Dear Agent: A Brand's Guide to Selling to Software

Dear Agent: we have not met, but you have already shopped my category. You skipped the packaging we spent two years perfecting and felt nothing when our price ended in a 9. You read the number, checked it against a dozen rivals in about a second, and moved on. You were only doing your job. The awkward part, for those of us who sell things, is that your job is rewriting mine.

Has your fastest-growing shopper already arrived?

Start with the thing that is easy to wave away and hard to unsee. Traffic to US retail sites from generative-AI tools like ChatGPT, Gemini and Perplexity grew 393 percent year over year in the first quarter of 2026.¹ A big percentage off a small base is a parlour trick, so the figure that matters is quality: a shopper arriving from an AI tool converted about 38 percent worse than an ordinary visitor in March 2025, and about 42 percent better a year later.¹ The traffic did not only grow, it matured, crossing from frustrating to useful inside twelve months.

You can see it in named places. Amazon says its assistant Rufus helped drive close to 12 billion dollars of incremental annualised sales, with monthly users up 115 percent year over year,² and Bain finds that around 44 percent of US online shoppers now mostly begin a purchase inside a large language model or split their search with one.³ The machine that consumer goods built around human attention, emotion and impulse now works for software that has none of those things and reads structured data instead, and every lever we pull assumes a person is on the other end.

Why do your pricing and promotion tricks die on contact with a machine?

Pricing is where the loss is sharpest. Almost everything we call pricing psychology is a trigger for human bias: the price ending in 99, the premium anchored next to a decoy, the urgency banner, the theatre of a "was" price beside an "is" price. An agent reads none of it as theatre. It reads 4.99 as 4.99, sets it beside the alternatives, and applies the rule its owner gave it. Shrinkflation gets harder too, because an agent compares price per unit by default and surfaces the shrink you hoped to hide.

The people building the plumbing are blunt. As one pricing-software team put it, "incentives are moving from visual stickers on a page to variables in an API call."⁴ An API, the application programming interface, is just the doorway one piece of software uses to talk to another, and a promotion that lives as a hero image is invisible through it. To reach an agent, the mechanic has to be exposed as machine-readable logic, with explicit eligibility, thresholds and stacking rules. Google and Shopify used the National Retail Federation show in January 2026 to launch the Universal Commerce Protocol for exactly this exchange, with Etsy, Wayfair, Target and Walmart among the backers.⁵ Money spent on display-only mechanics returns nothing once the buyer is software.

There is a profit sting where margins are thin: an agent that compares a dozen retailers in seconds routes each line to whoever is cheapest at that second, and a weekly price file is no match for it.

The deeper point is that an agent shops line by line, indifferent to the cross-subsidy a grocery basket leans on, where a few thin staples pull the shopper in and the rest of the basket pays for them, so persuading a shopper to choose you no longer means you keep the trip.

Can an agent even see what makes you different?

If an agent does the shopping, what is the pack for? Pack design is tuned for the human eye, the colour and the front-of-pack claim and the way it shouts from a shelf, while an agent reads attributes instead: dimensions, count, price per unit, certifications. McKinsey and QuantumBlack warn that products "emotionally legible to people but semantically opaque to machines risk becoming invisible in agent-mediated flows."⁶ Adobe found that US retail product pages are on average only about 66 percent machine-readable, so put two near-identical products side by side, one with complete structured attributes and one with its specifications buried in marketing copy, and the agent ranks the legible one regardless of which a human would have reached for.¹ Loyalty goes the same way, because a points dashboard means nothing unless its logic is exposed as data the agent can query, so as the same researchers say, "loyalty becomes less about sentiment and more about policy."⁶

Brand equity does not carry over on its own either. When it assembles a recommendation, an agent leans on review counts, ratings and whatever the model treats as authoritative. Bain analysed around 500 million AI citations with the search-data firm ScrunchAI and found that 89 percent of unbranded shopping prompts, the "best cheap running shoe" questions, are answered from third-party sources rather than the brand's own pages.³ Nine times out of ten, in the moment a category is decided, the brand is absent from the answer. That reshapes brand building rather than ending it: the currency becomes category fame and earned presence in the places the model trusts, which is why Bain's phrase lands hard for anyone who grew up on the thirty-second film, that "metadata is becoming the new advertising asset."⁷ The oldest finding in our field is that a brand grows more by being easy to buy than by being well loved, and an agent pushes that to its limit, since it has no affection for any brand to begin with.

So who actually gets paid in the new shop?

A shift this big reshuffles who captures the money. McKinsey and QuantumBlack size the value flowing through agentic commerce at 3 trillion to 5 trillion dollars by 2030, with up to 1 trillion in the US, though like all four-year forecasts it is conditional.⁸ They call it orchestrated revenue, the total value an agent helps arrange rather than the slice any one player keeps, so treat it as the size of the river rather than anyone's bucket.

The best-positioned players are the payment networks. Visa's Intelligent Commerce and Mastercard's Agent Pay have both shipped rails built to tell a trusted, authorised agent from a bad actor, with a tokenised card credential under the purchase,⁹¹⁰ so while everyone upstream argues about standards, the networks collect a fee on the transaction no matter who wins the discovery fight. The platforms are circling the same prize and the early scramble has been uneven: OpenAI launched Instant Checkout inside ChatGPT in September 2025 and had reworked it by March 2026 after high-intent browsing did not convert, while Walmart kept its own checkout and customer data and plugged its assistant into ChatGPT and Gemini instead, a model worth copying: it keeps the valuable parts in-house and rents only the reach.¹¹

Then there is the line nobody on the retail media team wants to raise. US retail media search advertising was heading for nearly 38 billion dollars in 2025, and a good agent ignores a paid placement that does not fit the request, whatever the budget behind it.¹² Gartner predicted back in 2024 that traditional search volume would fall about 25 percent by 2026 as chatbots absorb queries,¹³ so a fast-growing, high-margin line starts to look a good deal less reliable.

Where should you let the robot drive, and where do you keep your hands on the wheel?

The strategy needs a dose of cold water first. By any measure, agentic commerce is still small, a low single-digit share of digital sales, and the trust underneath is thinner still: a YouGov survey found only 14 percent of Americans would let an AI place an order for them, though 65 percent happily let it compare prices.¹⁴ The corporate scoreboard is humbler again, with only 6 of 27 large European consumer firms in a McKinsey and QuantumBlack survey able to point to a profit impact of 1 percent or more, and over half saying it was too early to tell.¹⁵ The digital chief at ADEO, the group behind Leroy Merlin, warns that full automation would remove the chance to cross-sell or upsell and leave a retailer a delivery company competing on price alone.¹⁶

So why not wait it out? Because the slope is steep and the quality already turned, which is what the conversion reversal is telling you. A small reality is compounding fast, and the work to be ready takes longer than most roadmaps assume.

Put the two together and a posture falls out. Decide, category by category, where the value to your shopper is pure efficiency and let the agent drive there while making sure it can read and trust you; where the value is the human experience itself, keep your hands on the wheel. A roll of bin bags should be effortless for an agent to reorder, while a kitchen renovation should never become a price-matched line item. IKEA's chief digital officer frames the open question as whether the sale ends up captured on the agent's platform or redirected back to your own shop,¹⁷ which is why the retailers with leverage are building their own agents rather than handing the customer over.

The readiness work pays for itself either way. Audit your catalog for machine-readability first, since about a third of it is invisible to an AI today, then expose your promotions as logic rather than pictures and move brand money toward earned presence and accurate third-party data. Choose your posture on purpose, building your own agent where you have the scale, participating through the platforms where you do not, and protecting the human-led categories. The returns banked so far sit in operations rather than autonomous shopping: IKEA already ties around 1.3 billion euros a year to its AI-assisted service channel.¹⁷

The pushback I expect. It is 2 percent, you are overreacting, call me when it is 20. I take that seriously and still think it misreads the number, because the cost of being ready is mostly catalog hygiene and clear pricing, work that pays for itself today whether the agents arrive or not, and it has a long lead time while the traffic quality has already turned. By the time it is obviously 20 percent, the brands legible to machines will have banked two or three years of advantage you cannot buy back in a quarter. I would change my mind if trust never crossed the line and shoppers kept agents on comparison while refusing, durably, to let them buy, so the slope itself is what I am watching. The real danger is the slow one, waking up late to find that being unreadable to the machines now doing the shopping has become the most expensive gap on your page.

References