Claude Code explained: what is it, what does it cost and what can your SME do with it?
Claude Code is Anthropic's AI agent that builds software in your own files. What it costs in Belgium, where it runs and what an SME can do with it.
Summary
- Claude Code is an AI agent that builds software: it reads your files, changes them and runs commands, in the terminal, the desktop app, the browser or your code editor
- It is included in every paid Claude plan: in Belgium from €15 per month with annual billing (Pro, excluding VAT, as of 17/11/2026); the free plan gives no access
- It also works without a terminal: the Code tab in the Claude desktop app and claude.ai/code work with ordinary instructions in plain language
- According to the 2026 Stack Overflow survey, 65.5% of developers who answered the question used Claude Code in the past year
- For an SME without a programmer, it works well for small internal tools and prototypes; as soon as customer data, payments or a complex codebase are involved, someone with technical knowledge should be part of it
Claude Code is Anthropic’s AI agent that builds software in your own files. You describe in plain language what you want, and Claude Code reads the project, changes files, runs commands and asks for your approval where needed. It is included in every paid Claude plan, in Belgium from €15 per month.
Most explanations of Claude Code are written for developers. This article is for the Flemish SME without its own programmer: what it is, what it costs in Belgium, where it runs, what you can build with it yourself and where the limit lies. ClickForest builds its own tools and this website with Claude Code, and that practice comes back throughout this article.
What is Claude Code?
Claude Code is an AI agent for software work. Anthropic describes it as a tool that “reads your codebase, edits files, runs commands, and integrates with your development tools”. The difference with a chatbot: Claude Code works directly in the files of a project, across multiple files, and carries out the steps itself.
Anthropic launched Claude Code on 24 February 2025 as a research preview, together with Claude 3.7 Sonnet. It has been generally available since 22 May 2025. In November 2026 it runs on Opus 5.5 by default, an Anthropic model released on 22 September 2026. If you prefer, you can choose another model, such as Sonnet 5.5.
Researcher Andrej Karpathy, co-founder of OpenAI, saw it at the end of 2025 as the first convincing example of an AI agent:
“Claude Code (CC) emerged as the first convincing demonstration of what an LLM Agent looks like - something that in a loopy way strings together tool use and reasoning for extended problem solving.”
— Andrej Karpathy, AI researcher, in 2025 LLM Year in Review (19 December 2025)
An AI agent gets a goal, plans the steps, uses tools and checks the result. With Claude Code that is about software, although people also use it for other work on the computer.
What is the difference between Claude Code, Claude Desktop, Cowork and ChatGPT?
Claude Code builds and changes software in the files of a project. Claude Desktop is the app in which you chat, share documents and use Cowork, and that app includes Claude Code as a separate Code tab. Cowork lets Claude carry out office work on your computer. ChatGPT is a chat assistant from OpenAI. For software work in a project, Claude Code is the specialised tool.
| Tool | What it does | For whom |
|---|---|---|
| Claude Code | Builds and changes software in the files of a project, runs commands, works with Git | Anyone who wants a website, tool or script built or adapted |
| Claude Desktop | The Claude app for chat, documents and projects, with a Code tab for Claude Code | Everyone with a Claude plan |
| Claude Cowork | Carries out office tasks on your computer, such as organising files or making reports | Anyone who wants to hand off repetitive work with files |
| ChatGPT | Chat assistant from OpenAI for text, analysis and questions | General use, just like the chat in Claude |
The names make it confusing, and beginners are not the only ones who think so. Developer Simon Willison, co-creator of the Django web framework, wrote:
“Claude Code is, with hindsight, poorly named.”
— Simon Willison, independent developer, on his blog (16 October 2025)
His point: Claude Code also automates computer work other than programming. How the desktop app and Cowork work exactly, you can read in our explanations of Claude Desktop and Claude Cowork. Want to know how Claude compares to other assistants? Then look at our comparison of AI tools.
What does Claude Code cost in Belgium?
Claude Code is included in every paid Claude plan. From Belgium, claude.com shows these prices on 17 November 2026, all excluding VAT: Pro €15 per month with annual billing or €18 per month, Max €90 and Ultra €180 per month. The free plan gives no access. For teams, a standard seat costs €18 per month with annual billing.
| Plan | Price in Belgium (excluding VAT) | Usage |
|---|---|---|
| Free | €0 | No Claude Code |
| Pro | €15 per month with annual billing (€180 upfront), or €18 per month | Basic, with a limit per five hours and per week |
| Max | €90 per month | Five times the usage of Pro |
| Ultra | €180 per month | Ten times the usage of Pro per week, without a five-hour limit for verified accounts |
| Team, standard seat | €18 per seat per month with annual billing, or €21.04 per month | For teams of 2 to 150 people |
| Team, premium seat | €90 per seat per month with annual billing, or €105.23 per month | More usage per person |
Source: claude.com/pricing, consulted from Belgium on 17 November 2026. The page shows the visitor’s currency: anyone opening it from the United States sees dollars. Enterprise is also listed in dollars in Belgium.
Note: the limits apply to Claude and Claude Code together. If you chat a lot during the day, you have less room to build in the evening. And if you have set an Anthropic API key on your computer, according to the help center you pay per use instead of through your plan. For heavy use, that can be a deliberate choice: you then pay per token with API credit.
Where do you use Claude Code: in the terminal, the desktop app or the browser?
Claude Code runs in five places: in the terminal, as an extension in VS Code, in the Code tab of the Claude desktop app for Mac and Windows, in the browser at claude.ai/code and as a plugin in JetBrains editors. For non-developers, the desktop app and the browser are the easiest start: you work with the same agent without a terminal.
- Terminal: the original form, on macOS, Linux and Windows. Powerful, but anyone who has never used a command line quickly drops out here.
- Desktop app: the Code tab in the Claude app. According to Anthropic, Claude Code is built into the app itself, so you do not install the terminal version separately. You need a paid plan.
- Browser: at claude.ai/code, also in the Claude app for iOS and Android. Handy for starting a task and reviewing it later.
- VS Code: a panel next to your files that shows the original and the proposed code side by side before you approve.
- Integrations: Claude Code also works with GitHub, GitLab and Slack, and can fetch data from other systems through MCP.
Simon Willison already saw that barrier when Anthropic launched Cowork in January 2026:
“What it really needs is a UI that doesn’t involve the terminal and a name that doesn’t scare away non-developers.”
— Simon Willison, independent developer, on his blog (12 January 2026)
That interface now exists. We ourselves build alternately in the terminal and in the VS Code extension. For beginners, the Code tab in the desktop app and the VS Code extension are the gentlest way in.
What can an SME without a programmer build with Claude Code?
An SME without a programmer can build small internal tools and prototypes with Claude Code: a calculator for quotes, a script that cleans up export files, a simple dashboard or a change to an existing website. The more customer data, payments or integrations are involved, the more important technical knowledge becomes to judge what you approve.
Ethan Mollick, professor at Wharton, puts his finger on the point that matters for an SME:
“Claude Code is particularly powerful because it works on your computer and your files.”
— Ethan Mollick, professor at the Wharton School, on his blog One Useful Thing (7 January 2026)
Working on your own files is the strength, and at the same time the reason to be careful. Mollick had already written about vibe coding, building by only describing what you want:
“vibecoding is most useful when you actually have some knowledge and don’t have to rely on the AI alone.”
— Ethan Mollick, professor at the Wharton School, on his blog One Useful Thing (11 March 2025)
From our own practice: ClickForest builds custom tools, internal software and websites with Claude Code for SMEs in Flanders. Our own browser and editor extension Claude Token Tracker was built entirely with it, as was the rebuilt order reporting for Bastiano. This website also runs on a setup that we build and maintain with Claude Code, as we explain in our article on Astro as an alternative to WordPress. We worked out the difference between building on gut feeling and building with structure in vibe coding versus Claude Code.
In Belgium there is still room. According to Eurostat, 28.8% of small Belgian companies (10 to 49 employees) used at least one AI technology in 2025, compared with 76.4% of large ones. That gap is the second largest in the European Union.
Would you rather have a tool built? ClickForest builds founder-led custom software with Claude Code for SMEs in Mechelen, Antwerp and the rest of Flanders. See how we build with AI or get in touch.
How safe is Claude Code, and what may it do without permission?
Claude Code asks for permission according to the mode you choose. In recent versions it starts in auto mode by default in the terminal and VS Code: many actions run automatically, with safety checks in the background. In manual mode it only reads and asks for permission at each step. Anthropic itself says auto mode offers no guarantee of safety: you remain responsible for what you approve.
| Mode | What Claude Code may do without asking | When to use |
|---|---|---|
| Manual | Read only, every change asks for permission | When you are just starting or working with sensitive files |
| Plan mode | Investigate and propose a plan, change nothing | Before a larger change |
| Auto | Everything, with safety checks in the background | For tasks whose direction you trust |
| bypassPermissions | Everything | According to the documentation, only in isolated containers and virtual machines |
The documentation is clear about this: “Auto mode reduces permission prompts but does not guarantee safety.” A command you approve can moreover write anywhere your user account has access to. If you want extra certainty, turn on the sandbox with the /sandbox command.
Anthropic defends auto mode with its own research: among 1,053 paying testers, human review caught 13.6% of the dangerous commands, auto mode 89%. That is research by the vendor itself, so read it as an indication. A safe habit: start a new task in plan mode, and do not let Claude Code loose on customer data without a copy and a backup. More on how to capture such working methods in our explanation of Claude Skills.
What do independent figures say about Claude Code?
The independent figures are about developers. In the 2026 Stack Overflow survey, 65.5% of the 12,255 developers who answered the question used Claude Code in the past year. Trust remains conditional: 48% trust AI when the result is easy to check, 7% for important decisions. A productivity gain for SMEs has not been measured.
That question of trust is justified. In early 2025, research organisation METR had sixteen experienced open-source developers work with and without AI assistance, and they were 19% slower with AI. A follow-up study from August 2025 gave a different outcome, but METR itself calls that data unreliable due to selection effects and writes that developers probably do work faster with AI now.
A second lesson comes from Anthropic’s own research, published on 27 October 2026. Fifty analysts carried out a statistical analysis with Claude Code and followed the agent’s preference in 86% of the choices, where you would expect about 50% by chance. Anthropic adds that it does not know how much that says about the real world. Still, it is a useful warning: if you approve quickly, you also adopt the AI’s choices.
The creator of Claude Code also remains cautious. About customers such as NASA, with very large codebases, he said:
“They have really big, really complicated codebases, and for them it’s not solved yet; the model still makes mistakes, and its code isn’t always perfect.”
— Boris Cherny, creator of Claude Code at Anthropic, in an interview with Platformer (26 May 2026)
How do you get started with Claude Code as an SME?
Start small and concrete. Choose a paid Claude plan, open the Code tab in the desktop app and give Claude Code one well-defined task in a separate folder, such as a tool that cleans up an export file. Let it make a plan first, read it and only then approve changes. Work with Git from day one, so you can undo every step.
- Choose a plan. Pro is enough to try it out. If you often hit the limit, Max is the next step.
- Start in the desktop app or the browser. That way you do not need a terminal. For the isolated sessions in the desktop app, Anthropic does require Git.
- Give one concrete instruction. Describe what the tool should do, for whom, and how you can tell it works. What you know about good instructions works here too; read more about it in our article on writing better prompts.
- Use plan mode. Let Claude Code first propose what it is going to do.
- Record fixed agreements. A
CLAUDE.mdfile in your project contains the instructions Claude Code reads at every session, such as which folder must not be touched. - Check the result. Test the tool with real, non-sensitive data before you continue working with it.
If you want your whole team to learn to work this way, a workshop with their own tasks helps more than a one-off demo. What such training should deliver, you can read in our guide to an AI workshop for your team.
When is it better to have it built than to do it yourself?
Have it built as soon as mistakes really cost money or trust: with customer data, payments, integrations with your accounting or webshop, or software that has to last for years. Building it yourself pays off for small tools that are only used internally. A middle way often works best: have the foundation built and then adapt it yourself with Claude Code.
In September 2026, Boris Cherny predicted that everyone will one day write code as well as he does. That is a distant horizon. Today the question is rather who judges the choices Claude Code makes: which data is stored where, what happens when an integration fails, how you maintain the software. That is where a builder’s work lies, and it does not disappear because the code gets written faster.
ClickForest builds custom software with Claude Code for SMEs in Flanders. Frederiek Pascal directs Claude Code himself as an AI agent, without intermediate layers. After the final payment, the code we wrote for you is yours, as an export on request, so you can later continue yourself with Claude Code or with another party. Looking for an agent that carries out tasks independently in your marketing or administration? Then also look at our approach to AI agents and our article on AI agents in marketing.
What should you remember about Claude Code?
Claude Code is Anthropic’s AI agent that builds software in your own files. It is included in every paid Claude plan, in Belgium from €15 per month excluding VAT, and also works without a terminal through the desktop app or the browser. For small internal tools, an SME can start on its own. With customer data, payments and long-lived software, you need someone who can judge the choices.
ClickForest builds tools and websites with Claude Code for SMEs in Flanders, and helps teams that want to learn it themselves through tailored AI training. Curious what Claude Code can mean for your business? See our approach to building with AI or get in touch.
Grow with AI, practically and without unnecessary complexity
Ready to use AI to save time and accelerate your marketing? Discover our AI services
Discuss your challenge directly with Frederiek: Book a free strategy call or send us a message
Prefer email? Send your question to frederiek@clickforest.com or call +32 473 84 66 27
Strategy without action remains theory. Let's take your next step together.
Frequently asked questions
Claude Code is Anthropic's AI agent for software work. It reads the files of a project, changes them, runs commands and works with tools such as Git. You give instructions in plain language, in the terminal, the Code tab of the Claude desktop app, the browser or a code editor such as VS Code.
Claude Code is included in every paid Claude plan. In Belgium, claude.com shows on 17/11/2026: Pro €15 per month with annual billing or €18 per month, Max €90 and Ultra €180 per month, all excluding VAT. Team costs €18 per seat per month with annual billing. You can also pay per use through the API.
No. The free Claude plan gives no access to Claude Code; the pricing page literally states under Free that Claude Code is not included. You need at least a Pro plan, or you pay per use with API credit in a Console account.
That depends on the task. For software work, Claude Code is popular: in the 2026 Stack Overflow survey, 65.5% of developers who answered the question used Claude Code in the past year.
Claude Desktop is Claude's app for chat, documents and Cowork. Claude Code is the agent that builds software in the files of a project. The desktop app includes Claude Code as a separate Code tab, so you can use it without opening the terminal.
For a small internal tool or a prototype, not necessarily: you describe what you want and approve the changes. The more is at stake, the more knowledge matters, because you remain responsible for what you approve. ClickForest builds custom tools with Claude Code for SMEs in Flanders and teaches teams to work with AI.
Yes. ClickForest builds custom tools, internal software and websites with Claude Code for SMEs in Flanders, founder-led. After the final payment, the code we wrote for you is yours: you get it as an export whenever you ask, so you can later continue yourself with Claude Code or with another party.
Sources and references
Claude Code: documentation and pricing:
- Anthropic: Claude Code documentation, overview · https://code.claude.com/docs/en/overview
- Anthropic: Claude Code product page · https://claude.com/product/claude-code
- Anthropic: Claude pricing (consulted from Belgium on 17/11/2026) · https://claude.com/pricing
- Anthropic Help Center: using Claude Code with your Pro, Max or Ultra plan · https://support.claude.com/en/articles/11145838-use-claude-code-with-your-pro-max-or-ultra-plan
- Anthropic: Claude Code documentation, model configuration · https://code.claude.com/docs/en/model-config
- Anthropic: Claude Code documentation, permission modes · https://code.claude.com/docs/en/permission-modes
- Anthropic: Claude Code documentation, security · https://code.claude.com/docs/en/security
- Anthropic: Claude Code documentation, VS Code extension · https://code.claude.com/docs/en/vs-code
- Anthropic: Claude Code documentation, desktop app · https://code.claude.com/docs/en/desktop
- Anthropic: auto mode as the default in Claude Code (August 2026) · https://claude.com/resources/articles/auto-mode-default-in-claude-code
- Anthropic: announcement of Claude 3.7 Sonnet and Claude Code (February 2025) · https://www.anthropic.com/news/claude-3-7-sonnet
- Anthropic: announcement of Claude 4 and the general availability of Claude Code (May 2025) · https://www.anthropic.com/news/claude-4
Research and figures:
- Stack Overflow: Developer Survey 2026, AI section (October 2026) · https://survey.stackoverflow.co/2026/ai/data.md
- Stack Overflow Blog: results of the 2026 Developer Survey (October 2026) · https://stackoverflow.blog/2026/10/06/the-results-of-the-2026-developer-survey-are-here
- METR: update on the productivity study with AI assistance (February 2026) · https://metr.org/blog/2026-02-24-uplift-update/
- Anthropic: "How AI agents shape experts' judgment calls" (October 2026) · https://www.anthropic.com/research/agents-judgment-calls
- METR: original study with experienced open-source developers (July 2025) · https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/
- Eurostat: "The use of AI technologies in the EU, Key results, 2026 edition" · https://ec.europa.eu/eurostat/documents/7870049/23260410/KS-01-26-009-EN-N.pdf/37d063cb-28cf-3b4e-91f3-c3784c970842
Sources of quotes:
- Andrej Karpathy: "2025 LLM Year in Review" (December 2025) · https://karpathy.bearblog.dev/year-in-review-2025/
- Simon Willison: blog post on Claude Skills (October 2025) · https://simonwillison.net/2025/Oct/16/claude-skills/
- Simon Willison: blog post on Claude Cowork (January 2026) · https://simonwillison.net/2026/Jan/12/claude-cowork/
- Ethan Mollick: One Useful Thing on vibe coding (March 2025) · https://www.oneusefulthing.org/p/speaking-things-into-existence
- Ethan Mollick: One Useful Thing on Claude Code (January 2026) · https://www.oneusefulthing.org/p/claude-code-and-what-comes-next
- Platformer: interview with Boris Cherny (May 2026) · https://www.platformer.news/boris-cherny-interview-ai-jobs/
- Computer History Museum: conversation with Boris Cherny (September 2026) · https://computerhistory.org/blog/in-conversation-with-boris-cherny-creator-of-claude-code/
