⚛️量子位•Freshcollected in 59m
Tiangong 3.1 Launches Skywork Design and Dynamic Workflows
💡Discover how Tiangong 3.1 is evolving agentic workflows to handle complex, multi-step business tasks.
⚡ 30-Second TL;DR
What Changed
Launched Skywork Design for creative and visual AI tasks
Why It Matters
The addition of dynamic workflows suggests a shift toward more autonomous, multi-step agentic processes, moving beyond simple chat interfaces.
What To Do Next
Explore the Tiangong 3.1 API documentation to integrate Dynamic Workflows into your existing agentic automation pipelines.
Who should care:Developers & AI Engineers
🧠 Deep Insight
Web-grounded analysis with 21 cited sources.
🔑 Enhanced Key Takeaways
- •Skywork Design integrates advanced visual AI capabilities like text-to-image generation, batch creation, inpainting/outpainting, and multi-image character consistency directly within the unified workspace, leveraging models such as Nano Banana Pro for high-resolution outputs.
- •The 'Dynamic Workflows' enable Skywork AI's 'Super Agents' to autonomously plan and execute complex, multi-step tasks by dynamically selecting from over 20 specialized tools and leveraging a 'DeepResearch' engine for real-time, verifiable information gathering across 600+ web sources.
- •Skywork AI's platform, also referred to as Skywork 3.0, functions as an all-in-one multimodal productivity workspace, integrating various leading AI models like Claude Opus 4.7, GPT 5.5, and Gemini 3 Pro, to generate diverse content from documents and presentations to code and websites.
- •The company's underlying technology stack centers on a powerful LLM augmented by a retrieval engine, multimodal capabilities, and a knowledge graph, with specialized 'Critic' models (e.g., Skywork-Critic-Llama3.1-70B) developed for pairwise preference evaluation and data improvement.
- •Skywork AI, incorporated in Singapore in July 2021, has secured investment from entities including Beijing Kunlun Internet Intelligent Industry Investment Fund Partnership and Beijing Artificial Intelligence Industry Investment Fund, supporting its development in AIGC and agentic AI.
📊 Competitor Analysis▸ Show
| Feature / Platform | Skywork AI | Microsoft Copilot (M365) | Salesforce Agentforce | OpenAI Frontier |
|---|---|---|---|---|
| Core Focus | Unified AI Workspace, AIGC, Deep Research | AI assistance within Microsoft 365 ecosystem | CRM-centric AI agents for sales/service | Enterprise agent orchestration & management |
| Agentic Capabilities | Super Agents for complex, multi-step tasks; dynamic tool selection (20+ tools); Dynamic Workflows | Intelligent automation for M365 tasks; multi-step actions across apps | Autonomous agents for CRM workflows; no-code/low-code agent building | Platform to build, deploy, manage AI agents; provider-agnostic SDK |
| Multimodal Generation | Yes (text, images, slides, websites, podcasts, data visuals) | Yes (content creation, data analysis, summaries within M365) | Limited (primarily text-based within CRM context) | Varies by integrated models/agents |
| Key Integrations | DeepResearch (600+ web sources), multiple LLMs (Claude, GPT, Gemini) | Deeply integrated with Microsoft 365 apps (Outlook, Teams, Excel, Word, PowerPoint) | Deeply integrated with Salesforce CRM and cloud services | Supports 100+ LLMs from various vendors |
| Target Users | Professionals, creators, analysts, educators seeking integrated productivity | Microsoft 365 users, enterprises | Salesforce-centric organizations, sales, customer service | Enterprises building and managing custom AI agent solutions |
🛠️ Technical Deep Dive
- Skywork AI's core technology stack is built around a powerful Large Language Model (LLM) augmented by a retrieval engine, specialized tools, and multimodal capabilities.
- The 'DeepResearch' engine is a key component, enabling parallel processing and autonomous browser sessions to navigate dynamic web content and gated information across over 600 web sources. It maps findings to verifiable sources to ensure E-E-A-T (Expertise, Authoritativeness, Trustworthiness) standards.
- The platform integrates multimodal models that combine vision and language, allowing for diverse content generation.
- AI agents within the platform can dynamically select from a curated set of over 20 specialized tools to plan and execute complex tasks.
- Skywork AI is developing a knowledge graph to enhance its AI's reasoning capabilities and to connect its platform with enterprise data.
- The company has developed specialized 'Critic' models, such as Skywork-Critic-Llama3.1-8B and Skywork-Critic-Llama3.1-70B, which are built on Meta Llama-3.1-70B-Instruct and Llama-3.1-8B-Instruct. These models are fine-tuned with diverse datasets for pairwise preference evaluation, data improvement, and reward modeling.
- The 'Shang Tsung' skill provides persistent memory and identity continuity for AI agents, utilizing a 'Second Brain' concept for daily logs and long-term memory, combined with SOULS session lineage.
- The agent architecture follows a closed-loop system that perceives inputs, reasons and plans, calls tools and retrieves knowledge, acts in an environment, observes outcomes, and continuously refines its actions, often employing ReAct (Thought → Action → Observation) patterns.
🔮 Future ImplicationsAI analysis grounded in cited sources
Skywork AI is poised to become a significant player in the integrated AI workspace market.
Its focus on a unified platform for multimodal content generation, coupled with advanced agentic capabilities and a robust research engine, addresses the market need for streamlined, end-to-end AI-powered workflows, reducing reliance on fragmented tools.
The emphasis on 'Dynamic Workflows' will accelerate the adoption of autonomous AI agents in professional settings.
By enabling AI to dynamically plan, execute, and adapt complex tasks with minimal human intervention, Skywork AI's approach can significantly boost efficiency and scalability across various industries.
Skywork Design's integrated visual AI capabilities will democratize high-quality content creation for non-designers.
By offering features like text-to-image, inpainting, and multi-image consistency directly within a productivity suite, it lowers the barrier for professionals to generate sophisticated visuals for diverse applications.
⏳ Timeline
2021-07
SKYWORK AI PTE. LTD. incorporated in Singapore.
2024-08
SkyworkAI Alignment Team develops Skywork-Critic-Llama3.1-8B and Skywork-Critic-Llama3.1-70B models.
2025-05
Skywork.ai platform launched globally as the 'DeepResearch-Powered Workspace'.
2025-08
Skywork AI receives investment from Beijing Kunlun Internet Intelligent Industry Investment Fund Partnership and Beijing Milaiwu Network Technology Co., Ltd.
2025-12
Skywork.ai details its flagship product, Skywork, as a pioneering multimodal productivity platform.
2026-03
Skywork Design features and pricing details are outlined, integrating visual AI capabilities.
📎 Sources (21)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 量子位 ↗