Claude Code for Business
Updated: 23 October 2026
Claude Code is Anthropic's AI agent that reads, changes and tests program code inside a project. It also handles documents of any kind well, from a technical specification to a commercial proposal. It has done well enough to bring Anthropic into the top three AI companies, and Anthropic has released a similar app for people who don't write code, Claude Cowork.
For a company weighing Claude Code for business use, three facts decide whether it pays off. A Team seat costs $20–125 per person a month, but it buys an allowance, not unlimited use, and Anthropic's own average for developers across enterprise deployments is $150–250 a month. The plan your staff sign in with decides whether Anthropic may train on your code: personal Pro and Max accounts allow it when a setting is on, while Team, Enterprise and API accounts do not. And every company success figure we could find is published by Anthropic, so the number that counts is the one your own pilot measures.
Contents
- What Claude Code is, in plain terms
- Claude Code pricing: what it costs a team
- Claude Code data terms, security and admin controls
- What companies report about Claude Code, and what independent studies say
- How to roll out Claude Code in a team
- When Claude Code is not the right tool
- What changed in Claude Code this month
- Official Claude Code sources
- FAQ
What Claude Code is, in plain terms
A chat assistant answers a question and leaves the work to you. Claude Code is an agent: a program that takes a task, plans the steps and carries them out itself. It reads the files in your project, edits them across several places, runs commands such as tests, and shows the result for a developer to accept. Anthropic describes it as "an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools."
It can run in four places: the terminal (the text window developers type commands into), extensions for the VS Code and JetBrains code editors, a desktop app, and the browser, where long tasks can run in the background. Inside a task it can hand parts of the work to subagents: helper agents with their own instructions and context. The same engine is also available as the Agent SDK, a toolkit for building your own agents with Claude Code's tools.
What companies use it for:
- The development team builds features, fixes bugs and writes tests with an agent doing the typing, while developers set the task and review the result.
- People outside engineering build internal tools and working prototypes they would otherwise queue for.
- Business agents for inbound support or internal operations run on the same engine through the Agent SDK. Examples follow in what companies report.
Claude Code vs Cursor vs GitHub Copilot
All three now work as agents, so "agent or autocomplete" no longer separates them. The differences are where each one lives, which models it uses and how its plans charge. Read from each vendor's own documentation on 23 October 2026:
| Claude Code (Anthropic) | Cursor | GitHub Copilot | |
|---|---|---|---|
| Agent in the code editor | VS Code and JetBrains extensions, desktop app | Cursor is itself an editor; its Agent edits code and runs commands | Copilot in VS Code and JetBrains |
| Agent in the terminal | the main, full-featured version | Cursor CLI | Copilot CLI (if Copilot comes through an organization, the organization must enable it) |
| Agents running in the cloud, in parallel | Claude Code on the web | Cloud agents in isolated virtual machines; they open pull requests (proposed changes for review) | Copilot cloud agent, started from GitHub; works on issues and pull requests |
| Home ground | Claude Code: terminal first | Cursor: editor first | GitHub Copilot: GitHub first |
| Models | Claude Code: Anthropic's Claude models | Cursor: several vendors (OpenAI, Anthropic, Google and others) plus Cursor's own | GitHub Copilot: several vendors (Anthropic, OpenAI, Google, xAI, Microsoft); the list depends on the plan |
| How plans charge | Claude Code: Claude plan seats with a shared usage allowance, or pay per token (per unit of text) through the API or a cloud provider | Cursor: a weekly usage pool covers Cursor's own models; other vendors' models are billed per token at API list price | GitHub Copilot: a monthly allowance of AI credits, used up by tokens; admins can set a credit limit |
There is no "better" column. A team that lives in GitHub issues and pull requests may prefer Copilot. A team that wants to choose between model vendors may prefer Cursor or Copilot. A team that works in the terminal and wants Anthropic's models gets Claude Code at its most complete. All three charge the same way at heart: the plan buys an allowance, and heavy agent work is paid by usage on top. The pilot described in the rollout section is how you find out which fits your team.
Claude Code vs Claude Cowork
Claude Cowork is Anthropic's agent for work outside code. Anthropic says it uses "the same agentic architecture that powers Claude Code", without the terminal that most people outside engineering find unfamiliar. Claude Code works inside a code project; Cowork works in the folders and tools you choose and hands back documents, spreadsheets, presentations and research for review. Since 16 September 2026 Anthropic has made Cowork part of one Claude app and is rolling this out across all plans gradually.
| Claude Code | Claude Cowork | |
|---|---|---|
| Who it is for | Claude Code: developers and the people who look after the code | Claude Cowork: staff outside engineering, such as marketing, finance, sales, legal and operations |
| Where it runs | Claude Code: terminal, VS Code and JetBrains extensions, desktop app, browser | Claude Cowork: the Claude desktop app (macOS, Windows; Linux in beta), web and mobile |
| What it works on | Claude Code: the code project, its files, commands and tests | Claude Cowork: folders and tools you choose, connected services such as Microsoft 365 or Google Drive, and a built-in browser for web tasks |
| What you get | Claude Code: code changes, tests and pull requests for review | Claude Cowork: Excel spreadsheets with working formulas, PowerPoint presentations, documents, research and scheduled reports |
| Plans | Claude Code: every paid plan (Pro, Max, Team, Enterprise) | Claude Cowork: every paid plan (Pro, Max, Team, Enterprise) |
On Team and Enterprise plans each member's Claude Code and Cowork work draw from that member's one seat allowance, so a developer who also hands Cowork long tasks leaves less of it for coding that week.
Claude Code pricing: what it costs a team
The short answer: Claude Code has no price of its own. It is included in every paid Claude plan, and the seat price is only the entry fee. Heavy agent work runs past a seat's allowance, and then you either wait or pay for usage on top.
| Plan (claude.com/pricing, read 23 October 2026) | Price per person a month | Who it is for |
|---|---|---|
| Free | $0, Claude Code not included | trying the chat assistant |
| Pro | $17 a month billed annually ($200 a year up front), $20 monthly | one person |
| Max | from $100 (5× or 20× Pro's allowance) | one person who works with Claude all day |
| Team, Standard seat | $20 annually, $25 monthly | teams of 2 to 150 |
| Team, Premium seat | $100 annually, $125 monthly; 5× a Standard seat's allowance | heavy users inside a Team plan |
| Enterprise | $20 a seat, billed annually, 20-seat minimum, plus usage at API rates | large companies |
Prices exclude tax and are the vendor's list prices in US dollars.
Why a Claude Code seat price is not the cost
On Team and Enterprise plans, each member's Claude Code usage "draws from a per-seat allowance that resets on a rolling five-hour window and a weekly window", and that allowance is shared with the rest of your Claude use, chat included. When a developer uses it up, there are two paths:
- Wait for the window to reset, which means lost hours on a deadline.
- Continue on usage credits, paid on top, with spend limits the company sets for the whole organization, a group or one member.
A third path is the API: no seat at all, and the bill is the tokens (the units of text the model reads and writes) the developer's sessions use. Anthropic suggests this route for heavy coding sessions. Companies with a cloud contract can also run Claude Code through Amazon Bedrock, Google's Gemini Enterprise Agent Platform or Microsoft Foundry: usage is then billed per token through that cloud, under commercial terms.
Anthropic's Claude Code costs documentation gives the reference point.
"Across enterprise deployments, the average cost is around $13 per developer per active day and $150-250 per developer per month, with costs remaining below $30 per active day for 90% of users."
Anthropic does not say how its $150–250 per developer average for Claude Code splits between seats and usage. Its advice is to "start with a small pilot group" and measure before a wider rollout.
Two things push the cost up. A coding session carries whole files and many steps of reasoning, so it uses far more tokens than a chat. And agent teams, several Claude Code instances working on one task, "use approximately 7x more tokens than standard sessions when teammates run in plan mode". Plan mode is a setting where the agent writes a plan for approval before changing anything. Agent teams are an experimental feature, switched off by default.
Enterprise is priced as "Seat price + usage at API rates", so usage is part of the bill from the first month. The costs documentation also describes a per-seat allowance on Enterprise, so ask Anthropic's sales team how the two combine for your contract.
A worked example: Claude Code for ten developers
Your inputs decide the answer, so start with them:
| Input | This example | Yours |
|---|---|---|
| Developers | 10 | |
| Seat type | Team Standard or Premium | |
| Billing | annual or monthly | |
| Share of heavy agent users | an assumption: how many of the ten hand the agent long, multi-step tasks most of the day | |
| Spend-limit policy | who may continue past the allowance, and up to what sum |
The options:
| Option | Per month |
|---|---|
| (a) 10 × Team Standard | $200 billed annually, $250 monthly |
| (b) 10 × Team Premium | $1,000 billed annually, $1,250 monthly |
| (c) Reference point, not a plan: Anthropic's enterprise average, 10 × $150–250 | $1,500–2,500 |
With 10 Team Standard seats (a), heavy users will reach the allowance, and each time they do they either wait for the window or start spending credits under the limit you set. 10 Team Premium seats (b) give five times the allowance for five times the price. The Anthropic-average reference point (c), $1,500–2,500, shows where real usage across enterprise deployments lands; treat it as a reference point. Enterprise needs at least 20 seats, so it is not an option for a team of ten.
The choice depends on how much of your team's day goes to long agent tasks, which is exactly what a pilot measures.
Claude Code data terms, security and admin controls
The short answer: the plan decides the data terms. Company code and documents on personal Pro or Max accounts fall under consumer terms, where Anthropic may train on them when the user's setting allows. Team, Enterprise and API accounts fall under commercial terms, where it does not.
| Personal accounts: Pro, Max | Commercial accounts: Team, Enterprise, API, cloud platforms | |
|---|---|---|
| Training on your code and prompts | yes, "when this setting is on"; the pricing page lists it as opt-out | no, unless the company opts in to share data |
| Retention | 5 years if the user allows training, 30 days if not | 30 days as standard; zero retention for qualified Enterprise accounts, enabled by Anthropic per organization |
| Terms | Consumer Terms | Commercial Terms |
In short: on personal Pro and Max accounts with training allowed, Anthropic keeps Claude Code data for 5 years; on Team, Enterprise and API accounts the standard retention is 30 days.
The practical trap is ordinary: developers start on their own Pro subscriptions, and the company's code ends up under consumer terms. Moving the team to a Team plan straight away is the fix.
If your developers are a contractor's team, ask which plan their seats are on. Personal Pro or Max accounts mean consumer terms for your code. Where else your code may go, and what each route means for it, is covered on our vibe coding page.
What the admin gets:
- Team: single sign-on (SSO, where staff log in with their company account), central billing and administration, and spend limits for the organization, a group or one member.
- Also on the Claude Code Team plan: server-managed settings, meaning tool permissions, file-access restrictions and connections to outside tools set once for the whole organization.
- Enterprise adds: role-based permissions, SCIM (automatic creation and removal of accounts from your staff directory), audit logs, a compliance API (a feed of activity records for your security tools), custom data retention, network access controls and IP allowlisting. Anthropic lists SOC 2 Type II compliance for its enterprise governance.
- Claude Code analytics on Team and Enterprise plans: lines of code accepted, how often developers accept the agent's edits, daily active users and sessions, and, with a GitHub connection, pull requests shipped with Claude Code's help. Anthropic calls those contribution figures "deliberately conservative" and "an underestimate". Read them both ways: lines accepted do not track code deleted later, contribution figures are in beta and cover repositories on github.com only, and they are switched off for organizations with zero data retention.
Claude Code Team vs Enterprise
| Team | Enterprise | |
|---|---|---|
| Seats | 2 to 150 | 20 minimum |
| Billing | per seat, monthly or annual; usage credits past the allowance | seat price plus usage at API rates, annual only |
| Admin | SSO, central billing, spend limits, server-managed settings | adds SCIM, role-based permissions, audit logs, compliance API |
| Data retention | 30 days | custom retention; zero retention for qualified accounts |
| Fits | most companies with 3–30 developers | regulated or large companies that need audit and provisioning |
What companies report about Claude Code, and what independent studies say
Every company figure below is published by Anthropic about its customers; none is independently audited.
Engineering teams: Claude Code results
- Rakuten (Japanese technology company): Anthropic reports that time to market for new features fell from 24 working days to 5, "a 79% reduction" (vendor-published). The figure measures delivery time for new features.
- Ramp (finance software): "1+ million lines of AI-suggested code implemented in just 30 days" (vendor-published). That is a volume. It says nothing about speed or quality.
Outside engineering: Claude Code results
- Pictet (Swiss private bank): around 700 people across engineering, product and business teams got access to Claude Code and Claude Cowork in a staged rollout, and generated "working prototypes in two hours instead of two weeks" (vendor-published). A prototype still has to be built into a product before customers can use it. If your interest is AI for the non-technical side of the company, that is a broader question than one tool, and it is where AI consulting starts.
Business agents built on the Claude Agent SDK
- JAKALA (data and AI services, Milan): Claude runs across 3,000+ employee seats and inside campaign-optimization agents built on the Agent SDK, where "a human pod handles review, approval, and execution". Anthropic reports the optimization cycle fell from 5–7 working days to under 24 hours on accounts using these agents (vendor-published).
Agents like these are billed differently from Claude Code seats: an Agent SDK agent runs on API keys and a per-token bill. How a business agent is put together, and what it costs to run, is in how to build an AI agent for your business.
If you want an agent for your own process, whatever it runs on, the AI agents we build also keep a person approving the actions that matter. We build agents for inbound and internal processes; we do not build cold-outreach, outbound-calling or voice agents.
Independent studies of AI coding speed
Two independent sources temper the vendor's numbers:
- METR, a randomized trial in 2025: experienced open-source developers took 19% longer to finish tasks when AI was allowed, yet believed afterwards that AI had made them 20% faster. The tools in that trial were mainly Cursor with Claude models, not Claude Code. METR's early-2026 follow-up says it is now "likely that developers are more sped up from AI tools", but calls its own data "only very weak evidence for the size of this increase".
- DORA 2025 (Google's survey of nearly 5,000 technology professionals): AI adoption goes with higher delivery throughput and, at the same time, lower delivery stability. Its summary: "AI doesn't fix a team; it amplifies what's already there."
For an owner, the conclusion is simple: feeling faster is not the same as being faster, and nobody else's percentage tells you what your team will get. Measure your own.
How to roll out Claude Code in a team
The short answer: measure before you start, pilot with a small group, keep code review human, and write the rules down. The tool is the easy part.
1. Take a baseline before a Claude Code pilot
Three numbers, from the month before anyone gets a seat:
- how long a change takes from first commit to merge (pull request cycle time);
- how often a change breaks something or has to be redone (change-failure rate, rework);
- how long a new developer takes to make a first useful change.
Leave the accept rate (how often developers accept the agent's edits) out of the baseline: it exists only once the tool is in use.
2. Guard against the two risks the METR and DORA studies name
Perceived speed is not measured speed (METR), and stability falls when review and tests are weak (DORA). So code review stays with people, tests run automatically on every change before the tool goes to the whole team, and the team keeps a written rules file the agent reads at the start of every session.
3. Learn from a public Claude Code pilot: Cian
Cian, a Russian property marketplace, described its enterprise pilot of Claude Code on Habr in June 2026, naming eight problems. Three of them are decisions for management:
- Ready-made subagents from plugin stores do their own job well but don't know your team's exceptions, so their quality in your project is unknown until tested.
- "Give it the whole Jira epic" (Jira is a task tracker; the idea is to hand the agent a large batch of tasks and get the result back) works only at first glance; Cian's advice is not to do it yet.
- Auto-approving the agent's actions saves interruptions, but Cian calls it one of the most serious risks of agent-based development. Close it with technical limits, so safety doesn't depend on one developer's attention at one moment.
For the wider pattern of moving any agent from a pilot to daily work, see how to implement AI agents in a company.
If you'd rather talk it through first, book a free consultation.
Set up AI-assisted development in your team
Our vibe coding service builds AI into each stage of your team’s work, with a person signing off each stage, and measures a pilot on one real feature against your own baseline. Sizing specification $1,500, credited against the pilot; pilots from $6,000. Tool subscriptions are paid to the vendor.
When Claude Code is not the right tool
The short answer: when the team has nothing for the agent to amplify, or when the job is too small for an agent.
- No tests and no version-control discipline. DORA's finding cuts both ways: the agent amplifies what is there, including the mess.
- A one-off script or a quick question. A chat assistant answers it for less.
- No one to own the rules and the review. Without an owner, the rules file goes stale and review becomes a rubber stamp.
- A budget that cannot absorb spikes. Usage varies with the work; if a month at double the average would be a problem, set hard spend limits before the pilot starts.
What changed in Claude Code this month
From the official changelog, 23 September to 22 October 2026. Only changes a business reader acts on; developer-level fixes are left out.
| Date | Change | What it means for a team |
|---|---|---|
| 19 Oct | Pro and Max warn when resuming an old conversation would use a noticeable share of the five-hour allowance window | users see the cost to their allowance before they spend it |
| 19 Oct | Code Review gets Quick, Standard and Deep levels per organization or repository, and an option to review every repository in a GitHub organization | automated review depth becomes an admin decision |
| 14 Oct | Claude in Slack can answer in channels shared with other organizations; off by default, an owner or admin turns it on | working with partners in Slack now needs a written rule on what Claude may see |
| 29 Sep | Project settings can no longer widen or switch off a sandbox (a fenced-off space for the agent's commands) that an admin requires | admin rules now beat local settings in one more place |
Official Claude Code sources
All read on 23 October 2026:
- Claude Code overview
- Claude pricing
- Manage costs effectively
- Data usage
- Track team usage with analytics
- Claude Code for Enterprise
- Claude Code changelog
- METR, early-2025 developer study and its February 2026 update
- Customer stories on claude.com: Rakuten, Ramp, Pictet, JAKALA
- Cian's pilot on Habr, June 2026 (in Russian)
- Google Cloud, the 2025 DORA report
FAQ
Can a company use Claude Code commercially?
A company can use Claude Code commercially: Team and Enterprise plans and API accounts fall under Anthropic's Commercial Terms. Personal Pro and Max plans fall under the Consumer Terms, so a company should put staff on Team or Enterprise rather than personal accounts.
How much does Claude Code cost for a business?
A Team seat costs $20–25 (Standard) or $100–125 (Premium) per person a month, and Enterprise is $20 a seat plus usage at API rates, with a 20-seat minimum. Anthropic's own average across enterprise deployments is $150–250 per developer a month, and it does not say how much of that is seats and how much is usage.
Is there a free plan?
The Free Claude plan does not include Claude Code: it is included in every paid plan, starting with Pro.
Does Anthropic train on our code?
Anthropic does not train on code from Team, Enterprise or API accounts, unless the company opts in. On personal Pro and Max accounts Anthropic trains on the data when the user's setting is on.
Team or Enterprise?
Team covers 2 to 150 seats with SSO, central billing and server-managed settings. Enterprise needs at least 20 seats, bills usage at API rates and adds SCIM, role-based permissions, audit logs, a compliance API and custom data retention.
Is it only for developers?
Claude Code is not only for developers. Non-developers use it to build internal tools and prototypes, and the same engine powers business agents through the Agent SDK. Anthropic reports that around 700 people at Pictet, including business teams, have access to Claude Code and Claude Cowork.
Is Claude Code worth the cost?
Only your own pilot can answer that. Every company success figure we could find is published by Anthropic, and an independent trial found developers felt faster while measuring slower. Take a baseline, pilot with a small group and compare.
Why is Max so expensive?
Max buys one person five or twenty times Pro's allowance. For a company, compare it with a Team Premium seat, which gives five times a Standard seat's allowance and comes under commercial terms.
Weighing Claude Code for your team?
In a free consultation we’ll look at your team, your plans and your data terms, and outline a pilot you can measure.
