Every serious cross-border seller I know runs the same single-point-of-failure operation. There is one person — the founder, a veteran account manager, a sourcing lead — who carries the actual truth of what is happening right now in their head. That person goes home, and the intelligence goes dark. They leave, and the knowledge walks out the door. So when I read that Tandem, an AI-native office leasing brokerage from Brendan Suh, is building an operation where every agent works with an “agentic brain” behind them — a system watching the market constantly for “what came available this morning, what a similar unit actually leased for last month, which landlords are flexible on term, what’s about to come back online” — I didn’t see a real-estate product. I saw a mirror of the disease that kills cross-border operations, and a working model for the cure.
The disease: market knowledge is one employee’s memory
In commercial real estate, the old model works like this: a landlord wants to lease a floor, a tenant needs slightly less, and the broker who knows both sides wins. The broker’s value is not the listing — public listings are everywhere. It is the private overlay: which landlord blinked on a previous deal, which tenant quietly accepted a concession package, which building manager is painful to work with, what the actual lease rate was against the advertised ask. That overlay lives in one person’s head. The firms that dominate the industry — CBRE and its peers — are essentially armies of those people, each holding a patch of human memory. The firm profits from human recall.
Now watch what happens in a cross-border operation. The equivalent of the broker is your top account manager, or the founder themselves. That person knows that the buy box only survives if you undercut the aggressive seller just enough to win it back. They know which freight route clears customs without inspection and which one triggers a hold. They know which keyword every new competitor attacks first, and they know that the supplier quietly bumps the unit price unless the order clears a certain quantity. None of that lives in the P&L. None of it exists in Amazon Seller Central as a report. It is an undocumented, unpaid database with one fragile copy. When that person takes a vacation, the operation goes partially blind. When they accept a competitor’s offer, the company loses its intelligence in a single afternoon.
Tandem’s insight is that this is not a people problem; it is an architecture problem. “Knowing the market stops being one person’s job, so it doesn’t stop when they go home, and it isn’t limited to the buildings they’ve personally walked.” Substitute “buildings” with “ASINs, suppliers, and ad accounts” and you have the most important operational sentence in cross-border commerce this month. The agentic brain is not a search engine. It is a colleague that never sleeps, has read everything, and is not limited to the buildings or categories one person has personally walked. That is not a feature. It is a new organizational form.
What Tandem changes: push intelligence instead of pull
The established data platform in commercial real estate is CoStar. It is monumentally useful, and it is query-based. A broker opens CoStar, types in a submarket, pulls comparables, interprets them, and makes a judgment. The system retrieves; the human interprets. Tandem inverts that posture. The agentic brain is “watching the market constantly” — it is not waiting for a query. When a new unit becomes available this morning, the brain notices. When a comparable lease closes, the brain logs the actual figure. When a landlord starts bending on term, the brain registers the signal. The output is not a dashboard that needs reading. The output is an agent who walks into the conversation with the market already loaded.
That distinction — push versus pull — is the entire ballgame, and it is exactly where most e-commerce tooling still fails. Helium 10 will tell you, when you ask, that a ranking moved or a keyword is rising. Keepa will show you, when you open the graph, that a competitor’s price has been eroding for weeks. These are retrieval tools with good search bars; they wait for you to come to them. Tandem’s architecture is an assertion engine: it does not wait to be asked, and it does not return a table of results. It returns an informed state of mind.
Why Amazon sellers should care more than Shopify ones
I want to be fair to the DTC crowd. If you run a Shopify store, you should still find Tandem conceptually interesting, but your urgency is lower. On Shopify you own the customer, the email list, and the event-level analytics. Tools like Klaviyo give you cohort truth the same day. The data asymmetry between you and your own market is small; you can query your own database. The agentic brain is a luxury.
Amazon sellers live in the opposite world. Amazon controls the buy box, the search results, the BSR number, and the truth about your own customer. You are renting a storefront inside a black box. The intelligence that actually matters — what a competitor’s listing really converts at, whether a category is tipping into a race to the bottom, which sellers are gaming review velocity — is exactly what Amazon does not expose. That structural condition is what makes an agentic brain valuable. When the marketplace holds the data, the operator who can synthesize publicly observable signals into private knowledge wins the market. This is the same condition commercial real estate has lived with for a century: the listing is public, but the truth is layered, dispersed, and forever out of date the moment it sits in a static system. So read Tandem as a blueprint if you sell on Amazon, and as a thought experiment if you sell on Shopify.
Build your own agentic brain this quarter
Here is where I stop reviewing Tandem and start handing out assignments. The launch page gives a clean decomposition of what an agentic brain should watch, and it maps almost perfectly onto cross-border trade.
First, fresh supply — “what came available this morning.” In your world, this means new listings in your category, new sellers entering your search results, new suppliers posting in your sourcing lane, new categories opening on the marketplaces you sell. Most operators notice these changes on a lag of days or weeks, usually when a competitor’s sales rank spikes. The brain should be noticing on day one and flagging the threat with a recommendation attached.
Second, actuals — “what a similar unit actually leased for last month.” The key word is “actually.” Not the list price. Not the ask. The realized price after coupons, lightning deals, buy-box wars, and affiliate stacks. Keepa gives you the raw price history, but you still need the synthesis layer that tells you: this competitor is not being “flexible on price,” they are clearing a doomed inventory position.
Third, flexibility signals — “which landlords are flexible on term.” In cross-border terms, this is which suppliers are suddenly quoting shorter lead times, which freight forwarders are offering off-peak rates, which marketplace categories are loosening their entry barriers. These are the negotiations nobody is holding until the moment they matter, and by then it is usually too late.
Fourth, forward supply — “what’s about to come back online.” The most dangerous competitor is not the one who is live today. It is the one whose inventory is in transit and will come back online in a few weeks, often with a new price point and a new review strategy. If you are not tracking your competitors’ stockout patterns and restock signals, you are not tracking the market; you are tracking the aftermath of it.
The common error is buying more tools and still doing manual synthesis: open Helium 10, check the graph, refresh Seller Central, repeat. That is not an agentic brain; that is a person deciding to act like a CPU. Tandem’s operating principle is the opposite: delegate the watching to a system that never sleeps, and reserve the human for the judgment. The output of a good agentic brain is not an alert. It is a standing, always-updated market brief — three or four sentences that say what changed, why it matters, and what decision is waiting.
The weekly brief, not the real-time alert
A real-time alert is the enemy of actionable intelligence. A system that pings you on every price change teaches you to ignore it. What compounds is a synthesized brief, delivered daily in a volatile category or weekly everywhere else, with no charts and no raw data dumps. The brain says: “Your buy box is exposed next week because competitor X’s container is about to land.” The human says: “Then I pull my coupon budget forward.” If an AI product gives you charts, it is giving you homework. If it gives you a position, it is giving you leverage. Tandem’s implicit philosophy — the agent that walks in already knowing the market — is the same philosophy that should govern your internal reporting. Stop reporting data. Report decisions.
Where the math breaks
I have spent this essay praising the architecture, so now I need to spend time on the failure modes, because AI-infused brokerage — like AI-infused marketplace automation — looks beautiful in a launch description and gets ugly at the edge cases.
The comp-data trap
Tandem’s power claim rests on knowing “what a similar unit actually leased for last month.” In office leasing, comparable transactions are messy but ultimately knowable: the market is structured, leases are recorded, and a brain can learn the pattern. Cross-border e-commerce actuals are catastrophically polluted. The realized price on Amazon is not the buy-box price. It is the price after coupon stacks, lightning deals, exclusive discounts, review-velocity manipulation, and the dark patterns every sophisticated seller has seen and every platform denies. If you feed an agentic brain polluted comps, it will confidently narrate nonsense. It will tell you a competitor is “flexible on price” when in reality they are dumping stale inventory or using promotions to inflate conversion velocity before a rank push. A brain is only as honest as its inputs. The cross-border version of Tandem needs a validation layer the real-estate version may not need: a human who audits the comps before the brain’s conclusion changes a pricing decision.
The unit economics of watching
The second failure mode is cost. A genuinely always-on market brain is not one subscription. It is a data stack, a monitoring layer, an AI layer, and the human time required to validate outputs. For a seven-figure account, that stack pays for itself the first time it intercepts a supply disruption or a keyword cliff before the damage compounds. For a five-figure shop, the same stack is overhead with a sophisticated facade. The math breaks on unit economics, not on the elegance of the brain. Tandem is building for the commercial real estate equivalent of a high-ticket seller: a single office lease is a large transaction with fat fees, so a brain that prevents one bad deal covers its cost for a year. If you are a smaller operator, do not buy the entire stack. Borrow the one piece that moves your P&L: the weekly brief, assembled from the tools you already pay for.
What my judgment says is missing
Let me do the part that launch-day enthusiasm won’t do: say what Tandem is missing.
First, the moat is still the relationship, not the information. The agentic brain can watch the market, but office leasing closes on trust, on the side letter, on the landlord’s unstated willingness to bend, on the broker’s memory of a building owner who hates brokers. The brain cannot read the pause, the non-answer, the body language across the table. Cross-border trade is identical. The brain can tell you a factory has capacity, but it cannot build the relationship that gets your PO pulled ahead of a bigger customer’s in peak season. Tandem’s real contribution may be that it frees the human agent to spend more time on the trust layer and less on the retrieval layer. It does not remove the human. The launch page is honest about this — the brain works “behind” the agent, not in place of the agent — and that honesty is the most credible thing about it.
Second, the launch is thin. No pricing, no availability details, no named customers, no case studies. That is normal for an early launch, but it means a skeptical operator should treat Tandem as a directional signal, not a procurement decision. Our industry has a pattern where every tool with a chat interface calls itself “agentic.” Most are alert systems with a wrapper. Tandem earns more credit than that — the framing, the comp-data discipline, and the humility of the “behind every agent” positioning all signal a team that understands the problem — but it has not yet shown transaction outcomes. I would watch it before I would buy it.
Third, and this is the one I most want cross-border sellers to internalize: no agentic brain solves the imagination problem. The brain is only as good as the questions it is built to care about. If you have never thought to ask “what’s about to come back online,” no brain will ask for you. The tool amplifies a good operator. It does not manufacture one. The operator who copies Tandem’s architecture without first owning a point of view about their market will simply automate their own mediocrity.
What I’d watch / test next
Here is the experiment I’d start this week: a two-week test that costs nothing. Pick your most competitive category. Set up three signals: Keepa price and rank alerts for your five closest competitors, Helium 10 keyword tracking for your top conversion terms, and a daily scrape of your competitors’ listing changes. Then do your usual manual review each morning, and separately ask an AI assistant to write a three-sentence market brief from the same signals. Compare the two. What did the brief catch that you missed — and what did it confidently get wrong? That test tells you whether your operation is ready for an agentic brain, and it costs less than one day of buy-box losses. Meanwhile, I’ll be watching Tandem for the numbers launch pages omit: pricing, case studies, and whether the brain stays quietly “behind” the agent or starts moving to the front of the deal. And I’ll be watching Brendan Suh for whether Tandem starts publishing the market’s actuals instead of just describing them. If it does, this stops being a real-estate story. It becomes the template every cross-border operator should steal.
