Overview
Every AI assistant sometimes gets stuck — a tool that keeps erroring, an instruction it has misread, a request it keeps getting wrong. Most products leave you to argue with it. AgenticWebTeam builds in an escalation path: a supervisor.
The supervisor is a more capable AI model that reviews the whole conversation — not just the messages, but the tools that ran and their errors, the instructions the assistant was working from, what it remembers about you and your site, and your feedback. It can even read the AWT platform's own code, read-only, to tell a genuine product bug from a misunderstanding.
What you get back is a short, plain-language diagnosis in your own language, plus up to four proposed fixes in a card in the chat. You apply or dismiss each one individually — and every applied fix records exactly what it replaced, so Undo genuinely restores it.
How it works
- A review starts — you tap Call a supervisor in the chat; the assistant calls one itself when it is stuck and tells you it has; or one starts automatically after a failed run, repeated tool errors, a 👎 with a comment, or two frustrated messages in a row.
- The supervisor investigates — reading the conversation, the tool runs and errors, the assistant's instructions and memory, and your feedback.
- You get a diagnosis — a short plain-language explanation, in your language, with up to four proposed fixes in a card in the chat.
- You decide — apply or dismiss each fix. Applied fixes can be undone at any time.
The fixes it proposes
- Correcting or adding to the assistant's memory of your site.
- Updating your profile preferences — tone, detail, how you like to work.
- A guidance note the assistant follows for the rest of the conversation.
- A line for your team's style guide — tone, style, things to avoid — on plans that include one.
Team style guide changes need an admin: when a non-admin applies one, it goes to the team admins for approval and is listed under “Suggested by the supervisor” on the Style guide page.
When it finds a fault in the platform
Sometimes the problem is not your setup — it is ours. When the supervisor finds a fault in the AWT platform itself, it files a report, and medium and high-severity reports email the AWT team directly. Fixes from these reports have already shipped: new pages now start as drafts by default, and research tools now say honestly when they are unavailable instead of guessing.
Why teams rely on it
- Explains, not just apologises — a plain-language account of what actually went wrong.
- Fixes you control — every change is a proposal you apply, and Undo genuinely restores what was there before.
- Admins stay in charge — team-wide changes wait for an admin's approval.
- Sensible limits — at most one automatic review per conversation per hour, and one review at a time.
Availability
The AI supervisor is included on every plan, and reviews never use your credits — AWT pays for them.
Frequently asked questions
Does calling a supervisor cost credits?
No. Supervisor reviews are paid for by AWT on every plan — asking for a second opinion never eats into your credit pool.
Will it change things without asking?
No. Every fix is a proposal you apply or dismiss, and applied fixes can be undone. The supervisor never changes your website, content or code itself.
Is a human reading my conversations?
No. The supervisor is an AI, and reviews start only on the triggers above — it does not monitor conversations continuously. When it finds a fault in the platform itself, a report goes to the AWT team so it can be fixed for everyone; your individual chats are not read by staff as part of this feature.
When does a review start on its own?
After a failed run, repeated tool errors, a thumbs-down with a comment, or when you have sounded frustrated two messages in a row — limited to one automatic review per conversation per hour.
