Every AI lab is shipping an agent that promises to watch your business while you sleep. Before you let one watch your Amazon store, check what it can actually see.
There are three kinds of AI assistant an Amazon seller can use, and they’re easy to confuse because they often sit inside the same apps. Jeff Su lays out the three tiers clearly in his video on personal agents. Here is what each one means for a seller or an agency, and which one I’d pick.
Tier 1: Chatbots
ChatGPT, Claude or Gemini in a chat window. You ask, it answers, the conversation ends.
Chatbots are good at one-off work: rewrite a bullet point, explain a policy notice, draft an email to a supplier, summarize a report you paste in.
Two things hold them back for a seller. They start from zero every time, so you re-explain your catalog, your margins and your goals in every chat. And they can’t see your account, so every question about your own numbers starts with you downloading a report and pasting it in.
Tier 2: DIY agents
Claude Code, Claude Cowork, Codex and similar tools. An agent doesn’t just answer, it does the work: runs a sequence of steps, uses tools, writes files.
“DIY” is the important part. You set it up yourself. Your rules and memory live in plain text files the agent reads every time: which store is yours, how you calculate profit, what a good ACoS looks like for your products, what it may and may not change. You connect it to your data with connectors, including your Amazon data. You decide what runs on a schedule.
The cost is setup time. The payoff is control: the agent works the way your business works, the files are yours, and you can move them to a different AI if a better one comes along.
Tier 3: Personal agents
Meta’s Muse, ChatGPT’s dots, xAI’s Grok Bot and Anthropic’s Claude Duet. Think of a personal agent as the ready-made version of a DIY agent. Underneath it uses the same idea, memory and rules it keeps about you, but you set it up by talking to it instead of writing files.
Two things set this tier apart. Each one runs on its own computer in the cloud, so it keeps working when your laptop is closed. And it’s proactive: you give it a goal and it checks in on its own, messaging you when something needs you.
For an Amazon seller, that sounds like exactly what you want. An assistant that watches your store every morning and only speaks up when something’s wrong.
The gap: what can it see?
Here’s the catch. An agent that watches your store is only as good as the data it can see, and at the time of writing, none of the personal agents comes with a built-in Amazon seller connection.
I tried two of them on my own store. Both understood the job immediately. Both were ready to put it on a schedule. And both had to ask me where to get my Amazon data before they could check a single number.
You have three ways to close that gap:
- Let the agent log into Seller Central in its browser. It works, but it’s fragile. Pages change, sessions expire, and you’re handing an agent your Seller Central login.
- Add a connector. Personal agents can add outside tools as custom connectors, and a connector built for Amazon data gives the agent clean, structured numbers instead of screen scraping.
- Use a DIY agent, where connecting your own data is the normal way to work.
That’s why we built the Seller Labs MCP Server. It connects an AI agent to your Amazon sales, fees, inventory, advertising and profit data, so the agent answers from your numbers instead of asking you for them. It works with both Claude and ChatGPT. At the time of writing, we’re working on connecting it to personal agents like Muse and Grok Bot as well.
Which seller jobs need always-on watching?
Personal agents make a lot of being “always on.” For most Amazon work, a check once a morning is enough. What matters more is whether the agent can see the data the job needs.
| Job | How often it needs checking | What the agent needs to see |
|---|---|---|
| Profit by SKU (fees, cost of goods, ad spend together) | Daily or weekly | Sales, fees, COGS and ad spend joined at the SKU level |
| FBA stock cover and restock timing | Daily | Inventory and sales velocity |
| Ad spend against budget | Daily, more often for big budgets | Campaign spend and budgets |
| Fee changes and reimbursements | Weekly | Fee and settlement data |
| Losing the featured offer on a top seller | Several times a day | Live pricing and offer data |
| A suppressed or blocked listing | Several times a day | Listing status |
The last two reward watching more often. The rest move over days, and a scheduled daily run catches them. Either way, the schedule is the easy part. The data is the hard part.
One more thing a watchlist can’t do: catch the problem you didn’t think to put on it. Buy Box, stock and ad budget are the alarms every seller knows. Profit leaks hide where those alarms don’t look, which is why an agent that can see the whole picture is worth more than one that watches a short list closely.
A note for agencies
Agencies should start closer to the DIY end.
Personal agents are built for one person’s life: one inbox, one calendar, one set of memories. An agency runs many clients, and each client’s sales and ad data needs a hard wall around it. A DIY setup lets you build those walls deliberately: which data, which tools, which client, written down in files you control.
It’s also worth reading what the platforms say about connecting outside tools. When Muse offers to add a custom connector, Meta shows this warning first:
Meta doesn’t review custom connectors or how they use your information. The agent may take unexpected actions. Grant access with caution and review their privacy policies.
That’s a fair warning for one person. For an agency holding a client’s data, it’s a question the client will ask, and you’ll want a better answer than “the agent set it up.”
Which should you pick?
Use a chatbot for one-off questions and writing. Don’t expect it to know your account.
Start with a personal agent if you’ve never used an agent before. Setup is a conversation, it remembers you, and it messages you first. It’s the easiest way to learn what an agent is good for. Expect three tradeoffs: you’ll need to connect your Amazon data yourself, you work inside the system it gives you, and your setup is tied to that company’s product.
Graduate to a DIY agent when you hit one of these walls:
- You want it working on your Seller Central data every day, not on what you paste in.
- You run more than one account or more than one marketplace.
- You manage clients.
- You want to decide exactly what it can see and change.
The tiers are moving toward each other. The labs are adding proactive features to their DIY tools, and personal agents are adding connectors. Whichever one you use, the question to ask isn’t how often it watches.
It’s what it can see.
