Agentation

Kilo Code is an open-source AI coding agent for VS Code, JetBrains, and CLI that generates code from natural language, automates complex tasks, and runs terminal commands. It supports 500+ AI models at provider rates and features an Orchestrator mode that breaks complex tasks into coordinated subtasks across planner, coder, and reviewer agents. With 1.5M+ users and 25T+ tokens processed, it's the #1 coding agent on OpenRouter.

GStack is an open-source Claude Code workflow system created by Y Combinator CEO Garry Tan, designed to encode specialized AI roles for planning, code review, QA, and shipping. Built with Bun and Playwright, it uses route-aware QA and structured release workflows to connect code changes with actual application behavior. Enables developers to achieve 10,000+ lines of quality code per week.

Cursor Composer 2 is an advanced agentic coding model built into the Cursor IDE, released March 19, 2026. With a 200,000-token context window, it autonomously plans and executes complex, multi-file coding tasks across an entire codebase — handling file edits, terminal operations, and tool use at frontier-level performance. It is used by over 50,000 businesses including Stripe.

Warp is an AI-native agentic development environment built from the ground up for developers. It features natural-language command generation, multi-agent orchestration via Oz (launched February 2026), and IDE-style terminal editing — letting developers describe tasks in plain English and have AI agents build, debug, and ship code autonomously. Over 700K developers use Warp with Pair and Dispatch agent modes for local and cloud-based agentic workflows.

z.AI's flagship open-weight model released February 2026 with 744B total parameters (40B active). Designed for complex systems engineering and long-horizon agentic tasks. Integrates DeepSeek Sparse Attention for efficient long-context workloads. Strong in reasoning and code generation.

Qwen 3 Coder is Alibaba's open-source AI coding model optimized for agentic software engineering. It supports a native 256K token context window (extendable to 1M tokens), enabling it to process vast codebases in a single session. Qwen 3 Coder scored 44.3% on SWE-Bench Pro and powers the Qwen Code CLI agent for terminal-based development.