The Autron AI Agent never changes anything in your account without your explicit approval. There are two confirmation patterns, depending on what you're changing — plus a small set of things the agent will simply refuse to do.
Two confirmation flows
Goal changes — chat preview, then "yes"
When you ask the agent to create, update, enable, disable, or change a goal, it follows a two‑step pattern:
The agent describes the change in chat — e.g. "I'll set your garlic press to a Maximize Sales goal with a 25% ACoS target. Confirm?"
You reply with explicit confirmation — "yes", "do it", "confirmed".
Only then does the agent apply the change.
If you reply with anything else, or want to tweak the proposal, the agent waits — it won't apply the change without that explicit "yes".
Amazon Ads changes — Approve / Deny card
When the agent is about to change something in your Amazon Ads account directly — create or pause campaigns, update budgets, change bids, add or remove targets, archive — it shows an Approve / Deny card with the full set of changes laid out:
Click Approve to apply the whole batch in one go.
Click Deny and nothing happens — you can ask the agent to revise and re‑propose.
One approval covers the whole batch — even if it's 50 campaigns or 1,000 bid updates.
Why two flows? Goals are Autron settings — the chat preview gives you a chance to fine‑tune wording before you commit. Direct Amazon Ads changes are bulk operations on Amazon's side — the structured card gives you the exact list of mutations and a single audit trail per approval.
Reads run automatically
Anything that just looks at your data — performance metrics, campaign lists, target lookups, keyword opportunities, profitability — runs automatically with no approval prompt. Only changes need approval.
Reviewing what's been done
Every Amazon Ads change you approve is recorded in the Action log — a filterable audit trail with timestamps, sellers, marketplaces, tools, and statuses. Open it any time to see exactly what the agent has done on your behalf, including which batches succeeded, partially succeeded, or failed.
When something fails
If a write fails after you've approved it (Amazon's side rejects it for a real reason), the agent gives you one short, plain‑English explanation of what got blocked — for example:
"Amazon couldn't find that target — it may have been archived or belong to a different campaign."
"That bid is below Amazon's minimum for this marketplace."
"Amazon blocked the campaign create — looks like a duplicate name."
It will not retry automatically and will not chain into another speculative attempt. You decide whether to retry, adjust, or move on. The Action log captures the failure with Amazon's full response so you can dig in later if needed.
What the agent will refuse to do
The agent is scoped to Amazon Ads, Autron metrics, and Autron features. It will politely refuse:
Off‑topic questions — anything outside Amazon Ads / Autron (general product advice, tax/investment advice, unrelated channels).
Reactivating archived items — archive is a terminal state on Amazon's side. The agent won't pretend it can be undone.
Reading data it doesn't have — if a tool isn't available in your session (rare), the agent will tell you instead of guessing.
It also won't:
Fabricate numbers — if it doesn't have data for a question, it says so.
Show internal IDs — you'll see product names and ASINs, never raw UUIDs.
The short version
Action | How it's confirmed |
Reading data, listing entities, checking metrics | No approval — runs automatically |
Goal changes (create / update / enable / disable) | Chat preview + your "yes" |
Amazon Ads changes (campaigns, bids, targets, ads, archive) | Approve / Deny card — all approved changes recorded in the Action log |