what it is
An Anthropic product for building and deploying agents at scale. It pairs a tuned agent harness with production infrastructure, pitched as “prototype to launch in days.”
The UI showed these building blocks.
- Deep research, multi-step web research with cited sources
- RAG retrieval over a vector store or document collection
- An intent router that sends each request to the right sub-agent
- A draft generator for emails, docs, summaries and reports
- A reflection loop that critiques output until it clears a quality bar
- A task planner that breaks a goal into ordered sub-tasks
- A summarizer for long thread history
You can start from an existing spec, type / to add skills or sub-agents, or let Claude interview you. The example prompt in the demo was “Build an agent that evaluates acquisition targets.”
what I found
I spotted it in an Instagram reel on 2026-04-20, with the caption “Anthropic just killed 1,000+ startups.” The questions I wrote down were latency, tool-use reliability and cost against a custom Claude Code harness.
compared to
- Claude Code is the harness I build on today.