Coding agents have moved from autocomplete to doing whole tasks. When Anthropic analysed 500,000 interactions for its Economic Index report on software development (April 2025), 79% of Claude Code conversations were classed as automation, where the AI performs the task directly, against 49% on Claude.ai. That shift is why rollout matters: an agent that edits files and runs commands needs rules before it needs prompts.
This guide is for owners, engineering leads and IT managers rolling Claude Code out across a UK team: plans and costs, setup, the features that encode your way of working, security settings and UK GDPR. Facts come from Anthropic's documentation or the named source, checked on 16 November 2026.
- Claude Code needs a Pro, Max, Ultra, Team or Enterprise plan, or a Console account; the free plan does not include it.
- Team and Enterprise plans run under commercial terms: Anthropic does not train on your code, and standard retention is 30 days.
- CLAUDE.md is guidance, not enforcement. Use hooks and deny rules for anything that must happen every time.
- Recent versions start sessions in auto mode, so decide your permission policy and enforce it with managed settings.
- Anthropic's own figures put enterprise spend at around $13 per developer per active day.
What is Claude Code?
Claude Code is Anthropic's agentic coding tool: it reads your codebase, edits files, runs commands and works with your development tools. It runs in the terminal, in VS Code and JetBrains, in a desktop app and in the browser, and you can mention Claude in Slack with a bug report and get a pull request back.
It is a working example of agentic AI, and not only for developers: people who maintain internal tools or a vibe-coded app use it too. Anthropic's data shows how hands-off that use already is:
Share of conversations by interaction pattern, sampled 6–13 April 2025
| Pattern | Share |
|---|---|
| Claude Code, automation | |
| Claude.ai, automation | |
| Claude Code, directive | |
| Claude.ai, directive | |
| Claude Code, feedback loop | |
| Claude.ai, feedback loop |
Source: Anthropic, Economic Index: AI's impact on software development, 2025
Startup work made up 32.9% of Claude Code conversations and enterprise work 23.8%. Our Codex vs Claude Code comparison covers how it stacks up against OpenAI's agent.
How much does Claude Code cost for a UK business?
Claude Code is included in Anthropic's paid plans rather than sold separately, from $20 a month for one person on Pro and $20 per seat a month on Team billed annually. Prices on claude.com/pricing are listed in dollars and, in Anthropic's words, "Prices shown don't include applicable tax", so add VAT where it applies to your business and check the billing currency at checkout.
Lower figure billed annually, higher figure billed monthly; one figure where only one is listed
| Plan | From – To |
|---|---|
| Pro | |
| Team, standard seat | |
| Enterprise seat, plus usage | |
| Max | |
| Team, premium seat | |
| Ultra |
The Team plan is for teams of 2 to 150 and lists "Claude Code and Claude Cowork" among its features; premium seats give "5x more usage than standard seats". Enterprise is listed at US$20 per seat a month billed annually, with a 20-seat minimum, plus usage that "scales with model and task". On Team and Enterprise, each member's Claude Code usage draws on a seat allowance shared with Claude chat and Cowork, which resets on a rolling five-hour window and a weekly window. Admins can add usage credits with spend limits per organisation, group or member.
Alternatively, pay per token through an Anthropic Console account, or run Claude Code through Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry. For budgeting, Anthropic publishes average enterprise figures:
- $13 average cost per developer per active day across enterprise deployments
- $150–250 average cost per developer per month
- 90% of users stay below $30 per active day
Source: Anthropic, Claude Code docs: Manage costs effectively, checked November 2026
Anthropic advises a small pilot group first to set a baseline. For most small teams, standard Team seats plus premium seats for the heaviest users are the simplest start.
Which Claude plan should a business use for Claude Code?
Use Team or Enterprise for company code. Pro, Max and Ultra include Claude Code but run under consumer terms with no organisation to manage. Team and Enterprise add commercial data terms and spend controls, and Enterprise offers zero data retention for qualified accounts.
| Feature | Pro, Max, Ultra | Team | Enterprise |
|---|---|---|---|
| Claude Code included | Yes | Yes | Yes |
| Commercial terms, no training | No, consumer setting | Yes | Yes |
| Standard data retention | 30 days, or 5 years if training on | 30 days | 30 days |
| Zero data retention | No | No | Partial, qualified accounts |
| Org spend limits | No | Yes | Yes |
| Per-user usage reporting | No | Partial, CSV export | Yes, Analytics API |
Sources: Anthropic, Claude Code docs: Data usage; Anthropic, Claude Code docs: Manage costs effectively, checked November 2026
How do you set up Claude Code for a team?
Install Claude Code with Anthropic's native installer, which updates itself, sign everyone in through the company plan, then commit shared instructions and settings to each repository. It runs on macOS 13 or later, Windows 10 1809 or later, Ubuntu 20.04+, Debian 10+ and Alpine 3.19+, with at least 4 GB of RAM.
For teams, the stable release channel uses a version "typically about one week old, skipping releases with major regressions", and managed settings can enforce a minimum version. A sensible rollout:
- Choose the plan, seat mix and sign-in method
- Install with the native installer on the stable channel
- Commit CLAUDE.md, settings and deny rules to each repository
- Add team skills, hooks and MCP servers at project scope
- Enforce policy centrally with managed settings
- Pilot on real tasks, then review spend and settings after two weeks
What is the difference between CLAUDE.md, skills, hooks and subagents?
CLAUDE.md is always-on guidance, skills are knowledge and workflows loaded on demand, subagents are isolated workers that return a summary, and hooks are automation that fires on events without Claude deciding. Anthropic's guide to extending Claude Code sums it up: CLAUDE.md handles always-on context, skills on-demand knowledge, MCP external connections, subagents isolation and hooks automation.
| Feature | Loads every session | Runs without Claude deciding | Shared team location |
|---|---|---|---|
| CLAUDE.md | Yes | No, guidance only | CLAUDE.md in the repo |
| Skills | Partial, descriptions only | No, Claude applies them | .claude/skills/ |
| Subagents | No, when spawned | No | .claude/agents/ |
| Hooks | No, on trigger | Yes, fires on its event | .claude/settings.json |
| MCP servers | Partial, tool names only | No | .mcp.json |
Source: Anthropic, Claude Code docs: Extend Claude Code, checked November 2026
Plugins bundle skills, hooks, subagents and MCP servers into one installable unit, so a new repository or starter gets the same setup in one step.
What is CLAUDE.md and why does it matter?
CLAUDE.md is a markdown file of persistent instructions that Claude Code reads at the start of every session; commit one to every repository. Files can sit at several levels: a managed policy file deployed by IT, a personal file in your home folder, a project file in the repository and a CLAUDE.local.md for notes you do not commit. If a repository has an AGENTS.md and no CLAUDE.md, Claude Code reads the AGENTS.md instead, which helps teams that also use Codex.
Keep it short: how to run the app and tests, folder structure, conventions, what not to touch and how changes are reviewed. Anthropic suggests under 200 lines, because longer files reduce adherence. Remember the limit: Claude treats CLAUDE.md "as context, not enforced configuration". A rule that must hold every time belongs in a hook.
What are Claude Code skills?
A skill is a folder with a SKILL.md file of instructions, plus any scripts or reference files, that Claude uses when a task matches its description or when you type its name, such as /deploy. Only the description loads into each session; the full content loads when it is used, so long reference material costs almost nothing until needed. Project skills live in .claude/skills/ and are shared through the repository. Claude Code skills follow the open Agent Skills standard, and older custom slash commands have been merged into skills.
Good first skills are tasks people repeat and get slightly wrong: "add a page the way we do it", "write a database migration", "prepare a release note". Write descriptions carefully, because vague ones make Claude load the wrong skill.
What are Claude Code hooks?
Hooks are actions that run automatically at points in Claude Code's lifecycle, and they are how you turn a guideline into a guarantee. A hook can be a shell command, an HTTP request, an MCP tool call, a prompt to a model or a subagent. The hooks reference lists more than 30 events, including SessionStart, UserPromptSubmit, PreToolUse, PostToolUse, PermissionRequest and Stop.
Typical uses: lint every file Claude edits, block edits to a .env file, run the tests before Claude finishes, or post to Slack when a task ends. Take Anthropic's warning seriously: "Command hooks execute shell commands with your full user permissions." Review hooks like any other code and be wary of hooks in repositories you did not write. Administrators can set allowManagedHooksOnly in managed settings so that only approved hooks run.
What are subagents in Claude Code?
A subagent is a specialised assistant that runs in its own context window, with its own system prompt and allowed tools, and returns only a summary to the main conversation. Claude Code ships with built-in subagents, including a read-only Explore agent for searching code and a Plan agent for research in plan mode. Your own subagents live in .claude/agents/ for the project or in your home folder for all projects; organisation-wide subagents set in managed settings take priority.
A reviewer subagent with read-only tools and your checklist gives every change a consistent second look, and routing simple tasks to a smaller model such as Haiku keeps costs down. If you are building agents beyond coding, see our AI agent examples for UK small businesses and our AI agent development service.
Is Claude Code secure enough for business use?
Yes, if you configure it deliberately: permission modes, deny rules, sandboxing and managed settings give you layered control, and Anthropic's security documentation points to its SOC 2 Type 2 report and ISO 27001 certificate in the Anthropic Trust Center. One recent change matters: from Claude Code v2.1.283, auto mode is the built-in starting mode for interactive terminal and VS Code sessions, with a classifier reviewing actions instead of a person.
| Mode | Edits files without asking | Runs commands without asking | Use it for |
|---|---|---|---|
| default (Manual) | No | No, except read-only ones | Sensitive repositories |
| acceptEdits | Yes | Partial, file commands | Fast work in trusted repos |
| plan | No | Partial, read-only | Exploring before changes |
| auto | Yes, classifier-checked | Yes, classifier-checked | Routine work, fewer prompts |
| dontAsk | Partial, pre-approved only | Partial, pre-approved only | Unattended runs |
| bypassPermissions | Yes | Yes | Isolated containers or VMs |
Source: Anthropic, Claude Code docs: Configure permissions, checked November 2026
- Deny first. Rules "are evaluated in order: deny, then ask, then allow", and an allow rule cannot carve an exception out of a deny. Deny reads of .env and secrets folders and destructive commands.
- Managed settings. Administrators can deploy settings that user and project settings cannot override, including turning off auto mode or bypassPermissions for the whole organisation.
- Sandboxing. Shell commands can run inside an operating-system boundary, Seatbelt on macOS and bubblewrap on Linux, with filesystem and network isolation.
- MCP servers. Anthropic does not security-audit MCP servers, so connect only ones you trust. Our MCP guide explains the risks, and our MCP server development team builds least-privilege servers for your own systems.
For a wider framework, the NCSC's Guidelines for secure AI system development (November 2023) cover secure design, development, deployment, and operation and maintenance, following a "secure by default" approach. Keep production credentials off machines where the agent runs and keep human review as the gate. Our security checklist for AI-built apps covers what to check before anything ships.
What does UK GDPR mean for Claude Code data?
Under UK GDPR your business is normally the controller for any personal data in prompts, code or logs, so choose a plan whose terms match your data. On Team, Enterprise and the API, Anthropic's data usage policy says it "does not train generative models using code or prompts sent to Claude Code under commercial terms", and standard retention is 30 days. On consumer plans, retention is 30 days, or 5 years if you allow your data to be used for model improvement.
- Zero data retention is available to qualified Claude for Enterprise accounts, enabled per organisation by Anthropic's account team.
- Local transcripts are stored in plaintext under ~/.claude/projects/ for 30 days by default, so laptop encryption and the cleanupPeriodDays setting matter.
- Feedback transcripts sent with /feedback are retained for 5 years; set DISABLE_FEEDBACK_COMMAND=1 if staff should not send them.
- Telemetry metrics never include your code, prompts or file paths, and DISABLE_TELEMETRY=1 turns them off.
The ICO's guidance on AI and data protection is the UK reference point, though the ICO notes it is under review following the Data (Use and Access) Act. Keep real customer data out of development, record your plan and retention decision, and assess whether a data protection impact assessment is needed.
How should you train a team on Claude Code?
Train on your own code, with your CLAUDE.md, skills and permissions already in place. A practical format is a half-day session where each person completes a real backlog task with Claude Code, covering clear task writing, plan mode, diff review and when to do it by hand. Review after two weeks: which skills are missing, which hooks would have caught a mistake, and what the usage reports show.
Circlemount is based in Telford, Shropshire, and agrees the format with each team: remote sessions, in-person workshops or a mix. That suits teams across the West Midlands, including Birmingham, as well as businesses in Manchester, London and elsewhere in the UK. See our Claude Code training and setup service for what a programme covers, or our Claude Code agency for ongoing build work.
Frequently asked questions
How much does Claude Code cost for a UK business?
Claude Code is included in Anthropic's paid plans. In November 2026 its pricing page lists Team standard seats at $20 a month billed annually or $25 monthly, premium seats at $100 or $125, and Enterprise at US$20 a seat plus usage. Prices exclude applicable tax, so add VAT where it applies.
Can I use Claude Code on the free Claude plan?
No. Anthropic's setup documentation says Claude Code requires a Pro, Max, Ultra, Team, Enterprise or Console account, and the free plan does not include it. Businesses can also run it through Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry if they already have an agreement with one of those providers.
What is the difference between skills, hooks and subagents?
A skill is a SKILL.md file of instructions that Claude loads when a task needs it, such as how your team writes migrations. A hook runs automatically on an event, such as linting after every edit, without Claude deciding. A subagent is a separate assistant with its own context window and tools that returns a summary.
Does Anthropic train on code from Claude Code?
Not under commercial terms. Anthropic's data usage policy says it does not train on code or prompts sent to Claude Code from Team, Enterprise or API accounts unless the customer opts in. Consumer plans let users choose, and retention rises to five years when training is allowed. Use a business plan for company code.
Is Claude Code compliant with UK GDPR?
No tool is compliant on its own; your business normally stays the controller for personal data in prompts and code. Use a Team or Enterprise plan for commercial terms and 30-day retention, keep real customer data out of development, protect local transcripts on laptops, and assess whether a data protection impact assessment is needed.
Are Claude Code hooks safe?
Hooks are as safe as the commands in them. Anthropic's documentation warns that command hooks execute shell commands with your full user permissions. Keep hooks in version control, review them like code, avoid hooks from sources you do not trust, and use the allowManagedHooksOnly managed setting if only approved hooks should run.
How long does it take to train a team on Claude Code?
Most teams can be productive after a focused half-day session on their own code, as long as project instructions, skills and permissions are set up beforehand. Plan a review after two weeks to refine CLAUDE.md, add missing skills, adjust permissions and check usage reports against what the team actually does.
Sources
Checked on 16 November 2026. Product features and figures change quickly, so check the current source before you rely on a detail.
- Claude pricing
- Claude Code documentation: Manage costs effectively
- Claude Code documentation: Advanced setup
- Claude Code documentation: Extend Claude Code
- Claude Code documentation: Hooks reference
- Claude Code documentation: Configure permissions
- Claude Code documentation: Data usage
- Anthropic Economic Index: AI's impact on software development (2025)
- NCSC: Guidelines for secure AI system development (2023)
- ICO: Guidance on AI and data protection
Rolling out Claude Code to your team?
Our Claude Code training and setup gets your team productive on your own code, with CLAUDE.md, skills, hooks, subagents and permissions in place from day one. For ongoing build work, see our Claude Code agency service, or Join the Circle for a free business analysis and proposal.





