Agentic Commerce and the Transformation of Marketing

Sustained quarter-over-quarter growth in real transaction volume (not just traffic) on an AI surface, the adoption of a universal checkout approach that solves the merchant opt-in barrier, and live agentic moves by the card networks in Türkiye would be signals that could make us more assertive; conversely, a second major "checkout retreat," a regulatory tightening driven by agent bias or fraud, or AI referral remaining limited for independent merchants over a long period could require us to be more cautious.

A Revolution in Discovery, Hesitation at Checkout

In online shopping, the buyer is no longer always a human. AI agents that search for, compare, select, and increasingly purchase products on the user's behalf — what Bessemer calls "the delegated buyer" — are spreading fast. And this is no longer merely a forecast; we can already see it in the data. According to Adobe's U.S. retail figures, traffic to retail sites from generative-AI tools rose roughly sevenfold year over year during the 2025 holiday season; moreover, these visitors converted at a higher rate and generated more revenue per visit than traditional sources did. In its own measurement, Salesforce says AI and agents influenced about a fifth of all retail sales during the 2025 holidays — some $262 billion in revenue (a vendor-sourced figure, to be read with caution).

But the full picture is more measured. In a study by IBM and NRF covering more than 18,000 consumers across 23 countries, roughly half of consumers use AI in some part of the purchase journey; yet this still sits mostly at the research-and-comparison stage — users continue to manage the final decision and the moment of purchase largely themselves. So the shift is fast, but for now mostly at the "assistant" level; it is not a full handover of the decision to the agent. That is precisely this article's thesis: agentic commerce is most likely a genuine platform shift, but still at an early stage; and for an investor, value is more likely to accrue not where the noise is loudest (protocols and payment rails) but where it is quietest (data, trust, and transaction execution).

Strong Discovery, Weak Conversion

The sector's most critical yet least-discussed reality is this: the discovery side is ready, the transaction side is not. In September 2025, OpenAI launched direct purchasing inside ChatGPT ("Instant Checkout"); but in March 2026 it pulled the feature and reverted to a model that redirects the user to the merchant's own site. The reason was instructive: only a handful of merchants had gone live, the system had not even set up state sales-tax infrastructure, and people were researching but not buying within the chat. Today ChatGPT is effectively a "referral engine" — creating value from traffic rather than from transactions. This is not the first time: in 2016, Facebook Messenger bots, and later Google Assistant's shopping features, were pulled back in much the same way. The crypto side tells the same story; despite a broad ecosystem valuation, Coinbase's x402 protocol was processing an extremely small — and largely test — daily transaction volume.

This "agentic readiness gap" — discovery being ahead while payment authorization and integration lag behind — is critical for an investor: even as traffic metrics explode, completed transaction volume is still small. The fact that market-size estimates swing across a range of up to roughly twenty-five-fold — from $1.7 trillion to $3–5 trillion for 2030, depending on methodology — is a reminder that these figures indicate a direction but do not carry certainty. In short, the early but real signals of a platform shift should not be confused with the evidence of a completed transformation.

Commoditizing Infrastructure, Fragmenting Standards

A standards race is under way at the infrastructure layer: OpenAI and Stripe's ACP; Google's AP2 and — together with Shopify, unveiled at NRF in January 2026 — its UCP; Anthropic's MCP, now donated to a Linux Foundation body and reaching more than ten thousand servers; Coinbase's x402; and the card networks' solutions (Visa Intelligent Commerce/TAP, Mastercard Agent Pay). Most position themselves as open standards, and this has two consequences. First, these openly available standards tend over time to lose their power to differentiate and to set prices — that is, they become commoditized and cease to generate value; the assistant layer the user sees is likewise concentrating in a few large model companies. For this reason, a venture built solely on top of a single protocol may have limited long-term defensibility. Second, the real risk is fragmentation: because every protocol requires the merchant's own opt-in, millions of small stores remain invisible to agents. And that very invisibility is a business opportunity in its own right.

As Marketing Inverts, Power Shifts to Clean Data

When discovery is mediated by a handful of agents, the marketing game inverts. AEO/GEO takes the place of SEO: the goal is no longer to be found by a human, but to be selected and recommended by a machine. The product feed stops being a catalog and turns into a contract; the agent chooses the product whose hard data (price, stock, delivery speed, return policy) is complete and machine-readable — often without ever sending the user to the site. As far back as 2024, Gartner forecast that traditional search volume would fall by roughly a quarter by 2026 (to be read as a direction). Bessemer sums up the shift strikingly: the buyer now often arrives better informed than the seller, impulse buying erodes, loyalty shifts from the brand to the agent, and the winners are not those with the best advertising but those with the cleanest data. From now on, brands will optimize two doors at once: one for agents, and one for the humans who still want to walk in.

Here there is an optimistic "leveling" thesis (associated with a16z): agents democratize information, let the consumer bypass platform "taxes," and make the clean-data small player visible — power shifts to the buyer. But a serious investor should also hold the counter-thesis. First, distribution inequity: for most independent merchants, AI referral volume is still microscopic, and a single model update can wipe out the gains. Second, platform bias: generative-AI surfaces are not neutral traffic distributors — their incentive is to keep the user inside the chat (the zero-click pattern and self-preferencing). Third, agent bias: academic studies (for example, ACES from Columbia) show that agents are not rational comparison engines; they carry a strong position bias — choosing items in the top half of the list by roughly 77% to 23% — and tend to pick the first acceptable offer. In short, "the agent decides" does not automatically mean "the best product wins."

Where Value Accrues: An Investor's Layer-by-Layer Analysis

Let us sharpen the investor's view a little. The commoditizing layers — open protocols and the AI surface concentrated in a few labs — are relatively less attractive from an investment standpoint; often it is more appropriate to treat them as mere infrastructure. The crowded but necessary layer is the payment-and-identity layer, which attracts the most funding but is dominated by the card networks together with Stripe/PayPal. The real whitespace, instead, appears in three places: checkout execution and the layer that connects the product feed to checkout — providing the universal merchant reach the opt-in protocols do not cover; agent-visibility (AEO/GEO) tools; and "know your agent" (KYA) and fraud-prevention infrastructure. Durable competitive advantages, meanwhile, can be expected to come mostly from a proprietary preference/data graph that travels with the user, from depth of merchant integration and distribution, from trust/fraud network effects, and from accumulated agent-readiness data.

The real risk for early-stage investors, though, lies hidden in Bessemer's own warning: the most valuable pieces of the stack may be captured by incumbents with strong incentives and existing distribution — payment networks, large e-commerce platforms, model providers. For this reason, our preference may be to stay close to areas the big players cannot easily copy: application-layer tools that deliver a fast return and differentiate through the data they hold. By contrast, it may be wiser to approach with greater caution businesses built solely on a single protocol or a single payment rail (crypto-only or card-only), thin intermediary layers ("wrappers") that a large platform could turn into a feature tomorrow, and agent-to-agent micropayment ideas for which real demand has not yet emerged.

Türkiye: Ready Infrastructure, Empty Shelf

So where does Türkiye stand in this picture? A striking asymmetry stands out: the payment infrastructure appears largely ready, but agentic products and dedicated local investment in this area are not yet at the same level of maturity. The card and instant-payment infrastructure is strong — according to BKM data, card payments rose about 49% year over year, and online reached nearly a third of card spending; the national scheme TROY is growing with state backing and supports tokenization; the central bank's (CBRT) FAST system and the digital-lira sandbox already cover programmable and machine-to-machine (M2M) payments — infrastructure highly relevant to agentic commerce. By contrast, Trendyol and Hepsiburada today mostly offer AI-based shopping assistants and merchant tools; as far as we know, not yet a fully autonomous agentic checkout.

For us, the most important structural point is this: Türkiye's local payment environment — an order in which Apple Pay, Google Pay, and PayPal are effectively absent and domestic wallets dominate — could create a barrier to entry for the U.S.-centric ACP/UCP players, which in turn could mean an opening for a local one. On the capital side, too, there appears to be a relative gap. In publicly available sources, investment news about recently founded, Türkiye-based ventures focused solely on agentic commerce is not common. This may point — especially for early-stage and application-layer-focused investors — to an opportunity area that is not yet sufficiently saturated and is worth evaluating.

Conclusion: Boğaziçi Ventures' Stance

In our view, this looks like a genuine platform shift; but we are still largely in the discovery phase, not the transaction phase — the buyer is increasingly a machine, yet the machine cannot yet reliably reach the checkout. For this reason, our current inclination is to stand, roughly, "ahead of the infrastructure, behind the hype." The areas closest to our application-layer thesis and of most immediate interest to us may be AEO/GEO and product-feed-optimization tools, especially those localized for Turkish marketplaces (Trendyol, Hepsiburada); after that — and confined to teams that can demonstrate real data and integration depth — checkout execution and KYA/fraud prevention could follow. We would also read Türkiye's local payment constraint less as an obstacle than as a factor that could turn into an advantage for a local player. Sustained quarter-over-quarter growth in real transaction volume (not just traffic) on an AI surface, the adoption of a universal checkout approach that solves the merchant opt-in barrier, and live agentic moves by the card networks in Türkiye would be signals that could make us more assertive; conversely, a second major "checkout retreat," a regulatory tightening driven by agent bias or fraud, or AI referral remaining limited for independent merchants over a long period could require us to be more cautious.