Skip to content

    Agentic coding: Claude Code as an AI coding agent

    |7 min read
    Topographic map graphic with a dotted path winding up a long slope to a terracotta goal, a symbol of agentic AI work with Claude Code

    A coding assistant suggests the next line and leaves the rest to you. Agentic coding with Claude Code hands the whole ticket to an AI coding agent, which reads the codebase, changes the files that need changing, runs the tests and writes the commit. That difference decides how much of a developer's day actually moves, which is why engineering leads across Europe are asking about it now.

    The guide below covers what agentic coding means, what an AI coding agent does inside a repository and how a team steers it with one file and a few settings. Teams that want help moving agents from a trial to production can see the fixed prices for a Claude consultant.

    What is agentic coding with Claude Code?

    Agentic coding means handing a goal to an AI coding agent, a program that plans the change, acts on the files, checks the result and tries again without anyone typing each instruction. A chat window answers a question about code, whereas an agent opens the files, makes the change and runs the check that tells it whether the change worked.

    Claude Code is Anthropic's agent for software work, and it runs in the terminal, in VS Code and JetBrains, in the Claude desktop app and in a browser.1 You describe the task in plain language, for example to write tests for the auth module and fix any failures. Claude Code then plans the approach, edits across several files and verifies the result.

    Adoption among developers moved faster than almost anyone predicted. The Pragmatic Engineer surveyed 906 engineers in early 2026, and 46 per cent named Claude Code their most loved tool, against 19 per cent for Cursor and 9 per cent for GitHub Copilot.2 OpenAI's own coding agent is compared with Claude Code task by task, with prices in euros, in Claude vs Codex.

    Gartner expects 40 per cent of enterprise applications to include task-specific AI agents by the end of 2026, up from under 5 per cent in 2025.3 The same firm also predicts that over 40 per cent of agentic AI projects will be cancelled by the end of 2027, because of rising costs, unclear business value or weak risk controls. That second number is the more useful one for anyone planning a budget.4

    What does an AI coding agent do in a real codebase?

    The work falls into four jobs that most teams recognise from their own backlog, and each one still needs a person who reads the diff before it merges.

    The module nobody dares to touch

    Every company has a module written years ago by someone who has since left, with no documentation and a test coverage figure nobody says out loud. Claude Code reads the module and its callers, writes tests that pin the current behaviour, and then refactors step by step with the tests running after each change. A human still decides whether the new structure is the one the team wants to maintain, since the agent has no view of where the product is heading.

    Tests, lint and dependency updates

    Writing tests for untested code, fixing lint errors across a project, resolving merge conflicts and updating dependencies are the chores Anthropic lists first in its documentation.1 Those tasks are dull enough that nobody schedules them, which is exactly why handing them to an agent for an afternoon clears weeks of backlog.

    Commits and pull requests

    Claude Code works directly with git, meaning it stages changes, writes commit messages, creates branches and opens pull requests.1 In a CI pipeline, meaning the automated build that runs on every push, it can review code and triage issues through GitHub Actions or GitLab. The review still belongs to a person, because an agent that approves its own pull request has removed the one check that catches its mistakes.

    Connections to your other systems through MCP

    MCP, the Model Context Protocol, is an open standard for connecting AI tools to other systems, so Claude Code can read design documents in Google Drive, update tickets in Jira or pull data from Slack.15 Each connection is a small server you add and approve, and it reaches only what you give it access to. Teams that want a connection into their own systems built for them can see on managed services how we quote each one as a fixed price after a scoping call.

    How do you steer an agent that edits your code?

    A file called CLAUDE.md in the project root is the steering wheel, meaning a plain text file with coding standards, architecture decisions and review rules that Claude Code reads at the start of every session.6 Writing ten lines in that file, for example which test command to run and which folders never to touch, changes the agent's behaviour more than any prompt typed during the session.

    Skills package repeatable work for the whole team, such as a /review-pr command that follows your own checklist. Hooks run a shell command before or after an action, for example a formatter after every file edit.1 Permission settings decide which commands Claude Code may run without asking, and a new team should start strict, since loosening a rule takes one line while undoing a bad deploy takes an evening.

    Claude Code prices and GDPR settings

    Money and data are the two questions every engineering lead asks before the first seat, and both have their own sections in our Claude Code guide with 2026 prices. Company code belongs on a commercial account such as a Team seat, because Anthropic does not train models on code sent under commercial terms and keeps that data for 30 days as standard.7 Every paid Claude plan from Pro at €18 a month includes Claude Code, so the seat a developer already has usually settles the budget question.8

    Where to start this week

    Agentic coding pays off on dull, well-tested work first and on risky, poorly tested work last, so the first week should go to chores and not to the payment module. Count the tickets the team closes per week for four weeks before the agent arrives and four weeks after, because that one number answers whether the seats pay back. Wondering who buys and runs the seats? Claude run as a managed service for the whole team puts them on one euro invoice.

    Open your issue tracker, pick the three oldest tickets labelled as chores or tech debt, and write a CLAUDE.md of ten lines naming your test command and the folders Claude Code may not touch. In thirty minutes you have a file and a short list, which is everything the first session needs.


    Sources

    Footnotes

    1. Anthropic (2026). Claude Code overview. Claude Code Docs. https://code.claude.com/docs/en/overview ↩ ↩2 ↩3 ↩4 ↩5

    2. The Pragmatic Engineer (2026). AI tooling for software engineers in 2026. https://newsletter.pragmaticengineer.com/p/ai-tooling-2026 ↩

    3. Gartner (2025). Gartner predicts 40% of enterprise apps will feature task-specific AI agents by 2026. https://www.gartner.com/en/newsroom/press-releases/2025-08-26-gartner-predicts-40-percent-of-enterprise-apps-will-feature-task-specific-ai-agents-by-2026-up-from-less-than-5-percent-in-2025 ↩

    4. Gartner (2025). Gartner predicts over 40% of agentic AI projects will be canceled by end of 2027. https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027 ↩

    5. Model Context Protocol (2026). What is the Model Context Protocol (MCP)? https://modelcontextprotocol.io/docs/2026-07-28/getting-started/intro ↩

    6. Anthropic (2026). Store instructions and memories. Claude Code Docs. https://code.claude.com/docs/en/memory ↩

    7. Anthropic (2026). Data usage. Claude Code Docs. https://code.claude.com/docs/en/data-usage ↩

    8. Anthropic (2026). Plans and Pricing. https://claude.com/pricing ↩

    Frequently asked questions

    Joachim Sahlin, CEO & AI Advisor

    About the author

    Joachim SahlinCEO & AI Advisor

    Joachim helps European companies understand and use AI in everyday work. As founder of satori., he trains teams, coaches leadership groups, and shows how tools like Claude become a natural part of work.

    Cookies

    Analytics cookies show which pages are read. Marketing stays off unless you tick it under Details.