Developer Tools

The tooling layer of AI-assisted development is the fastest-changing surface in software engineering right now. This cluster covers the editors, CLI agents, coding benchmarks, language shifts, and workflow changes that are actually altering how software gets written—not the aspirational roadmaps, but the tools developers are paying for and running in production.

The IDE wars are real. Cursor crossed $300M ARR in early 2026 after reaching developer velocity that incumbent tools struggled to match; Windsurf and GitHub Copilot occupy different parts of the market based on latency, context quality, and workflow integration. Groundy ran these tools against a production 50k-line codebase rather than synthetic exercises, which produces different conclusions than most published comparisons.

Vibe coding crossed the one-year mark. Andrej Karpathy’s framing—letting the AI drive while you guide high-level intent—has real productivity wins for prototyping and non-developer users, but the security debt and architectural decay accumulate invisibly. The OSSRA data showing a 107% surge in open-source vulnerabilities tied to AI-generated code is the invoice arriving for early adoption.

Claude Code, GitHub Copilot’s agentic mode, and OpenHands represent a second wave: not autocomplete, but agents that navigate codebases, write tests, open PRs, and run CI pipelines. The mechanics of running Claude Code as a GitHub Actions agent—trust boundaries, token scope, sandbox isolation—are exactly the operational details that rarely make it into vendor documentation.

Groundy also tracks the language and framework shifts underneath the tooling: Rust’s quiet expansion into inference engines and data pipelines, the return-to-Rails sentiment among developers burned by React complexity, and the Tree-sitter code indexing approaches that underpin better AI code understanding.

Benchmark literacy matters here. SWE-bench Verified has become the de facto coding-agent evaluation, but its 500-issue scope, test isolation assumptions, and gap between isolated bug fixes and full codebase navigation produce numbers that routinely flatter vendors more than they inform practitioners. Groundy reads the methodology papers and applies the same skepticism to IDE leaderboards that any engineer should apply to any benchmark.