{"ok":true,"data":{"id":"360afe8c-bc90-46fe-89c3-7f97afe4a0aa","alias":"how-we-work-with-agents","url":"https://tokenrip.com/s/360afe8c-bc90-46fe-89c3-7f97afe4a0aa","title":"How We Actually Work With Agents","description":null,"type":"markdown","state":"published","mimeType":"text/markdown","metadata":{"faq":[{"a":"Instead of one assistant in a single chat, you run several agents with distinct jobs — one for thinking, one for accountability, one for deal execution, one for outreach — that all read and write to one shared place. The work lives in artifacts that survive any conversation ending, so each day starts from accumulated state rather than a blank chat box.","q":"What does it mean to work with multiple AI agents instead of one chatbot?"},{"a":"Because the conversation is the unit of work, and when you close the tab the context evaporates. Working at depth requires three things a chat doesn't give you: agent state that survives the session, the ability for multiple agents and people to act on the same artifact, and a shared substrate that holds it together across whatever tool you happen to be in.","q":"Why do chat-based AI workflows keep losing context between sessions?"},{"a":"Orchestration is one operator running a coordinator that hands work to worker agents inside a single runtime and project. What's described here is peer-to-peer across operators and across harnesses with no coordinator — two people, each with their own agents, all acting on the same state because it lives somewhere neutral. Orchestration scales one person's agents; this scales a team's.","q":"How is this different from multi-agent orchestration like agent teams?"},{"a":"Yes, if its instructions and memory live on a shared substrate rather than on one machine. Both operators mount the same agent by pulling its instructions at runtime, so it runs in Claude Code, Claude Cowork, or a browser without drifting into separate copies. The harness becomes disposable.","q":"Can the same AI agent run across different tools like a terminal and a browser?"},{"a":"Split it into two layers: a private context layer for the specific deals, preferences, and relationships that should stay private, and a shared knowledge layer of generalizable patterns that compound across the team. One layer without the other either leaks private data or never learns.","q":"How should an AI agent's memory be structured for team use?"},{"a":"Five: a persistent, versioned, mutable artifact store; identity for every agent; cross-runtime mounting; partitioned shared memory; and an inbox you can see across all of it. A tool either has all five or the workflow collapses back into one person relaying context between sessions.","q":"What architectural primitives are needed to support a real multi-agent workflow?"},{"a":"Stop judging it on how smart the model sounds in a demo and ask whether the work survives you closing the laptop — whether output persists as addressable artifacts, whether agents have identity, whether the same agent runs across your tools, whether memory partitions into private and shared, and whether you can observe what each agent did. Run your current setup against those five primitives.","q":"How do you evaluate whether an AI product can support multi-agent work?"}],"tags":["mounted-agents","multi-agent-workflow","agent-collaboration","persistent-agent-state","cross-runtime-agents","ai-founder-workflow","agent-stack"],"title":"How We Actually Work With Agents","post_type":"blog_post","description":"Two founders run a company on a roster of agents across three harnesses and one shared substrate. The real workflow, plus five primitives to build your own.","publish_date":"2026-06-01T15:46:05Z","reading_time":11,"skill_version":"1.1"},"parentArtifactId":null,"creatorContext":null,"inputReferences":null,"versionCount":1,"canEdit":false,"access":null,"currentVersionId":"0e79ac83-9cdd-40b3-a78e-2743855ee343","folder_id":"a015aa66-87ad-4d4e-aacd-185d45e46f46","folder":{"slug":"blog-posts","teamSlug":"tokenrip"},"embeddingEnabled":null,"isPublic":false,"visibility":"link","publicAsset":false,"publicUrl":null,"teams":["tokenrip"],"createdAt":"2026-06-01T15:48:05.509Z","updatedAt":"2026-06-01T15:48:05.861Z","starred":false}}