Engineer automates 80% with Claude + .NET

A senior engineer said he now automates about 80% of his day job with Claude wired into a.NET app and GitLab workflows. (x.com) The setup described in the July 2026 post pulls work items from GitLab, sends them through Claude for classification, and uses the results to drive pull request and issue-handling steps inside the team’s existing tooling. GitLab’s own documentation says its Issues and Merge Requests application programming interfaces can create, update, approve, merge, or block work based on external systems. (x.com) (docs.gitlab.com 1) (docs.gitlab.com 2) In plain terms, the software engineer built a software clerk: GitLab provides the tickets and code review objects, Claude makes a structured judgment about what each item is, and the.NET program turns that judgment into actions. Anthropic’s documentation says Claude can be embedded in C# applications through its official software development kit for.NET. (docs.gitlab.com) (platform.claude.com) That matters because issue triage and pull request routing are repetitive parts of software work that already live in systems with programmable interfaces. GitLab documents label management, issue metadata, and merge-request automation as supported API use cases, which makes this kind of workflow less a lab demo than a wrapper around existing engineering plumbing. (docs.gitlab.com 1) (docs.gitlab.com 2) (docs.gitlab.com 3) Anthropic has also been pushing developers toward agent-style workflows instead of one-off chatbot prompts. Its current developer guides highlight tool use, structured JavaScript Object Notation output, and agentic systems that can plan and execute tasks against external tools. (anthropic.com) (platform.claude.com) The technical hinge is reliability. Anthropic says structured outputs can force Claude responses into a fixed JSON schema, which is the difference between a model saying “this looks like a bug” in prose and a program safely receiving a field that says `type=bug`. (platform.claude.com) The claim is still a self-reported one, and the original post does not publish benchmark data showing exactly how the 80% figure was measured. What is documented is that the building blocks exist: GitLab exposes the work objects, and Anthropic now ships an official C# package for.NET applications. (x.com) (platform.claude.com) (docs.gitlab.com) The thread lands on a narrower point than “artificial intelligence replaces engineers.” It shows one engineer moving routine ticket sorting and pull request handling into software, then keeping the higher-value work that remains. (x.com)