On August 27, Alibaba officially launched the new Qoder—an agentic workbench built on coding capabilities, designed for everyone. This marks a strategic upgrade from the original Qoder IDE launched in August 2025, shifting its positioning from an “AI coding tool” to an “agent capable of independently completing entire tasks.”
From “Coding Assistance” to “Autonomous Delivery”
The original Qoder IDE, launched in August 2025, was an AI coding tool that helped developers write code, autocomplete, and debug. One year later, it has surpassed 6 million global users and over 100,000 enterprise customers, including China National Petroleum, Nanjing University, China Eastern Airlines, FAW Group, and CITIC Securities.
The new Qoder changes the core workflow: users no longer start from code. They simply describe their goal in natural language, and Qoder automatically understands the context, formulates a plan, invokes tools, executes, and validates — with full visibility and the ability to adjust or take over at any time.
Four Technical Foundations
Harness Engineering Execution Environment: Qoder adopts the industry’s latest Harness technology, upgrading traditional chat into a structured task runtime. The agent’s execution process is structured into auditable artifact pipelines, with clear ownership and status for each step.
Read-Modify-Verify-Iterate Autonomous Loop: Qoder can search codebases, edit across files, run tests, and self-correct based on real results. Plan Mode constrains sessions to analysis and planning, while Permission Mode determines tool invocation permissions — auto-approve, ask for confirmation, or reject.
Granular Permissions and Tool Allowlists: Sensitive operations require explicit authorization, ensuring the agent doesn’t act autonomously beyond its bounds.
Cross-Session Long-Term Memory: Qoder supports periodic self-refinement of skills — the more you use it, the better it understands your projects and habits.
Model Ecosystem and Connector Ecosystem
Qoder comes pre-integrated with multiple frontier models including Qwen3.8-Max, featuring Auto Smart Routing that matches models based on task characteristics, balancing performance, speed, and cost.
On the connector front, Qoder is integrated with 40+ connectors, 70+ plugins, and 20,000+ skills, bringing code repositories, project management tools, cloud services, DingTalk, Notion, and other third-party systems directly into the AI workflow.
Two Modes and Voice Interaction
For different scenarios, Qoder offers two modes: Programming and General. Programming mode is for code-related tasks; General mode enables non-developers to complete tasks using coding and tool capabilities. Qoder also supports voice invocation, proactively delivering task updates, reminders, and next-step suggestions.
The new Qoder desktop app is now available — qoder.com for international users and qoder.cn for China.
Qoder’s evolution from AI coding tool to agentic workbench reflects a key trend in the AI development tools space: coding is the starting point, but completing whole tasks is the real destination. With 6 million users and 100,000 enterprises already using Qoder to write code, getting them to use Qoder to “get work done” is the logical next step.