Skip to content
←Back to Applications

Doubao Work: ByteDance's productivity agent, from decks to computer control

doubao-work

ByteDance's August 2026 productivity agent: PPT, documents and sheets, deep research, scheduled tasks — plus reading and writing local files, filling forms across pages, and phone-based task dispatch.

C
CONFIDENCE
Vendor Claim
Official model card or keynote only, no independent re-test
Docs, decks, sheets + computer control
KEY METRIC
What it delivers (vendor claim)
Vendor Claim · 2026-08
MATURITY
Product
research → demo → product → production
Editor's take

Doubao Work earns an office ladder slot because it pushes the boundary of an office agent down to the operating-system layer: reading and writing local files, filling forms across pages, and dispatching tasks remotely from a phone. Producing a deck or a report is something most assistants can do now; changing files in a real desktop environment and completing multi-step web flows is the line between "doing the work for you" and "writing for you" — the former has to handle failures, retries and dirty data, which is a different order of difficulty.

The second reason is that it joins both ends of content production inside one workbench: text (PPT / documents / spreadsheets / research) and imagery and video (Seedream 5.0 Pro / Seedance 2.5) without shuttling between tools. For e-commerce ads and explainer shorts, the step removed is precisely the most tedious hand-off.

Confidence: C (vendor claim). Checkable: the August 2026 launch, the individual/team/enterprise positioning, the four capability groups and skill examples, and the named image and video models (official site). Not checkable: the success rate of local-file and cross-page operations, and output reliability in professional domains such as medicine and investing — there is no third-party evaluation, so the card states what it can deliver and no accuracy figure. The professional examples are illustrations of use, not conclusions.

AI AgentOfficeProductivityByteDance
ByteDanceOfficial site3 min read7

Positioning in one line: a workbench from "idea" to "delivered"

Doubao Work's pitch is short: "a new work habit — you say it, Doubao does it." Officially it is described as an AI product and brand for productivity scenarios: it breaks a goal into tasks, calls tools, and completes research, content creation and data work — aimed at individuals, teams and enterprises. It launched in August 2026, with 30 days of subscription included on download.

Four capability groups — and "operating the computer" is the hard one

  • Office tools — generate professional PPT, documents and spreadsheets as usable deliverables;
  • Deep research — analyse complex questions, gather and cross-check sources, produce a research report;
  • Scheduled tasks — run analyses, reports and summaries automatically at preset times;
  • Operating the computer — read, create and modify local files; open web pages, fill in information and complete cross-page actions; and dispatch tasks remotely from your phone while away from the machine.

The first three are things plenty of products do today. The fourth is what separates this from a writing assistant. Reading and writing local files, and filling forms across pages, means operating in a real environment — dirty data, failures, retries. That is an order of magnitude harder than generating text, and it is the line between "an agent that can actually do the work" and one that cannot.

Creative delivery lives in the same workbench

It connects both ends of content production: Seedream 5.0 Pro for images, posters and comics, and Seedance 2.5 for storyboards and finished video, with iteration on feedback. A deliverable that needs images and video no longer has to be carried between three tools — for e-commerce ads, explainer shorts and film work, that removed shuttling is exactly the annoying part.

Skill coverage

The vendor stresses that it invokes domain-specific skills and follows professional workflows, and lists wide-ranging examples: e-commerce customer-service scripts, travel planning, cross-platform headlines and editorial calendars, de-AI-flavoured copywriting, product creative, stock screening, and clinical diagnosis with evidence. The breadth (investment / medical / creative / commerce) shows a skill-plugin route rather than one big general prompt.

Boundaries

This is a consumer and team product with broad vendor-claimed coverage. The actual success rate of local-file and cross-page operations, and the reliability of its output in professional domains such as medicine and investing, have no third-party evaluation behind them. The professional examples are usage illustrations, not citable conclusions — anything touching medical or investment judgement must go back to qualified sources.

More in Agents

4
MathResearchTopB

AlphaEvolve: pointing a model at problems that come with an objective scoring function

Google DeepMind's coding-agent-driven evolutionary search: Gemini Flash proposes candidates in volume and Gemini Pro in quality, automatic evaluators score them, and high scorers stay in the population as next-round context, i.e. evolutionary search with an LLM as the mutation operator; it applies only where an automatic evaluator exists, hence the math and agents filing. An evolved scheduling heuristic has run in production in Google data centres (Borg) for over a year, recovering 0.7% of Google's global compute (Google's compute, not all compute on earth), the card reading and the only result validated by long-running production. An evolved matmul kernel is 23% faster at specific sizes, and about 20% of 50+ open maths problems improved, including 4x4 complex matrix multiplication in 48 multiplications against Strassen's 1969 record of 49. Verification differs from the prover line: a Lean check is mathematical correctness, "23% faster" is an empirical reading on specific hardware. No public weights, Early Access is a waitlist, and the precondition is an evaluator you write yourself. Graded B (contested): a year of production behind the Borg result, nothing reproduced.

0.7%Google global compute recovered (production-verified)Contested · 2025-05
ProductionGoogle DeepMindSite
AlphaEvolve: pointing a model at problems that come with an objective scoring function
MultimodalDocs & SlidesTopC

Gamma: the orchestration tool that turns a body of source material into a presentable deck

What Gamma does is not "prompt for one slide" but reorganise existing material into something presentable: it ingests long documents, PDFs, pasted text or a URL, and returns a full deck with outline, pagination, imagery and layout, in a card-based structure where each page can be reordered, rewritten or re-illustrated on its own. It represents the kind of AI that is genuinely useful in the documents vertical - the value is in orchestration and editorial selection, not in generation quality. Its ceiling is equally clear: the output lives in Gamma's own layout system, and exporting to pptx loses a layer of fidelity - which happens to be a hard requirement for most organisations.

Web / pptx / pdfOutput formatsVendor Claim · 2025-01
ProductGamma TechSite
Gamma: the orchestration tool that turns a body of source material into a presentable deck
AgentsDocs & SlidesC

WorkBuddy: Tencent's all-scenario AI office workbench

Tencent's office agent: state the requirement and it plans, calls tools and generates files, leaving both process and result for review — connected to Tencent's IM, docs, mail, meetings and knowledge base.

15 minDeep-research turnaround (vendor claim)Vendor Claim · 2026-10
ProductTencentSite
WorkBuddy: Tencent's all-scenario AI office workbench

As an Amazon Associate, we earn from qualifying purchases.