The biggest mistake I'm seeing sellers make with AI isn't using the wrong prompts.
It's not giving it a big enough job to do.
AI jobs over prompts. That's the whole idea this week.
Most sellers treat their AI like a Google replacement. Ask a question, get an answer, move on.
Here's what happens when you give it a job instead.
Amazon suspended my account. So I put 2 frontier models in a room and had them argue about how to fix it.
They told me exactly how to reply to the canned responses. What to say to get it escalated to a supervisor. What to show them so they'd actually take action. They even told me to request a call.
I got it unlocked.
That's not a prompt. That's a job.
So here's my question for you this week: what are you underestimating about your AI that you could be doing right now?
Hit reply and tell me your #1 challenge with using AI in your ecom brand. I read every one.
"I didn't have the expectation that it could"
One of the sellers in our OpenClaw AI Ecom Accelerator has had a top-selling ASIN down since the 4th of July.
Amazon changed the compliance rules on him. For 8 years he submitted the paperwork himself. Now the testing center has to submit it. Five-week lead time on the lab, and FDA approval on top of that.
He's been going in circles with an appeal bot that keeps telling him to do the thing he already did. Nobody has read the file.
Here's the part that stopped me.
He has 2 AI agents running on a server right now. He never gave the problem to either one.
This week we had Andrew Erickson on as our special guest for the alumni call. Andrew's a multiple 7-figure seller who exited his brand, and he won the AI workflow hackathon at the summit last week.
He asked straight out: could your agent just own this?
His answer: "I didn't have the expectation that it could."
That sentence is the whole problem. Nobody on that call needed a better model. They needed to ask for more.
Here's how Andrew hands something over. This is close to word for word:
This is a major blocker
Identify the issue and understand it
Figure out who to talk to and how to expedite it
Tell me what to expect and by when
Follow up until it's done
Document how we solve this next time
Status update every Tuesday
That's not a prompt. That's a job description. It's what he'd hand a business partner.
Andrew has let go of VAs and replaced them with agents. I ran the entire 7 Figure Seller Summit last week with nobody in the EA seat.
But Andrew said something on that call I want you to hear, because it's the honest version: agents are better than a VA. They're not as good as a six-figure employee. Not yet.
Now the part where I eat my own cooking.
I burned through my $200 a month plan 3 times in 7 to 10 days. Hit zero. Locked out of my own stack until Saturday.
Andrew runs his whole operation for under $50 a month. Then he audited it down further. He doesn't buy one big subscription and run everything on it. He picks a different model for each job — a free one for regular chat, a cheap one for email triage, a mid-tier one for his memory summaries, and the expensive one only when the decision actually matters.
And one more thing that cuts against almost everything being sold right now. Anything with math in it should be code, not a model. Andrew threw his sales and inventory data at a chatbot and asked what to order. He ran it 20 times and got 20 different answers.
TAKEAWAY: 3 things to change this week. 1. Give your agent a job, not a task — if you'd hand it to an intern, you're wasting it. 2. Route by job, not by subscription. Most of what your agent does all day is worth a penny. 3. If it has math in it, make it code. Twenty runs should give you 1 answer, not 20.
Your keywords and your Amazon SEO don't matter anymore for Amazon Alexa. No matter how much PPC spend you throw at it.
A new study covered by Marketplace Pulse looked at 1,963 non-branded queries and 12,810 Alexa recommendations captured in May and June.
Here's what they found:
63.9% of Alexa's picks came from outside the organic top 10 Amazon search results
40.9% never appeared in the search results page at all
Only 14.3% were running a sponsored listing
For context, sellers paid $68.62 billion in Amazon ads in 2025.
Now the caveat, and it matters. It's only 1 study, 1 vendor, and 1 US account, over a 2 month window. It doesn't mean your ads stopped working.
But it does mean that ads don't buy you AI recommendations the way they buy you a search placement.
TAKEAWAY: This is why Alexa and Rufus optimization matters right now. Go ask Alexa "what's the best [your category]" and see if you come up. Costs you 2 minutes. If you're not there, that's your Q4 homework.
Close your Q4 purchase orders now
Heads up guys. As the calendar flipped to August, did you know right now is the time you should be closing your purchase orders with your suppliers for Q4?
Any later and you may miss the boat. Literally.
So if you haven't already, check your inventory, check your orders, and anticipate Black Friday and Cyber Monday sales so you don't miss out.
By the way — AI hackathon winner Andrew Erickson is offering a special 50% discount on his AI inventory forecasting tool. Enter code GARY50 to get it. There's a 30-day free trial too.
I watched him run it live on our call this week. A seller asked it a real Black Friday question — 20% promo bump, split shipments, 20% buffer. The answer came back: forget the 20% bump, you're already low on stock, and splitting your shipments is going to cost you more, not less.
That's the kind of answer you want in the first week of August. Not the first week of October.
TAKEAWAY: The Q4 purchase order is the decision right now. The promo strategy comes later. If you don't know your inventory position 3 months out, that's the thing to fix this week.
AI + Amazon Seller Meetup — Shanghai, Thursday Aug 13
Hey everyone. I'm in Shanghai next week and I'm hosting a small get-together for Amazon sellers and entrepreneurs who want to use AI in their business.
Forbes interviewed me last week on how business owners are using AI well. I'll share a short talk on what's working right now, and the mistakes I keep seeing sellers make with it.
Then we eat, drink, and talk shop.
📅 Thursday, August 13 🕕 6:00–8:00 PM 📍 Abbey Road Bar, 3 Taojiang Lu (桃江路3号), near Hengshan Lu, Xuhui
Who it's for: Amazon sellers, brand owners, sourcing agents, and anyone building with AI.
Free to attend. Just pay for your own drinks.
Keeping the room small so the conversations are actually good.
RSVP: hit reply with "SHANGHAI" and I'll send you the details.
Here's the Forbes piece if you want to read it: https://www.forbes.com/sites/elainepofeldt/2026/07/29/e-commerce-sellers-turn-to-ai-to-counter-challenges-to-profitability/
Not in Shanghai? Save these dates instead. October 13–14, Tokyo — the next in-person 7 Figure Seller Summit Live Mastermind. More details coming soon.
THE VITAL FEW
THE 20%
Everyone I know is paying $14 a month to talk to their AI instead of typing.
There's a free one. It's called FluidVoice, it's open source, and the speech never leaves your machine — it transcribes on-device, so you're not uploading your business conversations to anybody's server.
Mac only for now. Windows and iOS are on a waitlist.
QUICK POLL
What's actually stopping you from handing more work to an AI agent?
That's the 20% for this week.
See you next Friday.
Gary
P.S. — Hit reply and tell me your #1 challenge with using AI in your ecom brand right now. I'm building the next round of training around your answers, so be specific.
P.P.S. — Tokyo. 7 Figure Seller Summit LIVE Mastermind, October 13–14. Save the date, more details coming soon. Hit reply with "TOKYO" if you want me to save you a seat and early bird pricing.