Artifacts
Also written: Artifacts feature, Artifacts tab, Manifest, Agentic artifacts, Production-ready assets, Analysis assets, Artifact limit, Attachments
Tangible outputs from AI agents, like documents, code, or visuals, that can be saved, shared, and built upon.
Artifacts are the concrete results or deliverables that an AI agent produces. Think of them as the tangible things an AI creates that you can see, save, and use later.
For example, if you ask an AI to help you write a report, the final document it generates is an artifact. Other examples include code snippets, summaries of information, or even custom charts and diagrams.
Artifacts represent persistent, shareable outputs generated by AI systems, often as a result of agentic workflows or specific user requests. These outputs can range from textual documents, code, and data analyses to more complex deliverables like interactive visualizations or executable scripts.
Their significance lies in moving beyond ephemeral conversational turns to create lasting, actionable results. This enables collaboration, iteration, and integration into broader software development or business processes. The ability to manage and version these artifacts is crucial for reliable AI agent development.
Anthropic is heavily promoting its new 'artifacts' feature, which allows users to create and save outputs like documents, designs, and code directly within Claude. This move suggests a push towards making AI assistants more useful for persistent work and collaboration, moving beyond simple chat interactions. Other companies are also exploring ways to integrate AI-generated outputs into workflows, as seen with GitHub Copilot plugins.
- Getting started with Claude Code ProjectsAnthropic · Claude on YouTube ·
- Build an App With Claude DesignAnthropic · Claude on YouTube ·
- Artifact usage promotionAnthropic · Claude Help Center ·
- Enabling Consistent AI-Assisted Engineering with GitHub Copilot PluginsMicrosoft · Microsoft for Developers ·
- Get started with Claude DocsAnthropic · Claude Help Center ·
- Share artifactsAnthropic · Claude Help Center ·
- What are artifacts and how do I use them?Anthropic · Claude Help Center ·
- Anthropic merges Claude chat and Cowork in one interfaceAnthropic · TechCrunch AI ·
- Claude comes for Gemini with its own take on Docs and SlidesAnthropic · The Verge AI ·
- Education Report: How educators use ClaudeAnthropic · Anthropic Research ·
- .blend URL ViewerOpenAI · Simon Willison ·
- Testing Agent Skills Systematically with EvalsOpenAI · OpenAI Developers Blog ·
- Agent Plugins package your skills, tools, and moreGoogle · Google Developers Blog ·
- Custom visuals in chat and CoworkAnthropic · Claude Help Center ·
- Publish and share artifactsAnthropic · Claude Help Center ·
- Copilot Notebooks now works with Markdown, plain-text, and rich-text filesMicrosoft · Microsoft 365 Copilot Blog ·
- Data science workflows with ChatGPT WorkOpenAI · OpenAI News ·
- Operations workflows with ChatGPT WorkOpenAI · OpenAI News ·
- What Parameter Golf taught us about AI-assisted researchOpenAI · OpenAI News ·
- Everyday work with ChatGPT WorkOpenAI · OpenAI News ·
- Creating images with ChatGPTOpenAI · OpenAI News ·
- Google Antigravity: Hands on with our new agentic development platformGoogle · Google AI: Release Notes ·
- Introducing Google AntigravityGoogle · Google DeepMind Blog ·
- Lowe’s puts project expertise into every handOpenAI · OpenAI News ·
- Anthropic Claude
- Anthropic's Claude now features 'artifacts' for creating and saving outputs like documents, designs, and code directly within the chat interface, and is promoting their use.
- Google Gemini
- Google Gemini is exploring ways to integrate AI-generated outputs into workflows, potentially including features similar to artifacts.
- Microsoft Copilot
- Microsoft Copilot is enabling the sharing and updating of custom AI coding tools, which can be considered a form of artifact.
- Ephemeral Outputs
- These are temporary results, like a single chatbot response t
- Training Data
- This is the information used to teach an AI, whereas artifacts ar
- Model Weights
- These are the internal parameters of an AI model, not the externa
The term 'artifact' in software development refers to any output of the development process. In the context of AI, it has been adopted to describe the tangible results produced by AI agents or systems.
In Claude, try asking it to write a short Python script to print 'Hello, World!' and then save or share the resulting code. In ChatGPT, ask it to summarize a provided text and then copy the summary to use elsewhere. These saved outputs are examples of artifacts.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- Eval methodEvaluate agent skills with captured runs and checksTest agent skills by capturing runs and scoring them against predefined checks. · 30 min
- HeuristicBuild trust in agentic work with task-level abstraction and artifactsProvide context on agent work at a task level with verifiable artifacts. · 5 min
- HeuristicEnable asynchronous feedback for agent self-improvementAllow intuitive async user feedback across all agent surfaces and artifacts. · 5 min
- Setting tipEnable artifacts in Claude for code execution and file creationTurn on artifacts in Claude settings to allow code execution and file creation for enhanced tool building. · 1 min
- ChecklistClaude Code Projects setup and usage checklistSteps to set up, manage threads, review pull requests, and create artifacts in Claude Code Projects. · 5 min
Questions people ask
- What is the difference between an artifact and a chatbot response?
- A chatbot response is often temporary and part of a live conversation. An artifact is a saved, persistent output that can be shared, edited, or used in other applications.
- Can AI create different types of artifacts?
- Yes, AI can create various artifacts including text documents, code, data visualizations, summaries, plans, and even interactive elements.
First seen on Subvolts May 7, 2025. Updated Oct 16, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.