Where AI coding agents fit in SAP testing

Contoso's test team, led by Priya, carries a heavy regression load since Contoso's SAP S/4HANA go-live in August 2026: Fiori apps in the launchpad, classic transactions such as VA01 and MIGO in SAP GUI, OData services used by the custom core returns app, and about 7,500 custom ABAP objects. Two of Priya's automation engineers have started using Claude Code with Playwright, and their first results look impressive. Grace, the PMO lead, now asks a fair question: can that pair carry Contoso's regression suite through the feature packs and upgrades of the years ahead, or should the TCoE invest in a test platform? This course answers it with evidence rather than enthusiasm.

What Claude Code is, and what it is not

Anthropic describes Claude Code as an agentic coding tool that reads your codebase, edits files, runs commands and integrates with your development tools. It runs in the terminal, in VS Code and JetBrains IDEs, in a desktop app and in the browser. Agentic is the important word: you state a goal, and Claude Code plans, reads files, runs commands, checks the results and tries again, asking for permission according to rules you set. That makes it an excellent assistant to a person who writes or reviews test assets. It is not a test management tool, an execution grid or an evidence repository, and it doesn't claim to be.

• CLAUDE.md: a project instruction file Claude Code reads at the start of every session. Anthropic's docs are explicit that Claude treats it as context, not enforced configuration.
• Permissions: allow, ask and deny rules (evaluated deny first) and permission modes, set in .claude/settings.json and shareable through version control.
• MCP servers: tools connected through the Model Context Protocol, such as Microsoft's Playwright MCP server, which lets the agent drive a real browser.
• Hooks: shell commands that run at fixed points; a PreToolUse hook can block a tool call outright.
• Subagents and headless mode: specialized helpers with their own context and tool limits, and claude -p for scripts and CI pipelines.

The SAP test surface, and what an agent can reach

An SAP program proves quality on several surfaces at once, and an AI coding agent reaches each of them differently. Claude Code works through files, commands and connected tools. Anything that runs in a browser can be driven by Playwright code the agent writes. APIs can be called by scripts. ABAP source can be read and drafted, and SAP now ships an ABAP MCP server that, in SAP's words, lets any agent that supports MCP interact with ABAP code. SAP GUI for Windows is different: it is a desktop application, and the only programmatic route into it is SAP GUI Scripting, which you met in sap-tau-201.

Assistant, author or executor

It helps to separate three roles an agent can play. As an assistant, Claude Code drafts test cases, data plans and defect summaries that a person reads and corrects. The risk is low and the value is high. As an author, it writes automation code that an engineer reviews, commits and owns. The value is real, but someone now maintains that code for years. As an executor, it would run tests and decide pass or fail on its own. That is where SAP sign-off needs things an agent doesn't give by itself: the same steps and verdict on every run, an audit trail, and evidence a process owner can read.