Insights · Commerce · 9 min read
When your customer sends an AI agent to do the shopping
Agent-mediated buying is arriving fast. For most retailers it is a discoverability problem. For direct selling it lands directly on attribution — the link the whole channel is paid through.
The next structural shift in commerce is delegated shopping: a customer sets an intent and some guardrails, and an AI agent compares options, decides and completes the purchase on their behalf. The adoption curve is steep. Nearly half of consumers are expected to use AI agents in their brand interactions by the end of 2026, and around two-thirds of shoppers aged 25–44 say they would hand repetitive purchases to an agent outright.
The infrastructure is arriving to meet it. The major AI and payments platforms have shipped agent-checkout protocols — standard ways for an agent to browse a catalog, verify a price and complete a purchase without a human clicking anything. Retail is reorganizing accordingly: as Gap's CTO put it, brands now have to show up not just in search engines but in answer engines, because the agent's shortlist is the new shelf. Most of the resulting advice concerns structured product data, machine-readable availability and promise accuracy, and it is sound.
Direct selling has all of that work to do, plus a problem nobody else has.
The attribution problem
The channel runs on a specific promise: the person who introduced the customer earns from the customer's orders, for as long as the relationship lasts. Everything — commissions, rank qualification, retention economics, the entire logic of building a customer base — sits on the durable link between a customer and their distributor.
Now walk the agentic purchase path. The customer's agent is optimizing for the customer: best price, fastest fulfillment, cleanest checkout. It arrives at the corporate storefront directly, or through whichever channel the answer engine ranked. Nothing in that path knows or carries who introduced the customer. The referral link, the personalized storefront, the promo code — the mechanisms attribution leans on today are all session ceremonies, and a session ceremony is precisely the kind of step an agent routes around. The agent does not click the distributor's link. It does not browse the party. It executes the intent.
The result, at scale, is repeat-order revenue quietly migrating from attributed to unattributed. That is not a technical footnote. Every unattributed order is a distributor's earnings, and the field notices earnings with a speed and thoroughness no monitoring system matches.
Agentic commerce is a discoverability question for most brands. For direct selling it is a compensation question.
Subscriptions feel it first
The first place this lands is recurring revenue. Subscription and autoship programs are the closest thing the channel has to agent behavior already — a standing instruction to repurchase — and they are exactly what consumers say they want to delegate: the boring, repetitive purchases. When a customer's agent takes over replenishment, it will shop each cycle. Best price, available stock, fastest delivery, this week. A subscription that was attributed once at enrollment becomes a sequence of independent purchases, each one re-decided by software with no memory of the relationship that started it.
Channels that depend on subscription continuity — which, after a decade of moving the field toward recurring customer revenue, is most of the industry — have the most exposure and the most reason to move early.
The work, in order
First, make attribution durable rather than session-based. If the customer–distributor link lives in a cookie, a referral URL or a checkout code, it dies the moment an agent shops on the customer's behalf. It has to live in the customer record itself — asserted at enrollment, confirmed by the customer, carried through every channel and every purchase path including the ones no human clicks through. Attribution by identity, not by journey.
Second, treat the agent protocols as a sales channel and prepare it deliberately: structured catalog data, accurate availability, pricing an agent can verify, and promises that hold. Agents select for reliability and remember failure. In an agent-mediated market, promise accuracy stops being a service metric and becomes a selection criterion — the brands whose data is trustworthy get shortlisted, and the ones whose data is aspirational quietly disappear from consideration.
Third, arm your own side. The same delegation that threatens attribution is an enormous gift to the field. A distributor's agent — one that watches their customer base, spots the lapse before it happens, drafts the outreach, handles the follow-up — multiplies exactly the part-time hours the channel has always struggled to make productive. The channels that thrive will be the ones where the field got agents too, not just the customers.
Why act early
Only about one consumer in ten says they would let an agent act with no oversight, so this is an adoption curve, not a cliff — there is time to do the work properly. But attribution is architecture, and architecture moves slowly. Retrofitting identity-based attribution after agent traffic arrives means re-litigating every ambiguous order with the field in real time, which is the most expensive possible way to discover the problem.
The companies that will handle agent-mediated orders gracefully in two years are the ones deciding now where the customer–distributor link actually lives. It is a quiet piece of work. It is also, for this channel specifically, the difference between agentic commerce as a threat and agentic commerce as the best productivity story the field has ever been told.