On October 9, 2026, TRAE announced a major product change. TraeWork and TraeCode will merge into a single product called TRAE, upgraded to a full-chain development platform. The stated goal is to serve professional developers and, at the same time, people who want to create with AI without a deep engineering background. This article distills the announcement: what the merge means, how the two modes divide labor, and why the company frames this as a move from AI coding to AI development.

1. One Entry, Two Modes, Three Endpoints in Sync

The core of the merge is the end of mode-splitting. Previously, TraeWork leaned toward office work, task management, and content generation, while TraeCode focused on professional coding. The split matched the products but not the work.

A user's real task does not naturally divide into "office" and "coding." Building a product demo requires requirement breakdown, page design, code implementation, test checks, and a delivery document. Advancing a project can touch documents, data, tasks, code, and reporting at the same time. Two separate entries forced users to guess which tool a task belonged to. That guesswork is exactly the friction the merge removes.

The new TRAE uses one unified entry for these continuous tasks. It supports two working modes that switch seamlessly.

1.1 Agent mode

Agent mode is built for moving from an idea to a delivered result. A beginner can step through a demo one stage at a time. A professional can dispatch multiple agents in parallel across projects and coordinate their progress. The mode abstracts away the sequencing; the user states the goal, and the platform handles the steps. The handoff between agents is part of the workflow, not a manual chore.

1.2 IDE mode

IDE mode keeps the full intelligent development environment. It is the place for deep code editing and fine-grained debugging within a single project. When an agent hands over a working but imperfect result, the developer moves into IDE mode to refine it precisely. Fine-tuning, interaction fixes, and structural changes all belong here.

The split is deliberate. Agent mode owns global advancement; IDE mode owns deep development. For a beginner, the pair behaves like an all-round assistant that starts from an idea. For a professional, it behaves like a workbench that schedules agents across projects and tasks.

1.3 Three endpoints in sync

TRAE now covers desktop, web, and mobile. A user can start a task on a computer, check progress on a phone, add requirements, and return to the desktop for deeper work. Projects, tasks, and artifacts continue around the same workflow. Device switching does not interrupt the pipeline. The same task chain connects capabilities, so AI participates from idea to result rather than stopping at a single stage.

The three-endpoint model is not cosmetic. State lives in the workflow, not on the device. That is what lets a phone view continue where a desktop session left off, and what keeps a task recoverable after a machine or a meeting interrupts it.

2. The Name Still Means The Real AI Engineer

TRAE began as "The Real AI Engineer." The intent was never to ship another AI coding tool. The target was an AI engineer that understands goals, advances tasks, and solves problems. After the merge, users do not need to pick an entry point. They do not need to decide whether to start from code, a task, or a design. They state what they want to complete, and the platform starts.

That positioning explains the product decisions. A coding tool optimizes output. An engineer optimizes outcome. The merge, the unified entry, and the artifact model all point in the same direction: treat a request as a project to be finished, not a prompt to be answered.

3. Not Just Writing Code, but Getting the Work Done

The announcement states a principle: the value of AI coding is not generating code, it is solving problems inside a real project. The new TRAE strengthens six capabilities. They are project understanding, code generation, file modification, command execution, error analysis, and result verification.

Underlying them is what the announcement calls the Harness engineering layer. Stronger models provide the reasoning; the harness provides reliable execution. Together they let the AI do more than answer. They let it keep advancing a task until completion. Generation quality is a model problem; finishing a task is an engineering problem, and the harness is the answer to the latter.

There is also a shift in interaction. AI coding is moving from "the human gives an instruction, the AI does one step" to "the human sets a goal, the AI advances continuously." The new TRAE improves goal understanding, tool calling, result checking, and autonomous repair. That reduces how often a user must step in. Professionals finish complex tasks with less friction. More people can turn a single sentence into an openable, clickable, verifiable result.

The phrase "verifiable result" is worth stressing. A generated answer is cheap; a result that passes a check is what actually advances a project. This is why result verification sits alongside generation as one of the six core capabilities rather than an afterthought.

4. Upgrade Details and What Migrates

The upgrade is a merge of two codebases into one product. The new TRAE covers the full chain: requirement breakdown, solution generation, code development, and artifact delivery.

For TraeWork users, the in-client prompt shows an "upgrade now" button. Installing the new TRAE and opening it once completes data migration. For TraeCode users, the upgrade triggers automatically inside the client.

The migration preserves chat history, account data, profile, login state, local data, and business credits. No manual relocation is needed. The announcement asks users to keep the network stable and not close the upgrade window during the process.

The release is in grayscale testing. Upgrades reach users in batches. Issues can be reported to feedback@mail.trae.cn.

5. Agent Plus IDE: From Global Advancement to Deep Development

The dual-mode design covers the full range from task advancement to deep coding. In Agent mode, the workbench manages multiple projects at once. Several agents run in parallel on different tasks. Local tasks can launch across multiple directories, and SSH remote environments are supported, so work continues in more flexible contexts. When a task needs close attention to code detail, the user switches to IDE mode.

Parallel execution multiplies the platform's outgoing calls. Every agent request for a model or a tool travels the same infrastructure, so keys, limits, and accounting need a single consistent place. An API gateway such as Treerouter is where those rules live: one set of credentials, one rate-limit policy, and one metering ledger that every agent shares. Centralizing them keeps the dispatch layer simple as the number of parallel agents grows.

A test scenario in the announcement makes the dual-mode logic concrete: an event registration website.

The old flow required many steps. Define registration rules. Draw the page structure. Write the frontend. Handle form logic. Add tests. Write a document. Each step was a separate session in a separate tool.

The new flow starts in Agent mode. Break the requirements, produce a page plan, write the code, run tests, and assemble the delivery document. Complex tasks split across parallel agents: one structures the pages, one generates code, one checks the form logic, one organizes the delivery document. When the page needs fine-tuning, the interaction needs optimization, or the code needs a deeper change, the user returns to IDE mode.

The scenario shows the division of labor clearly. Agents handle breadth. IDE handles depth. The platform decides the handoff.

6. A Full-Chain Platform: Four Capabilities Under One Roof

After the merge, four capabilities enter the unified workflow. Together with Agent mode and IDE mode, they form a complete chain from task initiation to result consolidation.

6.1 Automation tasks

Repeated processes are handed to agents. Periodically organizing project progress, running fixed checks, summarizing test results, and generating reports can all run continuously without a human at each step. Automation turns standing work into a scheduled pipeline.

6.2 Template library

Common scenarios start fast. Product solutions, page building, data analysis, and technical documents can begin from a template instead of a blank page. The template is a starting point, not a constraint; the workflow adapts it to the task. Faster starts shorten the distance from idea to first working version.

6.3 My assistant

Delivery work outside code is brought into the chain. Documentation, reports, collaboration, and organizing tasks connect to the same task pipeline as coding. The assistant closes the gap between "the code is done" and "the project is delivered." Non-engineering work becomes part of the same workflow instead of a separate step.

6.4 Unified project artifacts

Code changes, HTML, PPT, Word documents, images, videos, and reports all land in one place called Artifacts. Results no longer hide in chat history, and files no longer scatter across tasks. The model turns each AI collaboration into a reviewable, presentable, and continuable asset. It is not just convenient file retrieval. It is a persistence layer for the whole workflow, and it is what makes multi-session work reproducible.

7. Why the Merge Is a Product Thesis, Not a Feature List

The announcement describes the change in terms of a broader direction: from AI coding to AI development. The new TRAE reconnects development links that had been split across tools. For professional developers, it means more efficient work on complex projects. For a wider group of creators, it means moving from an idea to a usable product, step by step.

One line in the announcement carries the positioning: AI is changing how people create, but what matters is the people willing to solve problems and make ideas real. The shift, in the company's words, is from "letting AI write code" to "building the product together with AI." The closing slogan states the aim plainly: let every idea have the chance to move forward.

8. A Technical Reading of the Unification

Viewed as an engineering decision, the merge is about reducing moving parts. A platform that manages multiple projects, dispatches parallel agents, and syncs across three endpoints generates a large volume of internal calls. Each agent request for a model, a tool, or an external service passes through shared infrastructure. That infrastructure must handle identity, limits, and accounting consistently, not ad hoc.

The same principle applies to how the platform's own calls are managed. When several agents run concurrently, teams typically front the underlying model and tool calls with an API gateway. An API gateway such as Treerouter centralizes keys, rate limits, and usage metering across the whole dispatch, so parallel agents share one control point instead of each configuring access separately. The gateway becomes the single place where policy meets traffic.

This is the engineering mirror of the product thesis. The product unifies the user's entry point; the gateway unifies the platform's outgoing calls. Both reduce fragmentation. Both make the system easier to operate as it scales.

9. What to Watch After the Upgrade

The announcement leaves three open threads worth tracking.

First, grayscale rollout. The upgrade reaches users in batches. How quickly coverage expands, and how the feedback loop at feedback@mail.trae.cn translates into fixes, will determine early quality.

Second, migration reliability. Chat history, account data, and business credits must survive a two-codebase merge. Any loss in that transfer is high-impact, so the batch rollout is the right cautious path.

Third, the asset model. Whether Artifacts becomes a genuine persistent workspace, rather than a shared folder with a new name, depends on how deeply it integrates with agents and templates. That integration decides whether the product thesis holds.

The direction is clear even before the details settle. Splitting office tools from coding tools no longer matches how people actually build things. A unified platform that treats an idea as a continuous workflow is the more durable model, and TRAE is betting the merge on it.

Learn more: https://treerouter.com