Claude Code is an Anthropic agent that reads files, edits them and runs commands from a plain-language request. For a business, it is included in every Claude Team seat, from $20 per month on annual billing, and on Enterprise at $20 per seat per month plus usage at API rates. These figures are checked on Anthropic's official pages as of 4 November 2026.
Transparency: Codito Ergo Sum is a certified Anthropic partner and uses Claude Code every day. For the comparison of plans and of Claude's privacy in general, see our Claude for business guide. Here, we cover what changes when the AI acts on your files and your systems.
What is the price of Claude Code in business?
Claude Code has no price of its own in the team plans: it is part of the seat, and only API access is paid directly by consumption. Your choice comes down to 3 access modes.
| Access mode | Price (excluding taxes) | How it is consumed | Suited to |
|---|---|---|---|
| Claude Team | Standard: $20 per seat per month on annual billing, $25 monthly. Premium: $100 and $125 | Allowance per seat, renewed in 5-hour windows and weekly, shared with chat and Cowork | Teams of 2 to 150 seats |
| Claude Enterprise | $20 per seat per month on annual billing, 20 seats minimum, plus usage at API rates | Cost follows the model and the task; spending limits per user and per organisation | Organisations with compliance needs |
| API, Bedrock, Google Cloud, Foundry | By volume of tokens, depending on the model | Direct invoice or through your cloud, caps per workspace or per cloud budget | Automations and technical integrations |
Sources: claude.com/pricing, support article on Claude Code with Team or Enterprise, cost documentation. Accessed on 4 November 2026. Prices may change at Anthropic's discretion.
How much should you budget per person?
Anthropic states an average of about $13 per developer per active day, which is $150 to $250 per month, across its enterprise deployments, with under $30 per active day for 90% of users. These figures are averages published by the vendor: they vary widely depending on the model, the size of the code and the use of automation.
Translated for 10 active developers: if usage follows this average at API rates, expect $1,500 to $2,500 per month. On Team, the seat caps this spend, then the usage credit, if enabled, takes over up to the limit you set. A Premium seat at $100 on annual billing can therefore remain more predictable for an intensive developer: the right trade-off is measured over 1 month of pilot.
The documentation also gives the habits that reduce consumption: starting from an empty session between 2 unrelated tasks, keeping the Sonnet model for most tasks and reserving Opus for complex reasoning, writing precise requests. Agent teams, which are hungrier, cost about 7 times more tokens than a standard session in plan mode, according to the same page.
What is Claude Code used for outside developers?
Outside developers, Claude Code is used to build small internal tools, process batches of files and automate repetitive work without waiting for the technical team. Anthropic published its own uses in July 2025: its legal team prototyped a system to direct employees to the right lawyer, and its marketing team built a pipeline that spots underperforming ads in a CSV file and generates new versions of them.
In an SME, we see the same families of uses:
- Management: building a custom management tool. At Gustave Rénovation, a construction company supported by Codito, the company head was trained in Claude Code and is developing their own application (quotes, invoices, planning, subcontractors).
- Finance and administration: reconciling 2 exports, checking files, producing a recurring report from files.
- Marketing and sales: adapting content, cleaning a contact database, preparing a tracking table.
- Operations: renaming and filing batches of documents, extracting data from PDFs.
The terminal is no longer required: Claude Code exists as a desktop app, on the web and in code editors. For office use without code, Cowork is often a simpler way in. The tipping criterion: as soon as the task produces a program, touches real data or has to run without you, bring in someone able to review it.
What are the security risks of Claude Code?
The main risk of Claude Code lies in its rights: it acts with the permissions of its environment, and an action approved too quickly can modify or expose whatever your account can reach. Anthropic's security documentation says it bluntly: the user is responsible for reviewing the proposed code and commands before approving them, and no system is completely immune to attacks.
| Concrete risk | Example | Countermeasure |
|---|---|---|
| Prompt injection | A web page or a file contains a hidden instruction for the agent | Do not give it untrusted content without precautions, isolate in a virtual machine or a container |
| Destructive action | Deleting or overwriting files, a command run on the wrong environment | Git, backups, sandbox, no direct access to production |
| Secret leak | API key written in plain text in a file or a repository | Secrets vault, deny rules on sensitive files, scan before publishing |
| Dubious MCP server | Third-party connector that accesses your tools | Allowlist; Anthropic states that it does not audit the security of MCP servers |
| Fragile code in production | Application that works in a demo and breaks under load | Review, tests, see our vibe coding guide |
The protections exist: a manual mode that starts read-only and asks for approval for each sensitive action, an automatic mode where a classifier blocks what it judges dangerous, a sandbox that isolates the file system and the network, encryption in transit. Runs in Anthropic's cloud take place in isolated virtual machines, with logging. All of this reduces the risk without replacing human review.
What happens to your data with Claude Code?
On commercial plans, Anthropic does not train its generative models on the code and prompts sent to Claude Code, unless the customer has chosen to provide its data, and keeps this data for 30 days as standard. Claude Code runs locally, but all prompts and model outputs travel over the network in TLS.
3 details are worth knowing, all taken from the data usage page:
- The
/feedbackcommand sends a copy of the conversation, code included, kept for 5 years; it can be switched off with an environment variable. - Transcripts are stored in plain text on the computer for 30 days by default: take this into account when protecting machines.
- Zero retention is not part of the standard Enterprise plan: it is activated organisation by organisation, after an eligibility check by your Anthropic contact.
An employee signed in with a personal Pro account falls under consumer rules, with a training setting of their own choosing. So require the business account for any company file.
What governance should you put in place for Claude Code?
Sufficient governance for Claude Code comes down to 7 written rules and a named owner, relying on the administration tools provided by Anthropic.
- Business accounts only, with SSO where possible.
- Managed settings. The managed settings, deployed from the console, a device management tool or a file, override user settings.
- List of permitted commands and MCP servers, maintained by the owner.
- No direct access to production: separate test environment, fictitious data, go-live after review.
- Git and review: every change goes through a branch and a human review, especially for authentication, payments and personal data.
- Caps and monitoring. Spending limits per organisation, group or person, and monitoring through analytics dashboards. The OpenTelemetry export reports costs, tokens and activity; the content of prompts is hidden there by default.
- Documented training: who is trained, on what, on what date. It also contributes to the AI literacy obligation of the AI Act, detailed in our policy template.
In companies of at least 50 employees, introducing a new technology also falls under consultation of the CSE (works council, under French labour law): have your legal adviser validate this point.
When should you not deploy Claude Code?
Do not deploy Claude Code widely if nobody in-house can read what it produces, or if your workstations give access to production without separation. It is also not suited to code subject to location constraints that your contracts forbid entrusting to a third party, unless specifically scoped. If your need is limited to writing and summarising, a seat and the chat interface are enough.
To move forward without exposing yourself, there are 2 routes at Codito Ergo Sum. The Premium AI Partnership, at €799 excl. VAT per month, no commitment, cancel at any time, trains your teams remotely and builds with them agents that they know how to take over, within 4 hours per month spread as you wish. If applications built with Claude Code are already running in production, the Enterprise Partnership, on quote from €2,000 excl. VAT per month, no commitment, provides their monitoring and security. To choose, let's take 30 minutes.
Official sources (checked on 4 November 2026)
- Pricing: claude.com/pricing
- Seats: Use Claude Code with your Team or Enterprise plan
- Costs: Manage costs effectively
- Security: Security
- Data: Data usage
- Administration: Deploy managed settings and Monitoring usage
- Uses: How Anthropic teams use Claude Code
- French Labour Code, article L2312-8


