Alibaba launches a new Qoder, an AI agent workbench built for everyone

On August 27, 2026, Alibaba introduced a new version of Qoder, upgrading what began as a developer-focused AI coding tool into an agent-driven workbench built for everyone, not just professional engineers. The redesigned interface is simpler and task-centered: users no longer need to start from code — describing a goal in natural language is enough for Qoder to call on its coding and tool capabilities to carry out development, prototyping or data-processing work.

From coding tool to a workbench for everyone

The new Qoder builds on a year of work on its Agent Harness, the execution framework that orchestrates its agents, and ships with several frontier models built in, including Qwen3.8-Max. A feature called Auto handles intelligent scheduling: depending on the nature of a task, it automatically matches the right model, balancing quality, speed and cost without the user having to choose manually. To ground agents in real project context, Qoder now supports more than 40 connectors, over 70 plugins and more than 20,000 skills, letting it plug code repositories, project-management tools, cloud services and internal company systems directly into the AI workflow.

Depending on the scenario, Qoder offers two modes: a programming mode built for developers and code-related tasks, and a general mode aimed at a wider audience that doesn't need to touch code directly to get work done through Qoder's coding and tool capabilities. It's this general mode that embodies the "for everyone" pivot Alibaba is emphasizing with this release.

Plan, Goal and side tasks: managing long-running work

For long and complex jobs, Qoder adds three complementary mechanisms. Plan lays out requirements, an approach and concrete steps before execution begins, so users know what to expect before the agent starts working. Goal keeps execution anchored to a clear objective while continuously verifying results along the way, to keep a task from drifting off course. Side tasks absorb secondary details or issues that come up mid-mission without interrupting the main thread of work.

• 40+ connectors covering code repositories, project management, cloud services and internal tools
• 70+ plugins and more than 20,000 skills agents can call on
• Several frontier models built in, including Qwen3.8-Max, orchestrated by the Auto feature
• A programming mode for developers and a general mode for everyone else
• Plan, Goal and side tasks to structure long, complex missions

A desktop pet and live voice to lower the barrier to entry

On the interaction side, the new Qoder introduces a desktop pet and real-time voice capabilities: users can start and follow up on tasks by voice, and check in, adjust or take over execution at any time. The tool also adds a scratchpad, voice-recorded meeting notes and live voice discussions. Alibaba illustrates the idea with a concrete example: a user asks out loud to "first analyze the refactoring plan for this module"; Qoder produces a plan based on the project's context; the user confirms the direction and adds a requirement such as "add caching," still by voice; the agent then continues execution and verification. Throughout, users can adjust instructions on the fly in conversation rather than having to restate a full instruction at every step — a design Alibaba frames as key to lowering the barrier for non-specialist users.

The Qoder product family has grown to include this new general-purpose Qoder, Qoder IDE, Qoder CLI, a Qoder plugin for JetBrains, a mobile version, the "digital employee" product QoderWake, and Qoder Cloud Agents for enterprises. Since its global launch in August 2025, Qoder says it has served more than 6 million users and over 100,000 enterprise customers worldwide — numbers Alibaba is citing to justify this broader pivot.

Against Cursor, Claude Code and Devin, Alibaba's answer in agentic coding

Over the past year, Western tools such as Cursor, GitHub Copilot, Anthropic's Claude Code and Cognition's Devin have turned "agentic coding" into one of the most contested segments of applied AI — agents capable of independently executing multi-step development tasks rather than simply completing a line of code. Qoder follows a similar playbook, with an Agent Harness, autonomous execution and continuous verification, but Alibaba is now explicitly betting on accessibility for a much broader audience, using natural language and voice to extend capabilities once reserved for engineers to non-technical roles. That shifts Qoder's competitive footprint beyond the developer-tools market and into the broader field of enterprise productivity software.

For a technical team evaluating the tool, the immediate stakes cut both ways. On one hand, a genuinely capable general mode could mean real productivity gains if product managers, operations or marketing staff can prototype or process data themselves instead of waiting on an available engineer, shortening internal turnaround times. On the other, it forces a faster rethink of access controls and code review: once non-technical users can trigger an agent able to modify a repository or query internal systems, deciding who approves, oversees and secures those actions becomes a governance project in its own right — not just a tooling decision.