Artifacts, the Claude feature Anthropic introduced in 2024, should become the primary interface for collaborating with AI agents — and most Claude users are not using it that way. That is the argument Thariq Shihipar, Member of the Technical Staff at Anthropic, made in a recent Latent Space podcast interview.

Shihipar's assessment is grounded in observable user behavior. A Claude AI Facebook group with nearly three million users rarely hosts discussions about Artifacts, a signal he reads as evidence the feature is poorly understood and underused as a project-management tool.

What does Shihipar want users to do differently?

He wants Claude users to stop treating chat as the sole channel for steering agents and start building persistent, live documents — dashboards — that agents can read, update and work against over the life of a project.

He said: "And so one thing that people are not doing yet that I'm trying to encourage is this idea of a dashboard artifact. So you have Claude working on a project long term. Maybe it's like a Kanban or something. It can store that Kanban data in a database. Multiple Claudes can access that data via the Artifact MCP. And that Artifact can talk to those Claudes as well."

He acknowledged the model is harder to build than a simple prompt-response flow: "But there's a lot of work to do because it's so much more complicated than a multiple choice question. There's a lot more detail in terms of diagrams and code snippets and schemas or whatever it is for that problem. But Artifacts is the more AGI-pilled way of doing the, 'ask you the question,' basically."

Did anyone push back?

Yes. Podcast co-host SWYX of Latent Space challenged the premise, saying he works with Claude directly, with zero Artifacts, and finds the division of labor unclear. His counterpoint: "I think one thing that's unclear to me about the Artifact stuff is what feedback should go in through the Artifact and what feedback should go through a Claude, the chat. Because the more AGI-pilled one is to just feed everything to the Claude."

Shihipar answered directly that the Artifact is the better long-term channel: "I think the more AGI-pilled one is to go through the artifact... I think that Artifacts will be your interface into the harness. You can comment on this live document of your plan of the work. You can see maybe multiple agents and different agents are doing this. And that Artifact is built for the current work that you're doing... I think on-the-fly interface for your harness is probably where things are headed."

How would the architecture actually work?

Co-host Vihu Sapra pressed on why users should not simply do everything through the current Claude Code setup, where chat already returns rich HTML mock-ups and diagrams. Shihipar's answer describes a separation of concerns across three layers:

• The display layer — an Artifact hosted somewhere, with its own database, that surfaces all agent work.
• The intelligence layer — inference running in the cloud, so users do not have to keep their computer on.
• The execution layer — "local hands," the ability for a cloud-based agent to access a user's computer when it is online, spin off multiple sub-agents, and have those sub-agents communicate with each other.

He framed this as "unpackaging the Claude Code experience," which today runs everything in one place. In his vision, a user messages a cloud-based Claude that can run either local or cloud sessions — similar to how Claude Tag works today, with local hands to be added over time.

What is confirmed versus speculative?

Confirmed by an Anthropic staff member: Artifacts shipped in 2024; Anthropic is developing the infrastructure toward multi-agent workflows with Artifact-based interfaces; adoption of Artifacts as a dashboard concept among Claude users is currently low. Speculative, and presented by Shihipar as direction rather than shipping product: the full separation of UI, inference and execution layers, and "local hands" for accessing a user's computer. He explicitly said Anthropic is "still getting there from an infrastructure perspective."

What should you take away?

For teams already running agents on long projects — SEO workflows, content pipelines, codebases — Shihipar's recommendation is concrete: use Artifacts as the persistent layer where requirements, plans and schemas live, instead of burying context in chat history. Artifacts help users see what agents are doing and guide their work more effectively, he argued.

He also conceded Anthropic has more work to do before his vision of agent collaboration through Artifacts is fully realized. Watch for signs of adoption — Artifact-based dashboards appearing in that three-million-user Facebook group, and Anthropic shipping the local-and-cloud session architecture he described — as the indicator of whether this interface shift actually takes hold.