Doubao Work Deep Dive: 5 Advanced AI Workflows for PPT, Data & App Building

1. Data-First PPT Creation Workflow

Guest Simon 阿文 demonstrated a counter-intuitive approach: for data-driven presentations, don't start with slides. Instead, invoke Doubao Work's data analysis skill to examine the dataset and uncover hidden viewpoints. Select a specific insight, then generate a full PPT where every table and chart supports that viewpoint. The output includes BI-level visualizations such as rectangle tree maps with hover drill-down detail, and each slide comes with pre-written speaker notes.

The author emphasizes that editable output is a critical capability for AI office tools — "uneditable AI output is no different from disposable chopsticks." Doubao Work provides two editing layers: (1) direct sidebar editing of text, icons, formatting, and animations on the generated PPT; (2) AI-assisted editing via a dialog triggered from the sidebar or main chat, allowing natural-language revision requests.

2. Browser Automation for Tedious Data Collection

Doubao Work can control a browser to perform repetitive web tasks. In the livestream demo, the author observed it: (a) open a procurement system link, navigate to the report center, apply filters, and export Excel; (b) scrape Douyin high-engagement posts, extracting titles, links, like counts, and favorite counts into an online spreadsheet; (c) enter a Feishu group, download five inconsistently formatted tables, clean them, and merge into a single unified table.

The author replicated this by combining computer control, browser control, and data visualization skills to build a "2026 Domestic GPU Half-Year Report Dashboard." Manual data collection alone would have taken a full day; Doubao Work completed collection and cross-verification in under one hour.

3. Serverless Shareable App Generation

The "App Generation" skill lets users create interactive, shareable web applications without servers or domains. A shareable link works for anyone, unlike a local HTML file. The livestream showcased three Three Squirrels (三只松鼠) internal tools built from scratch: (1) a recruitment website; (2) a sales analysis dashboard consolidating eight inconsistent channel tables from a Feishu group; (3) an AI customer-service training platform where a multi-dimensional table maintains a QA knowledge base, the AI simulates a difficult customer for five dialogue rounds, auto-scores each round, provides a turn-by-turn review, and suggests a gold-standard answer.

The author found the first two examples basic, but the training platform was revelatory: "AI must understand company business to produce usable apps." A key distinction was drawn: "Only viewable and scrollable is a webpage; embed AI capabilities, enable interaction, judgment, and bi-directional linkage with multi-dimensional tables — that's an application." Such solutions previously required weeks of dev-team scheduling; now they can be built in one evening with plain language.

4. Custom Skill Development: The 3-2-1 Framework

A guest shared a "321" framework for skill mastery:

Three questions to understand skills: which useful skills exist (copy from others), where to find them, and how to build your own.

Two steps to create a skill: first, work with Doubao to clarify every workflow step of the task; second, refine each step's output standard to a 90-point level by running multiple test cases, iterating feedback, and stabilizing before solidifying as a reusable skill.

One-click sharing: finished skills can be uploaded to the enterprise workspace and synced to the whole team.

The author advises relying more on Skill Creator than Find Skills, noting that the most frequently used and best-fitting skills are self-made.

5. Product Feedback and Verdict

The author lists three improvement requests: (1) inconsistent creation logic — new partner creation lives under "Partner Chat" instead of "Skills·Connectors·Partners → Work Partners·Squads"; (2) enterprise login forces work mode and disables the "Call" function; (3) when one phone controls two computers, the task list is cluttered and computer icons are identical, making it impossible to distinguish which machine a task ran on without opening the task detail.

Despite these issues, the author concludes: "Doubao Work is already an indispensable productivity agent, and the best domestic one I've used. Drop your stereotypes and try it — you'll thank me later."