Autonomy and spend
Autonomy decides whether Connect may act; the plan allowance and the workspace budget decide how much it can. The two compose as an intersection, never as an override: an autonomous channel with a spent allowance sends nothing, and raising the mode does not buy capacity. A refusal for spend is recorded with that reason, distinctly from a compliance one.
Two controls, one outcome#
| Autonomy | Allowance | What happens |
|---|---|---|
autonomous | Available | The action goes out |
autonomous | Spent | Nothing is sent; the refusal names the limit |
ask_before_send | Available | Held for a decision, then sent on approval |
ask_before_send | Spent | The approval is recorded and the send waits; the item stays until it goes |
draft_only or off | Either | Nothing is sent, and capacity is not the reason |
The fourth row is the one worth reading twice. Approving something the workspace cannot currently send does not fail and does not vanish: the decision is kept, and the action waits for capacity. That is why an approval queue can look worked-through while the sent folder stays still.
The daily ledger#
A customer workspace's use is bounded by its plan's allowances, counted in a ledger that resets. The Owner's own workspace has no plan and no such gates at all — that is the single commercial difference between the two audiences, and it is not a difference in what the feature can do. Everything else about autonomy behaves identically on both sides.
Plan and Usage is where the day's position is visible. Read it before loosening a channel: moving email from ask_before_send to autonomous usually raises the number of sends per day, and the ledger is what that increase runs into.
Calls are priced individually#
Voice is the channel where capacity is felt per action rather than per day. A call is priced on the audio it uses and metered against the workspace's AI quota, and a call the budget cannot cover does not happen. On a voice channel set to autonomous this is the gate most likely to produce a refusal you did not expect, because nothing about the recipient or the message is wrong.
Two practical consequences. Long calls cost more than short ones, so a channel that has been reworded to answer briefly goes further on the same budget. And a queue of due call-backs approved all at once can exhaust the budget partway through — which is visible as some calls placed and the rest refused, in the order they were attempted.
Reading a stoppage correctly#
- Everything stopped at once, mid-morning
- Look at the allowance first. A ledger that hits its limit stops every channel at once, which reads like an outage and is not one.
- One channel stopped, the others continue
- Not the ledger — that is per workspace. Look at the provider for that channel, or the channel's own mode.
- Calls stopped but email continues
- The AI budget for voice rather than the daily send limit. The refusal reason distinguishes them.
- Nothing at all is happening, including drafting
- Not a spend limit. Check whether Connect is switched on — a stopped runtime refuses nothing because it evaluates nothing.
Questions#
Does raising a channel to autonomous increase my allowance?
No. The two are independent controls, and the mode is the one about permission. Loosening it typically increases the number of sends a day, which means the limit is reached sooner rather than later.
If the allowance runs out, do I lose the replies Connect had written?
No. Work already prepared stays where it is, and an approval given during a stoppage is kept and acted on when capacity returns. The item stays in the queue until it actually goes.
Does the Owner workspace have the same limits?
No — the Owner has no plan and no such gates. Every other part of autonomy is the same on both sides, over one implementation; only the commercial ceiling differs.