Reliable Cross-Account API Integrations

Diving deeper into

Head of Product at SaaS startup on building a personal AI OS with Codex automations and Claude Cowork

Interview
a lot of stuff I built with custom API connections, because it's important for it to work across so many accounts
Analyzed 3 sources

The hard part is not getting an agent into one app, it is making one identity reliably act across many copies of the same app. In practice this setup is less like turning on a plugin and more like building a thin integration layer, a Google Cloud app, API scopes for Gmail and Calendar, account by account auth, a CLI Codex can call, and a health check that keeps tokens from silently expiring across personal, work, and side project accounts.

  • The Google Workspace integration shows what custom really means here. It required creating a Google Cloud app, enabling Gmail, Calendar, Slides, and related APIs, connecting multiple accounts, extending token life, and adding a daily connection check because auth failures kept breaking workflows.
  • This is why API connections matter even when browser use exists. Browser and computer use let the agent log into sites as the user, but API links are what make repeated read and write actions stable enough for daily jobs like scanning inboxes, moving calendar events, and drafting across accounts.
  • The broader pattern is that reliability falls as the tool chain gets longer. Another operator found one to three connected tools usually work, but four or five tool chains start passing errors downstream, which is why power users keep building narrow custom glue instead of trusting a fully general agent stack.

This is heading toward a split market. The winners will be products that hide OAuth, scopes, retries, and account routing behind a simple chat surface, while still giving power users logs and control when something breaks. Until then, personal AI OS setups will keep looking like handcrafted middleware wrapped around a model.