reset --hard).
Unreviewed changes →
Hooks that run formatters, linters and tests after each edit.
Only autocomplete →
Multi-agent workflows: triage, fix, test, document, open the PR.
No one to set it up →
ATI Lab (official Anthropic partner) installs it on one team first, then rolls it out.
Questions teams ask
Is this only for engineering?
Claude Code runs in your repo and terminal, so engineering, platform and QA benefit first. For ops and finance teams, see our Claude for business rollout. Claude for business →
Can we stop it touching production?
Yes. Tool permissions and hooks block the commands and paths you list, before they run.
How long does setup take?
We start with one team and its first workflows. First deployments typically take 6–12 weeks, then the same setup rolls out to other teams.