Chad Rubin sold Skubana in 2021 and is now betting on Profasee, which uses AI agents to execute decisions rather than recommend them. The bet is credible mainly because he runs his own eight-figure Amazon brand through the product, which is the part worth copying regardless of your channel mix.
Most of the people selling you AI agents in 2026 cannot tell you what happens when two of their agents disagree. Chad Rubin can, because he has watched it happen inside his own P&L.
In March 2026, one of the best people on Chad Rubin’s PPC team resigned. His name was Marko. Chad’s response was not to post the role. He told Marko he was going to build an AI agent modeled on him, name it after him, and keep the institutional knowledge inside the company. By Chad’s own account, Marko was both honored and slightly unnerved.
That story is a better window into what is happening in commerce operations right now than any product page. The thing being automated was not a task. It was a person’s accumulated judgment about which bids to move and when, the kind of knowledge that walks out the door with two weeks’ notice and takes six months to rebuild. Every operator reading this has lost that knowledge at least once.
Chad has been on the show three times in eight years, going all the way back to episode 13 in 2018. Across those visits, I watched the pitch move from multichannel inventory to dynamic pricing to what he now calls AI employees. That is why this piece exists. Eight years of watching one operator keep changing his answer tells you more than any single product page ever will.
The reason to pay attention is that the thing being replaced is not software, it is you. For most of the last decade, you were the intelligence layer sitting between disconnected tools: reading the inventory report, remembering it while looking at the ad dashboard, then deciding what to do about pricing. Profasee puts the average customer’s replaced spend at $7,800 per month across tools and agency fees, which is a real number but not the one of interest. The interesting number is the hours you spend being the glue.
This lands at a time when the category is nearly unusable. Everyone has AI in their bio. Chad is blunt about it, and his frustration is the most useful part of the conversation: the space is full of people who watched a few videos, wrapped a chatbot around a dashboard, and started selling courses. His test for separating the real from the theatre is simple enough to use on a sales call. Does the system make and execute a decision, or does it show you data you still have to decipher and act on yourself? Almost everything currently marketed as an agent fails that test.
For a Shopify operator, the relevance is not the Amazon product. It is that this pattern is arriving on your side of the business next, and the vendors will use identical language. The same test applies whether the pitch is about your marketplace pricing or your storefront. If you want the broader map of where this is heading across the ecosystem, our guide to agentic commerce for Shopify merchants covers the protocol layer that sits underneath all of it.
Chad Rubin is an Amazon seller who kept building the software he needed and then sold it. He got laid off from Wall Street during the 2008 crisis, took his family’s vacuum business online from a warehouse in Harlem, and grew it into Think Crucial, an eight figure Amazon brand he still runs today. He is a top 250 Amazon seller and co-author of the book Cheaper, Easier, Direct.
Along the way, he co-founded Skubana, the order and inventory management platform acquired by 3PL Central in 2021 and now operating as Extensiv, and co-founded Prosper Show, one of the largest Amazon seller conferences in the world. Profasee began as a dynamic pricing platform, which is what we covered in his 2025 appearance on Amazon pricing strategy. Profasee reports more than $82 million in profit unlocked for customers across that pricing work.
The detail that matters more than any of the exits is this one: he runs his own money through the product every day. He described himself as the support ticket rather than someone waiting on one. That is the difference between a founder who has a hypothesis about what merchants need and a founder who finds out on Tuesday morning that his own inventory forecast was wrong. It is also the honest reason to give the product a longer look than the category average deserves, and it has nothing to do with the marketing.
The decision was to stop selling pricing and start selling coordination, and Chad frames it as an industry shift rather than a company strategy. His argument is that operators are moving from a software stack to an agent stack, where agents replace tools and, in some cases, the agencies and headcount doing the work around those tools. Pricing was the wedge. The larger problem is that inventory never talks to pricing, pricing never talks to PPC, and none of it talks to the P&L.
What he rejected is as instructive as what he chose. Plenty of vendors responded to the same moment by shipping an integration to a chat assistant. Chad’s position is that an integration is the wrong shape, because it makes the operator do the assembly. The bet, instead, is to ingest the data with conventional software, then let purpose-built agents read it all. His analogy is a steak knife against a butter knife: someone who is not an engineer should be able to click a button and hire the capability.
The team is four live agents with two more announced. Claudia coordinates and delegates like a chief operating officer. Oracle owns margin-aware pricing. Marko runs PPC, which Chad calls the hardest role because the money moves fastest. Bruno handles demand planning and inventory. The coordination is the product: Bruno flags a stockout risk, Marko pulls ad spend on that ASIN, and nobody had to notice. Profasee’s published case studies include PF Harris adding roughly $215,000 in annualized profit across 15 SKUs, which the company reports as a 24 times return. Those are vendor-published figures, and they are the right ones to interrogate on a call rather than accept from a landing page.
The most important limitation is that the coordination story is further along in the pitch than in the product. Chad says plainly that the agents read the same ingested data today, but that agents holding their own stand-up meetings and cross-pollinating each other’s decisions is phase two, not shipped. He also says there is no playbook, that much of the platform is not yet figured out, and that the honest comparison is to a car with driver assistance rather than a robotaxi. Someone still sets strategy on day one and stays behind the wheel.
The fit gate is real, and it excludes most readers of this site. Ultra is built for Amazon brands at $50,000 per month or more with 20 or more SKUs, because below that the velocity is too thin for the agents to find signal. It is Amazon only today. Chad says multichannel and Shopify are the direction, exactly as Skubana went, but direction is not a roadmap date and should not be bought as one. If dynamic pricing on Amazon is not already a lever you are pulling manually, this is not your next purchase.
Price deserves a flag. Profasee’s own pages currently disagree: the Ultra overview and the Oracle pricing page both list a starting price of $698 per month, while the application page still describes $299 per month for the platform plus your first agent. That is roughly a doubling in five months against what was published earlier this year. Early access pricing moving is normal. Confirm the current number on the fit call, in writing, before you model any of this. In the interest of disclosure, the Profasee link in this piece is a tracked partner link, and it changes nothing about the assessment above.
Spend thirty minutes this week listing every tool and retainer in your operations stack, then draw a line between any two that actually share data, and count the lines. Most operators between $500K and $2M find they have eight or nine boxes and almost no lines, which is the real finding. I watched this pattern repeatedly during my years inside the Shopify ecosystem: merchants at that stage add tools faster than they add the process to connect them, and premature complexity, not underinvestment, is what stalls them.
If you are Amazon-heavy and above the fit line, the next step is to run something in read-only mode alongside your existing setup. Chad’s framing is that you do not have to fire your agency, you just have to check their work, and a week of watching the decisions an agent would have made is cheaper than a migration. Start in approval mode, widen autonomy only as the work earns it, and treat the agents as very well read interns rather than staff. You decide how much rope they get based on how much trust they have earned.
If you are Shopify only, the takeaway is the test, not the tool. When the agent pitch arrives on your side of the business, and it will, ask the vendor what happens when two of their agents want opposite things, then ask who arbitrates and on what metric. If the answer is a metric rather than the business outcome, or if there is no answer at all, you are looking at automation wearing an agent costume. And before you buy any of it, get your cost of goods clean, because every one of these systems is only as margin-aware as the margin data you feed it.
An AI employee is an agent that makes and executes decisions within your business, while an AI tool surfaces information you still have to interpret and act on yourself. The practical test is whether anything changes in your account without you clicking it. A repricing dashboard that recommends a price is a tool. A system that evaluates margin, inventory position, competitor stock levels and rank, then moves the price within your guardrails and logs why, is closer to an employee. Most products currently marketed as agents in 2026 fail this test, which is why the phrase has lost most of its signal value.
Profasee Ultra is Amazon-only today. The agents operate on Amazon Seller Central and advertising data, and there is no Shopify, Walmart or TikTok Shop coverage in the current product. Chad Rubin has said multichannel is the direction, following the same path Skubana took from Amazon into multichannel, but no date has been published. If you run a Shopify store with meaningful Amazon revenue, Profasee would cover the Amazon side only, and your Shopify operations stay on whatever you use now. Treat stated direction as intent rather than a commitment when you are budgeting.
Profasee sets the floor at $50,000 per month in Amazon revenue with 20 or more active SKUs. The reasoning is signal density rather than ability to pay: below that volume there are not enough transactions for the agents to learn demand elasticity, and a competent operator working manually will match or beat them. Profasee also states it is not a fit for sellers under $30,000 per month and reviews every application before granting access. If you are below the line, the honest answer is to fix your cost of goods tracking and your pricing discipline first, because both are prerequisites anyway.
Profasee’s published figures currently conflict, so confirm the number directly before budgeting. The Ultra overview and the Oracle pricing page both list a starting price of $698 per month for the platform plus your first agent, while the application page still describes $299 per month for the same configuration. Profasee says most customers run two to three agents, and puts the average replaced spend at $7,800 per month across tools and agency retainers. Early access pricing moves, sometimes significantly, so get the current figure in writing on the fit call rather than from a landing page.
Ask the vendor what happens when two of their agents want opposite things, and who arbitrates. A genuine multi-agent system has an answer, usually an arbitration layer weighing the business outcome rather than a single metric, because the conflict is structural: a pricing agent may want to raise the price for margin at the same moment an ads agent wants to lower it because cost per click is climbing. A chatbot on a dashboard has no answer because it has no second agent. Follow up by asking what the system does without you, and what it logs afterward.
