Theme 1 – Convenience vs. Vendor Lock‑in / Trust
Many commenters see the managed agent API as a way to avoid the hassle of running their own sandbox, but they worry it deepens reliance on OpenAI and raises trust issues.
- “To save yourself the hassle of running your own sandboxed VM.” – simonw
- “Because how does OpenAI earn more money then? … it seems to try more vendor lock‑in.” – pixl97
- “Not sure I exactly trust OAI to do that right.” – pixl97
- “Perfect for when you want your data to be stolen programmatically.” – chairhairair
Theme 2 – Pricing, Quotas, and Cost Concerns
Users report that interacting with Codex (or similar agents) quickly consumes their usage limits, making the service feel expensive or unusable for casual work.
- “Since a week or so everything I ask codex to do … uses at least 1% of my weekly limits and like 5% of my 5h limit.” – spwa4
- “It sits there and spins for a bit then as soon as it spits out something, my 5h is 5‑10% lower …” – viccis
- “The pricing on this is a bit confusing… Is there a way to deliberately shut down an environment so you don’t have to keep paying for it?” – simonw
Theme 3 – Technical Details: Sandbox Configuration, Security, and Abstraction
Discussion frequently turns to how the sandbox can be configured (network access policies), whether those controls are trustworthy, and the trade‑offs between using an SDK for development versus an API for deployment.
- “It offers three settings:
enabled,disabled,restricted… Given what happened with the wiki thing … how trustworthy is that restricted option?” – simonw - “I assume you'd develop via the SDK, then deploy it via the API.” – kakugawa
- “You can give the agent a tool … which waits for events. Agent calls it and the tool sleeps until an event happens then returns it to the agent.” – dist-epoch
- “The harness needs to be running … you can just make new sessions for each new task?” – krashidov (and replies about daemon‑style remote control)