TL;DR — I've created agents. The orchestration has been challenging, but a whole lot of fun now that it is working. I'm building a SaaS/PaaS platform as a solo dev, so dev eats all my time and marketing / outreach always gets dropped. The challenge was keeping me, Claude Cowork, Claude Chat, Claude Code, and my agents on the same context. 1. I created a context.md file that lives on a Google Drive mirror on my local machine. Both during and at the end of each Cowork session, Cowork updates it — strategy, new DB table names, edge functions, anything needed to talk about the whole platform. 2. Claude Chat can't natively read md files in Drive, so I deployed the context file to the cloud with a token for authorized access, so Chat can fetch it when I start a new conversation on my phone. 3. I built two skills: an outreach agent (direct email to users) and a content agent (social media copy, blog posts, video scripts). At the end of each Cowork session, Cowork updates the context doc and Code pushes it to the cloud. When I start the next Cowork session, it's like getting a briefing from a Chief of Staff — recap, leftovers, backlog, and proposed work orders for the agents. 4. I review the proposed work orders and tell Cowork to create them. Work orders are written into the context.md file, which the agent skills can read. 5. I activate the skills; each agent reads the context doc, finds its work orders, produces output in assigned folders, updates status to complete, and notifies me in the same Cowork session for review. If approved, the agents schedule the content.
Arquitectura de un fundador SaaS solista: contexto compartido entre Cowork, Chat, Code y agentes
Un desarrollador solo usa Cowork como capa de Chief of Staff: un único context.md en Google Drive como fuente de verdad, y Cowork delega work orders a agentes (outreach + contenido), mientras Chat móvil y Code leen el mismo archivo
Solo founder building a SaaS/PaaS platform with Claude Cowork as control plane
Escenario
Un desarrollador solo usa Cowork, Chat móvil, Code y varias skills de agentes en paralelo. Problema: cada superficie tiene su propio silo de contexto, y mantener al humano + todas las superficies + agentes sincronizados es casi imposible — así que marketing y outreach siempre se postergan.
Prompt
At the end of each working session, update context.md at /path/to/GoogleDriveMirror/ with: strategy changes, new database table names / edge functions, UX decisions, and any info needed to discuss the entire platform. When I start the next Cowork session, read context.md and brief me like a Chief of Staff would: last session recap, leftover items, backlog, and proposed work orders for my outreach and content agent skills. I'll review and tell you which work orders to commit; append approved ones to context.md so the agents can pick them up when I activate them.
Resultado Esperado
Construyó una arquitectura estilo Chief of Staff: 1. **context.md** (espejo local de Google Drive) = única fuente de verdad 2. **Cowork** — actualiza context.md durante y al final de cada sesión; al inicio de la siguiente hace un briefing tipo Chief of Staff 3. **Claude Code** — sube context.md a un endpoint con token para que Chat pueda leerlo 4. **Claude Chat (móvil)** — lee el mismo archivo mediante token autorizado 5. **Dos skills** (outreach + contenido) — leen context.md, localizan sus work orders, dejan la salida en carpetas asignadas, marcan el estado como completado y notifican al usuario en la misma sesión de Cowork **Resultado:** cada sesión de Cowork empieza con un briefing tipo Chief of Staff; marketing y outreach ya no se postergan; todos los agentes comparten la misma fuente de verdad.