Blog
Letters for builders moving Agents from local experiments into production systems.
A letter to builders on moving Agent Work from local experiments toward a Cloud Agent Session Lifecycle API.
From the end credits of Avengers: Endgame to the making of PUBG: agents with clean context, useful deviations, and the human time it takes to exceed expectations.
Mosoo Computer gives an agent a cloud workspace that accumulates useful state and connects it to the channels you already use, so the next task starts from what the last one left behind.
If Cloud Agents become mainstream, developers will need more than another layer of Agent CRUD. They will need a neutral control plane that can resolve, run, and govern Environments. This memo works backward from that future.
People are forming two very different expectations of agents: one has an identity and memory and is cultivated over time; the other is stateless and replaceable, using a freer reasoning pipeline to replace fixed workflows.
ghfind delegates GitHub project evaluation to one published Mosoo Agent while keeping its rubric, job queue, result validation, and product UI in the application.
Why having an API is no longer enough for your web application in the Coding Agent era—and why the CLI and CLI-Skill should become the default contract for development and testing from Day 1.
Before deciding whether an agent loop belongs in a control plane or a sandbox, define which facts must survive and which execution state may disappear. mosoo and Agent Driver provide a concrete, testable example.
OpenAI's Responses API brings state, compaction, tool programs, WebSocket continuation, and multi-agent coordination into one model runtime. Here is what changed—and what still belongs to the application.
Local coding agents showed what models can do with a real environment. The next cloud abstraction is a managed agent task: work in, progress and artifacts out.
