Yes, for configuration. In Affiliate Tabs you can ask the assistant to create a group, add a form field or invite a manager. It drafts the change, shows the exact values, and nothing happens until you confirm. It never approves or pays commissions, and credentials are created on their own pages.
Setting up a partner program is mostly typing the same kinds of things into forms: a name for the group, a commission percentage, a few application questions, an invitation to a colleague. None of it is hard. It is fiddly, and it is spread over a dozen screens.
An assistant that takes plain-language instructions can shorten that. Whether you should let it depends almost entirely on what it is allowed to do on its own.
What can you actually ask for?
In Affiliate Tabs, the assistant answers questions and makes configuration changes. Questions look like "How much did this partner earn last month?" or "Which applications are waiting?" Answers come back with the report or guide section they came from, so you can check them.
Changes look like the ones you would otherwise click through: "Create a group called Resellers with a 20% offer," "Add a required phone field to the application form," "Turn off the Black Friday trigger," "Invite sam@agency.com as a partner manager," "Point our webhook at the new URL." Behind the scenes, the assistant searches the product's list of operations, reads what each needs, looks up ids, and prepares the call.
For a first-time setup this is the useful part. You describe the program in a few sentences, and it builds the pieces in order instead of making you find them.
Who actually presses the button?
You do. This is the design choice that makes the rest acceptable. Reads, such as lists and reports, run immediately. Anything that changes data becomes a proposal: the assistant shows what it will do with the exact values, and you press Confirm or Dismiss. Nothing is changed until you confirm, and a proposal can be decided only once. The assistant does not say a change happened before you have confirmed it.
Deletions are marked in red, and the assistant has to say it intends to delete before it may draft one. Sensitive actions, such as publishing terms, deleting a trigger or erasing a person's data, send you through a password check when you confirm.
Read the first few proposals line by line. If one says a 20% offer and you meant 10%, you see it before anything is saved.
What does it never do?
Some things stay on their own pages on purpose.
- Approving or paying commissions, charging invoices and withdrawing balances.
- Choosing a plan, accepting agreements and submitting tax forms.
- Creating API keys, OAuth clients, tracking tokens and webhook secrets, so a secret never appears in a conversation.
Configuration can be undone by changing it back. Money, signatures and secrets cannot be undone so easily.
Does it have more access than you do?
No. It works with your role. If your role cannot change commission triggers, it will not offer to, and a confirmed change runs as you, with the same checks, audit log entry and notifications as the page that normally makes it. If your role changed after the proposal was drafted, confirming fails.
This matters for the record. When someone asks later who changed a rate, the answer is a named person who pressed Confirm, which is what you want. What a partner program audit trail should record goes through why.
What about Claude, ChatGPT or Cursor?
If your team already uses one of those, you can connect it over MCP. Connected tools can read live data under the key's permissions and can propose changes, but never make them. Their proposals wait under Assistant, then Pending changes, where anyone on the team can confirm or dismiss them under the same rules as a proposal made in chat.
That suits people who live in another tool all day: have it draft the setup for a new group, then review the proposal in the app.
When should you not use it?
Do the first commission design yourself, with a pencil. An assistant can enter any structure you describe; it cannot tell you whether 30% for 12 months suits your margins. Decide the economics first, using something like how to set commission rates partners respect, then use chat to enter them.
Also, be wary of vague instructions. "Set up a good affiliate program" produces something generic. "Create a group called Agencies, 15% of every payment, 12 months, applications require a website and a short description of how you will promote us" produces something you can check.
Messages count toward a monthly allowance by plan: 2,000 on Launch, 10,000 on Growth and 50,000 on Enterprise.
Common questions
- Can an AI assistant approve or pay commissions?
- Not in Affiliate Tabs. Approving or paying commissions, charging invoices, withdrawing, choosing a plan and accepting agreements are never run from chat. Those stay on their own pages, where a person looks at the numbers and decides.
- Does the assistant change things without asking?
- No. Reads run at once, but any change becomes a proposal that shows the exact values with Confirm and Dismiss buttons. Nothing changes until you press Confirm, and deletions are marked in red. Sensitive changes also ask for your password.
- Can it see more than I am allowed to see?
- No. The assistant works under your own role. It offers only what your permissions allow, and a confirmed change runs as you, with the same checks and audit log entry as the normal page. If your role changed since a proposal, confirming fails.
- Can I connect my own AI tool?
- Yes, over MCP. Tools such as Claude, ChatGPT or Cursor can read data under the key's permissions and propose changes, but never make them. Their proposals wait under Pending changes, where anyone on the team can confirm or dismiss them.