Development with Claude Code
Feature implementation, tests, migrations, refactoring: Claude Code works in the developers’ repository and terminal. Project conventions are written once, and changes go through review like any other.
Development with Claude Code, technical support, documentation, incident analysis: agents that speed up your product teams, inside a security framework your customers can audit.
An AI agent for a software company is a Claude agent built into the software delivery cycle: writing and reviewing code with Claude Code, answering technical support tickets, writing documentation, analysing incidents. It works inside the team’s tools, and every code change goes through human review before merge.
A software vendor answers to its customers for its code, whoever wrote it. Four frameworks govern the use of agents in the product and in the delivery pipeline.
The Cyber Resilience Act (Regulation (EU) 2024/2847) sets cybersecurity requirements for products with digital elements: installed or embedded software, and the remote processing they cannot work without. Standalone SaaS is generally outside its scope. Reporting of actively exploited vulnerabilities and severe incidents has applied since 11 September 2026; most of the text applies from 11 December 2027. Code written with an agent is still code the vendor answers for.
A vendor that builds Claude into its own product may become a provider or deployer of an AI system under the EU AI Act. Obligations depend on the end use: transparency for a conversational assistant, a stricter regime if the product is used for hiring, credit assessment or grading students.
Your customers ask for evidence: ISO/IEC 27001, SOC 2, security questionnaires. The use of an AI model in development or in the product has to appear there, with the sub-processors involved and the data shared.
If your product processes personal data on behalf of your customers, calling an AI model adds a sub-processor. It belongs in your data processing agreement and in the list shared with customers, who must be able to object.
Feature implementation, tests, migrations, refactoring: Claude Code works in the developers’ repository and terminal. Project conventions are written once, and changes go through review like any other.
The agent reads the ticket, the logs and the documentation, reproduces the issue when it can, proposes an answer or a fix and escalates to the engineer with the context already gathered.
API documentation, user guides, release notes drawn from tickets and commits: the agent keeps documentation in step with releases, the product team validates.
During an incident, the agent correlates logs, metrics and recent changes and suggests hypotheses; afterwards, it drafts the post-mortem. The on-call engineer keeps control of production actions.
No generated code reaches the main branch without review. Tests, static analysis and secret scanning run on the agent’s changes exactly as on the team’s.
Allowed commands, off-limits files, network access: Claude Code is configured per project. Deny rules, backed by the sandbox, keep secrets out of reach, and sensitive actions require confirmation.
Ticket-to-production time, review rework rate, change-related incidents: existing delivery metrics are enough to judge the real effect, without counting lines of code.
We start from your processes and your regulatory constraints to find the first agent worth building, and the framework that goes with it.