Qoder Unveils AI Agent Workbench: From Code Completion to Task‑Oriented Automation

Preface

Alibaba pushes Qoder a big step forward: it is no longer just an AI coding tool built around a code editor, but is repackaged as an "agent workbench for everyone".

The shift changes the default AI‑coding scenario from IDE, terminal, code‑completion, and auto‑bug‑fix to describing a goal in natural language, after which the system invokes coding and tool capabilities to accomplish development, prototyping, data processing and other tasks. The entry point is the task itself rather than a code file.

New Position: Coding Core but Not Just for Developers

The core ability remains Coding, but it is placed inside a broader task framework. Users can start by describing a goal such as "create an internal data‑cleaning script" or "add a cache solution to this module"; Qoder then calls models and tools, breaking the goal into executable steps.

This reflects a trend where AI programming moves beyond the "editor plugin" form. Complex work involves understanding requirements, querying repositories, invoking cloud services, reading project‑management systems, handling data, and generating documentation—tasks a simple completion plugin cannot cover.

Qoder aims to pack all these actions into a single workbench.

Auto Scheduling: Models No Longer Manually Chosen

Building on a year of work on Agent Harness, Qoder includes several cutting‑edge models such as Qwen3.8‑Max and provides Auto intelligent scheduling.

Developers often wrestle with choosing a fast but cheap model versus a strong but expensive one. Auto scheduling hands this decision to the system, matching the task’s characteristics to the appropriate model based on effectiveness, speed and cost.

This marks a maturation stage for AI agents: early efforts focused on "which large model is integrated", while later stages focus on model routing, task decomposition, tool invocation and failure recovery.

Connectors, Plugins, and Skills: Agents Need Real Work Systems

Qoder supports over 40 connectors, 70 plugins and more than 20 000 skills, allowing code repositories, project‑management tools, cloud services and internal utilities to be integrated into AI workflows.

A typical development task involves many contexts: code repository state, requirement descriptions in a project‑management system, cloud resource configuration, internal tool permissions and processes, and team conventions about which files or commands may be modified.

Connectors bring systems in, plugins extend specific actions, and skills encapsulate stable procedures; together they form a usable agent environment.

Two Modes: Programming Users vs General Users

Qoder offers a programming mode for developers (code understanding, generation, debugging, refactoring, testing, tool calls) and a general mode for broader business scenarios where users may not write code but can still leverage coding and tool capabilities for data cleaning, prototype generation, workflow automation, or cross‑system information aggregation.

This separation acknowledges that many automation needs come from operators, product managers, data analysts, and project managers who do not know how to code.

Plan, Goal, and Side Tasks: Structure for Long‑Running Tasks

For long‑running or complex tasks, Qoder introduces three structures:

Plan : outlines requirements, solutions and steps before execution, preventing direction drift.

Goal : continuously executes while verifying results, maintaining a clear target, current progress, validation status and next steps.

Side Tasks : handle details or derived issues without interrupting the main task flow.

These mechanisms address the difficulty of keeping AI agents stable over extended, multi‑step processes.

Voice and Desktop Pet: Lighter Interaction

Qoder adds a desktop pet and real‑time voice capabilities, allowing users to start and follow tasks via speech and to view, adjust or take over execution on the fly.

This reduces the need to craft long, structured prompts for every interaction, enabling more conversational, incremental task refinement.

Product Matrix: Covering IDE, CLI, Mobile and Cloud Agents

The Qoder family now includes Qoder IDE, Qoder CLI, Qoder JetBrains plugin, Qoder mobile, digital‑employee product QoderWake and Qoder Cloud Agents, each targeting different entry points—from daily coding to automated scripts, lightweight mobile access, digital‑employee scenarios and continuous cloud execution.

Since its global launch in August 2025, Qoder serves over 6 million users and more than 100 k enterprise customers, indicating that AI coding agents are moving into production‑grade usage.

How Teams Should Evaluate Agent Workbenches

When comparing Qoder, Codex, Claude Code or other agents, look beyond demo output and assess five dimensions:

Context integration: ability to read code repositories, requirement systems, cloud services and internal tools.

Task planning: ability to generate a plan and await human confirmation.

Execution boundaries: control over file reads/writes, command execution, secret access and deployment actions.

Verification loop: automatic testing, failure explanation and re‑verification.

Cost control: model selection per task rather than always using the most expensive model.

Domestic teams should unify model endpoints (e.g., set ANTHROPIC_BASE_URL to https://code.ai80.vip) so that different tools share the same governance, key isolation, logging and cost tracking.

Conclusion

The most notable change is not another AI coding product, but the evolution of AI coding tools into agent workbenches: task‑oriented entry, auto‑model scheduling, a rich ecosystem of connectors/plugins/skills, voice and desktop interaction, and structured long‑term task management via Plan, Goal and side tasks.

For developers, this shifts work from "let AI write code" to designing collaborative workflows where humans set goals, review plans, control permissions and verify outcomes, while AI handles context retrieval, tool invocation and execution.