Every action is another chance to go wrong
When the AI picks each action as it goes, a job is only as good as the whole chain. Get nineteen actions in twenty right, and a twenty-action job still comes out right only about a third of the time.
Compare
Faber vs ChatGPT and Claude.
Both use AI on your work. The difference is whether you have to be there: a chat assistant works while you are in the conversation, and Faber works when you are not.
| Faber | A chat assistant | |
|---|---|---|
| When it works | On a schedule, when work arrives, or on demand, with nobody watching. | While you are in the conversation. |
| Before it runs | A full preview run you approve, with nothing sent or changed. | There is nothing to preview. It acts as you talk. |
| Approvals | Mark any action to require your approval after each run. The rest goes on its own. | You are watching it happen, which only works while you are watching. |
| What decides the actions | Decided once, when the task is built, and you approve them before anything is real. | Decided fresh, every time you ask. |
| Recurring follow-up | Reads the reply, acts on it, and keeps going until the job is done. | Each conversation starts again. |
| Record of each run | A plain-English record of what it checked, did, set aside, and skipped. | Chat history. |
| Setting it up | Describe the work once and the AI builds it. You answer a couple of questions. | Explain it again each time, or keep a prompt somewhere and paste it in. |
| Built for | Recurring work you would rather stop doing yourself. | Thinking, drafting, and answering, with you in the loop. |
When not to use Faber
A chat assistant is better at plenty of things.
Faber is not built for one-off work and does not try to be. Open a chat for the thinking, the drafting, and the thing you need an answer to before lunch. Faber earns its place when the same work comes back next week, and the week after, and you would rather it just happened.
If you are already using agent mode
Why long jobs come apart.
Faber decides once, not every run
The AI works out the actions when it builds the task. After that it follows them, using AI only where an action needs judgment, like reading a reply or drafting a message.
Which is what makes the preview worth reading
A preview only tells you about tomorrow if the actions are settled. You read the whole first run before any of it is real, and that is what keeps happening. Only the mail and the rows change.
Stop explaining the same job every week.
Describe it once, read the preview run, and let it happen.