Spend a little time searching for a travel backpack these days and the experience is both familiar and frustrating. A flood of listicles appears — “Top 13 Travel Backpacks for 2024” — many stuffed with affiliate links, recycled specs, and half-hearted reviews. It’s harder than it should be to find genuinely useful guidance, especially if you’re not a gear enthusiast. The web still feels like it's working for itself, not for you.
A lot of this traces back to the mechanics of online discovery. Ever since affiliate marketing took off, content has often been optimized less for accuracy or insight and more for search engine visibility and conversion rates. Product recommendations are shaped by who pays and who tracks — not always who adds value. Over time, this incentive structure has eroded trust. You can still find good advice online, but you have to hunt for it.
This economic layer, built around attribution (who gets credit for a purchase), wasn’t designed with today’s AI-rich landscape in mind. It’s fragile. “Last-click” attribution — where whoever touched the shopper last gets paid — means reward goes not to the source of real influence, but to whoever was lucky or aggressive enough to grab the customer at checkout. That might be a promo code plugin or a deal site — not the writer or friend who truly shaped the decision.
As AI agents begin playing a bigger role in consumer behavior, this structure gets even shakier.
For routine or considered purchases — a laptop, a bike, even laundry detergent — AI could be a powerful companion. A well-designed agent could compare specs, surface price history, check availability locally, and remind you which product actually fits your lifestyle. In theory, it could use your household inventory, past purchases, or loyalty program data to make smarter suggestions.
But we’re not quite there yet. Despite their promise, AI systems today still hallucinate. They struggle with long-tail product knowledge — niche camera lenses, less-known bag brands, specific SKUs — and real-time accuracy. Information can be outdated, wrong, or missing entirely. And our systems don’t yet integrate well with the fragmented, messy infrastructure of commerce: many websites lack structured data, even fewer are built with APIs that allow seamless transactions.
Impulse buys — that scented candle you didn’t think about until you saw it — don’t fit cleanly into an AI agent’s workflow. AI currently reacts better than it inspires. That spark of desire is still very human. But the middle ground — purchases that are neither snap decisions nor deeply involved — is where the opportunities are clearest.
Examples come to mind easily: a new side table, a travel pillow, a dish soap that won’t dry out your hands. Individually small decisions, but ones we overthink or second-guess. These are areas where AI could meaningfully reduce friction. Whether it’s tracking price drops (CamelCamelCamel-style), optimizing card rewards, or reconciling subtle feature tradeoffs, the digital assistant becomes a kind of buying co-pilot.
For AI to become a durable part of the shopping process, though, it has to do more than advise. It needs access. That means new infrastructure — commerce with legible pages, browsable inventories, permissioned APIs. Platforms like Rakuten and Honey have started navigating this, but they’re narrowly optimized for couponing, not thoughtful buying journeys. CamelCamelCamel, which simply tracks price history well, feels more aligned with the kind of trust and transparency that consumers crave. A good agent should feel more like that — useful, unintrusive, and not trying to sell you anything it's not confident about.
That’s why models like Costco resonate in these conversations. Their entire business is trust. They make money from membership, not markups. They don’t list everything. They curate. And they commit that nothing they sell is something they’d be embarrassed to offer. It’s a refreshingly direct relationship: you pay them to filter, fairly. In a world flooded with content noise and questionable endorsements, that kind of clarity stands out.
It may also be AI-resistant in the sense that an agent can’t easily improve the Costco model — it’s already reducing complexity and signaling quality. But we can imagine agents learning to recommend “Costco-like” equivalents outside of their ecosystem — things with fair pricing, high satisfaction, limited downside.
There’s a horizon here where new commerce-native and AI-native platforms emerge. They won’t look like traditional price comparison websites, and they won’t follow the junk-SEO playbooks of yesteryear. They’ll focus on helping humans decide, not just buy. They might use structured data from Shopify stores, UPC codes from manufacturers, and natural language input from shoppers. Their goal isn’t just transactions — it’s better decisions.
It’s tempting to imagine a full-pipeline agent — one that quietly tracks your intent, finds the best answer, and completes the order with the best card and shipping method already applied. But there are real challenges to get there: accurate attribution, real-time data integrity, merchant permissions, and the soft squishiness of human taste.
Even so, something is shifting. We’re seeing signs that the SEO-driven, surface-deep affiliate cycle has peaked. Google is losing informational traffic to LLMs. Consumers are getting savvier, sensing when they’re being tricked or talked down to. And the structure of the web — built for humans browsing — may need to flex to accommodate both humans and agents collaborating.
At Dellecod, we think a lot about what it means to build for this environment — one that respects the shopper, empowers their choices, and doesn’t hijack their intent. We don’t have all the answers yet. But we do believe the next chapter of commerce will be shaped less by who yells loudest and more by who earns trust.
And maybe — just maybe — that’s where technology can start to restore something we’ve lost: clarity.
A lot of this traces back to the mechanics of online discovery. Ever since affiliate marketing took off, content has often been optimized less for accuracy or insight and more for search engine visibility and conversion rates. Product recommendations are shaped by who pays and who tracks — not always who adds value. Over time, this incentive structure has eroded trust. You can still find good advice online, but you have to hunt for it.
This economic layer, built around attribution (who gets credit for a purchase), wasn’t designed with today’s AI-rich landscape in mind. It’s fragile. “Last-click” attribution — where whoever touched the shopper last gets paid — means reward goes not to the source of real influence, but to whoever was lucky or aggressive enough to grab the customer at checkout. That might be a promo code plugin or a deal site — not the writer or friend who truly shaped the decision.
As AI agents begin playing a bigger role in consumer behavior, this structure gets even shakier.
For routine or considered purchases — a laptop, a bike, even laundry detergent — AI could be a powerful companion. A well-designed agent could compare specs, surface price history, check availability locally, and remind you which product actually fits your lifestyle. In theory, it could use your household inventory, past purchases, or loyalty program data to make smarter suggestions.
But we’re not quite there yet. Despite their promise, AI systems today still hallucinate. They struggle with long-tail product knowledge — niche camera lenses, less-known bag brands, specific SKUs — and real-time accuracy. Information can be outdated, wrong, or missing entirely. And our systems don’t yet integrate well with the fragmented, messy infrastructure of commerce: many websites lack structured data, even fewer are built with APIs that allow seamless transactions.
Impulse buys — that scented candle you didn’t think about until you saw it — don’t fit cleanly into an AI agent’s workflow. AI currently reacts better than it inspires. That spark of desire is still very human. But the middle ground — purchases that are neither snap decisions nor deeply involved — is where the opportunities are clearest.
Examples come to mind easily: a new side table, a travel pillow, a dish soap that won’t dry out your hands. Individually small decisions, but ones we overthink or second-guess. These are areas where AI could meaningfully reduce friction. Whether it’s tracking price drops (CamelCamelCamel-style), optimizing card rewards, or reconciling subtle feature tradeoffs, the digital assistant becomes a kind of buying co-pilot.
For AI to become a durable part of the shopping process, though, it has to do more than advise. It needs access. That means new infrastructure — commerce with legible pages, browsable inventories, permissioned APIs. Platforms like Rakuten and Honey have started navigating this, but they’re narrowly optimized for couponing, not thoughtful buying journeys. CamelCamelCamel, which simply tracks price history well, feels more aligned with the kind of trust and transparency that consumers crave. A good agent should feel more like that — useful, unintrusive, and not trying to sell you anything it's not confident about.
That’s why models like Costco resonate in these conversations. Their entire business is trust. They make money from membership, not markups. They don’t list everything. They curate. And they commit that nothing they sell is something they’d be embarrassed to offer. It’s a refreshingly direct relationship: you pay them to filter, fairly. In a world flooded with content noise and questionable endorsements, that kind of clarity stands out.
It may also be AI-resistant in the sense that an agent can’t easily improve the Costco model — it’s already reducing complexity and signaling quality. But we can imagine agents learning to recommend “Costco-like” equivalents outside of their ecosystem — things with fair pricing, high satisfaction, limited downside.
There’s a horizon here where new commerce-native and AI-native platforms emerge. They won’t look like traditional price comparison websites, and they won’t follow the junk-SEO playbooks of yesteryear. They’ll focus on helping humans decide, not just buy. They might use structured data from Shopify stores, UPC codes from manufacturers, and natural language input from shoppers. Their goal isn’t just transactions — it’s better decisions.
It’s tempting to imagine a full-pipeline agent — one that quietly tracks your intent, finds the best answer, and completes the order with the best card and shipping method already applied. But there are real challenges to get there: accurate attribution, real-time data integrity, merchant permissions, and the soft squishiness of human taste.
Even so, something is shifting. We’re seeing signs that the SEO-driven, surface-deep affiliate cycle has peaked. Google is losing informational traffic to LLMs. Consumers are getting savvier, sensing when they’re being tricked or talked down to. And the structure of the web — built for humans browsing — may need to flex to accommodate both humans and agents collaborating.
At Dellecod, we think a lot about what it means to build for this environment — one that respects the shopper, empowers their choices, and doesn’t hijack their intent. We don’t have all the answers yet. But we do believe the next chapter of commerce will be shaped less by who yells loudest and more by who earns trust.
And maybe — just maybe — that’s where technology can start to restore something we’ve lost: clarity.