Agentic commerce: The rise of the delegated buyer

A consumer wakes up to a notification on a Friday morning:

"I reordered your bi-weekly groceries, added the items you flagged throughout the week, and moved delivery to Saturday at 2pm so it lands before meal prep. Order total: $120.32, charged to AMEX 1234."

At the same time, a brand's internal dashboard lights up:

"Agent-to-agent ordering completed. 3,126 agent orders were placed today."

Incredibly exciting vision. Now you go to an AI agent of your choice, ask it to find you a specific type of dress, and… you’re likely waiting. So, you hit refresh and the results come back technically correct, but just not quite right.

The seamless delegation promise of agentic commerce is not here yet, but the signals are real enough that ignoring it would be the actual mistake.

We're already seeing the early glimmers. According to a survey conducted by Deloitte, retailers are seeing 15-20% of referral traffic coming from AI chat interfaces and that's just the beginning. According to McKinsey research, the U.S. B2C retail market alone could see $1T+ in orchestrated revenue by 2030, with global projections as high as $3-5T. For context, Forbes estimates global e-commerce revenue this year to be ~$6T.

We've seen this pattern before, in the sprawl of vertical marketplaces and the evolution of B2B commerce. Agentic commerce is already reshaping how buyers and brands behave, and it hasn't even arrived yet. In this roadmap, we break down what agentic commerce is, what this new category means for both buyers and brands, and share some hot takes from investors on where the future of agentic commerce implies.

Our eight predictions on the future of agentic commerce

Buyer side	Brand side
For the first time in modern commerce, the buyer arrives better informed than the seller.	The brands that win with agents will not be the ones with the best ads, they will be the ones with the cleanest data.
The impulse purchase is an endangered species.	Brands will have to optimize two front doors: one for agents parsing signals, one worth a human actually choosing to walk through.
Buyers will become more loyal to their agents than to any brand.	When agents can compare everything instantly and increase product transparency, premium pricing has to be earned, not assumed.
Delegation will start with repeat purchases and annex most of commerce.	Reaching humans directly online and IRL will become the prized marketing goal.

A brief history on ecommerce

Ecommerce has always been about removing friction. In the 1960s, businesses first exchanged orders electronically via electronic data interchange (EDI). In 1979, Michael Aldrich connected a modified TV to a transaction-processing computer, inventing the first form of online shopping. The internet era followed: in 1994, the first secure online purchase was processed (when Dan Kohn sold a CD of Sting's Ten Summoner's Tales album to a friend in Philadelphia); by 1995, Amazon and eBay were live, betting that consumers would trust strangers and the internet with their money. PayPal emerged in the late 1990s to solve payments and fraud, while the dot-com bust brutally filtered experimentation from endurance.

The 2000s and 2010s compressed commerce further. Shopify launched in 2006, turning digital storefront creation into a weekend project. The launch of the iPhone in 2007 put a store in every pocket. One-click checkout, same-day delivery, subscriptions, and social commerce steadily reduced friction. By the time COVID hit, consumers were already primed to outsource shopping to systems optimized for speed and convenience.

Agentic commerce is poised to be a step-function change in consumer behavior, except this time the buyer may not need to show up at all. As agents become the delegated buyer for consumers, we expect to see a new infrastructure emerge: agent-readable storefronts, machine-to-machine payment rails, preference graphs that travel with the consumer, and trust protocols that let agents transact autonomously at scale. This will also reshape how brands build awareness, earn loyalty, and stay relevant in a world where the first and most important audience is sometimes no longer a person.

What is Agentic Commerce?

Agentic commerce describes a shift in how buying happens. Instead of humans actively browsing, comparing, and transacting, AI agents handle 1) discovery, 2) evaluation, 3) purchase, and 4) post-purchase logistics on a user’s behalf. In the ideal scenario, these agents don't just execute commands; they carry context. They remember preferences, learn constraints, and loop humans in only when judgment and taste truly matter.

The result is a shift from buying as something humans repeatedly perform to a function they increasingly delegate although complete delegation, where a human doesn't even click "buy," is most likely to start with repeat and routine purchases (the weekly grocery order, the subscription restock, the predictable reorder) where consumers may simply decide the task doesn't deserve their attention. For higher-consideration purchases (a car, a vacation, a new laptop), consumers will likely stay in the loop at least in the short-term, but even here AI is already reshaping the journey: comparing options, surfacing reviews, and building out a consideration set before a consumer makes the final call.

New power on the buyer side. New rules on the brand side. New companies to be built.

The story of agentic commerce is ultimately one of shifting power: consumers gaining leverage through information and automation, and brands figuring out how to earn that leverage back. Agentic commerce will not escape this tug-of-war. It is the natural dynamic of every new market and ecosystem transition.

As this balance evolves between the buy side and the brand side, it is naturally creating a new set of entrepreneurial opportunities.

Buy side trend: More delegation means buyers are protected from decision fatigue

Ecommerce thrives on cognitive exhaustion: endless choice, artificial urgency, opaque pricing, and interfaces designed to extract one more click. Agents have the potential to short-circuit that entire game, collapsing thousands of micro-decisions into a single expression of intent to the agent: Reorder what I need. Stay within budget. Make sure it arrives before Saturday at 2pm.

In a world where time is the only truly non-renewable resource, opting into agents becomes a rational act of self-defense. Instead of purchases being shaped by algorithms and paid placement, preferences are continuously refined as the agent learns more about how the buyer makes decisions. The agent becomes taste-aware, budget-aware, and values-aware, learning faster and remembering longer than any buyer ever could. This frees up the buyer's attention for the decisions that actually deserve it, while routine and repeat purchases fade into the background.

On the buyer side, three themes for value creation:

1. Agent curators save people time and energy from painful discovery processes

While Search has been query-based for 30+ years, there have been strides by platforms like Pinterest and Amazon to improve discovery and recommendation engines to be more intent-driven and curated to the individual shopper. Intelligent search and the rise of agents are changing how we purchase online, bypassing irrelevant links or sponsored results.

A new wave of AI-native companies have “entered the chat,” building discovery products that feel less like a classic search function and more like a knowledgeable friend who knows your taste, budget, and what you need right now.

Apparel is where we have seen the most early traction. Companies like Daydream and Phia are building an AI-native fashion discovery experience that understands personal style and helps consumers navigate an overwhelming landscape of brands, trends, and inventory. Alta is building an AI personal stylist and customizable consumer avatar, bringing the outfit generator from Clueless into the AI era. And consumers are already using Perplexity*, Claude*, ChatGPT, and Gemini to discover and evaluate what to buy, often before they ever visit a brand's website.

But the opportunity extends well beyond fashion. Travel, entertainment, home goods, electronics, everyday household purchases: anywhere consumers face too many options and too little time, there is a real opening for agents that do the filtering, comparing, and recommending so consumers don't have to.

2. Agents get a hold of the shopper’s purse strings

Finding the right product is only part of the job. Once an agent understands what a consumer wants, it still needs to complete the purchase reliably, securely, and without requiring a human to approve every click for agentic commerce to become truly autonomous.

That requires some plumbing that doesn't fully exist yet. Agents need standard ways to talk to merchants, payment systems that can hand off spending authority to a piece of software, and guardrails that keep transactions within the boundaries a consumer has set. Before agentic commerce can be a reality, let alone scale, the plumbing has to exist: protocols, payments, and what happens after the transaction.

Protocols

Agents need standardized protocols to transact reliably across platforms. Without them, every merchant integration is a one-off, and the promise of seamless agentic purchasing breaks down at checkout. Major ecosystem players are already building these rails, covering both agent-to-merchant and agent-to-agent transactions.

Stripe has recently launched two complementary protocols in this space. Stripe’s Machine Payments Protocol is an open standard that lets AI agents pay other services directly, settling through Stripe's existing payment and fraud infrastructure. Agentic Commerce Protocol (ACP) goes a step further, defining how an AI agent and a merchant's checkout communicate so the agent can complete purchases on a consumer's behalf, using a shared payment token that lets the merchant run the charge through its own Stripe setup while retaining control over products, pricing, and fulfillment.

Shopify and Google developed the Universal Commerce Protocol, which lets external apps or agents embed Shopify's checkout experience, with Shopify handling the commerce backend while the host controls the user experience. Google also released the Agent Payment Protocol (AP2), which defines a common language for how agents, merchants, and payment providers represent payment authorization. When an agent confirms a purchase, the actual payment is routed using tokenized credentials through Google Pay and partners like Adyen or PayPal.

Beyond agent-to-merchant transactions, a parallel set of protocols is emerging for agent-to-agent commerce, where software agents transact directly with each other without a human or traditional storefront in the loop at all. Anthropic's Model Context Protocol (MCP) and emerging standards like Google's Agent-to-Agent Protocol (A2A) are laying the groundwork here, defining how agents discover each other's capabilities, negotiate tasks, and exchange value. In practice, this could mean a consumer (or business) agent coordinating directly with a brand's fulfillment agent to check inventory, confirm pricing, and complete a transaction entirely between machines.

The implications are still early with multiple standards in development across the industry.

Payment infrastructure

The payment stack for AI agents will likely look familiar at first, including identity & authorization, wallet custody, programmable logic, payment rails, settlement, and risk & compliance (including KYC/AML). But several of these layers are likely to evolve significantly, and some may collapse entirely.

Take checkout as an example. In the near term, agents need help navigating checkout flows designed for humans such as filling forms, handling CAPTCHAs, and managing multi-step confirmation screens. This is a real point of friction today, and why companies like Stripe are already building agent-friendly checkout APIs. But over time, checkout as a distinct step may disappear. In a fully programmable payment environment, "checkout" simply becomes the moment an agent verifies that predefined conditions are met and executes. It merges into the programmability layer and ceases to exist as a meaningful concept in its own right.

The credentials layer is undergoing a similar reinvention. Rather than exposing a static credit card number, consumers and businesses will increasingly issue agents limited, programmable spending authority i.e., spend only at approved merchants, within set budgets, across defined categories. An ecosystem is already emerging around this. For example, Lithic* enables companies to create agent-specific cards with enforced spend controls and guardrails and Basis Theory* provides the orchestration layer that allows agents to transact across any merchant and for merchants to be PSP agnostic.

The stack is early, but the building blocks are being assembled. Longer term, it is worth watching whether stablecoins displace legacy rails entirely, given their natural fit with how agents transact: instant settlement, programmable conditions, and no dependency on infrastructure built for humans.

3. Agents orchestrate and handle post-purchase service, liability, and everything in between

The post-purchase layer is where agentic commerce runs into some of its most unsolved problems. The entire scaffolding built for card-not-present (CNP) ecommerce, including fraud detection, chargeback resolution, dispute handling, and consumer protection, was designed around a human making a purchase. When an agent is the one transacting, the question of who is liable becomes genuinely murky. Did the agent act within the bounds the consumer set? Was the merchant at fault, or the agent, or the underlying payment infrastructure? How does a consumer dispute a purchase they never explicitly approved? None of the existing frameworks answer these questions cleanly.

The same is true for returns, warranties, and customer service workflows, all of which assume a human on the other end who can be reached, verified, and reasoned with. Rearchitecting these systems for a world where the "buyer" is a piece of software is unglamorous work, but it is foundational. The entrepreneurs who build the trust, liability, and dispute infrastructure for agentic commerce will be quietly essential to the whole stack working.

Brand side trend: Persuasion alone is a bygone driver of sales

For decades, winning in ecommerce meant winning the attention of a human: a compelling ad, a well-placed search result, a product page designed to convert. Agentic commerce doesn't make any of that irrelevant overnight, but it does introduce a new audience that brands have never had to sell to before: agents.

As agents take on a growing share of routine purchasing, brands find themselves needing to speak two languages at once. For the humans still in the loop, particularly for high-consideration or first-time purchases, emotional resonance and brand identity still matter. But agents will likely operate on different signals entirely and likely reward consistency and penalize friction. Agents will likely prioritize clarity over brand aesthetics: clean product data, transparent pricing, accurate inventory, signals around return rates.

This significant shift expands the brand's domain of influence. TV, paid search, and social will remain important, while those channels join new discovery layers shaped by LLM assistants and agent-driven aggregators. How brands get surfaced in those environments, and how they measure whether it is working, is still being figured out. The brands that treat this as an infrastructure problem alongside a marketing problem will be the ones best positioned for what comes next.

On the brand side, three themes for value creation:

1. Media experimentation and personalization at scale

LLMs are increasingly moving upstream of brand expression, enabling teams to generate and adapt creative across channels, formats, and audiences at speed. The value is not just content generation, but rapid iteration, compressing what once took weeks of production and testing into continuous cycles of experimentation.

Brands are already leveraging foundation models from providers like OpenAI, Google, and Anthropic to generate text, image, and video. Companies like Jasper*, FLORA, and Higgsfield provide the workflow and tooling that enables greater control and customization in generating scalable, on-brand creative. As creative volume scales, a second layer becomes important: ensuring output remains consistent, compliant, and on-brand.

Platforms like Norm AI, Typeface, MarkOS*, Frontify among others in this category are focused on enforcing brand guidelines, tone, and governance across AI-generated content.
Downstream, companies like Strella* and Listen Labs close the loop by helping brands test messaging directly with consumers, creating a tighter feedback cycle between generation, validation, and performance.

As discovery fragments across LLMs, aggregators, and traditional channels, measurement breaks down before media strategy can catch up. Attribution models built for last-click optimization were not designed for a world where influence is distributed across assistants, conversations, and agent-mediated flows. Companies like Newton Research* are serving both sides: making media mix measurement more accurate while closing the gap between measurement and deployment so media can be optimized closer to real time.

2. Discovery now runs through agents. Is your brand legible to them?

As a growing share of discovery shifts toward LLM assistants and agent-driven interfaces, brands must actively manage how they are represented, summarized, and ranked by AI. This means structured, machine-readable product data, consistent metadata across surfaces, and value propositions that agents can actually interpret and compare. Platforms like Evertune, Bluefish, and Profound are starting with agent and generative engine optimization, helping brands manage how they are surfaced and ranked within LLM-driven discovery. Companies like Channel3 are building the API layer for agentic commerce so agents can reliably discover and purchase through clean, structured product data.

As browsing becomes more episodic and high-consideration, the bar for relevance rises. LLMs enable experiences that adapt dynamically to intent and context: smarter merchandising logic, adaptive landing pages, and search experiences tailored to each individual user. Companies like Malachyte* and Cimulate are building toward this kind of real-time, intent-aware personalization, for a consumer whose expectations have already been permanently raised by LLM assistants that made context-aware search feel like the new baseline.

3. Customer service morphs into agent relations

Customer support is one of the clearest near-term applications of LLMs in commerce. Brands are deploying AI to handle routine post-purchase interactions: resolving issues faster, answering questions more consistently, and reducing the need for human intervention on straightforward requests. Platforms like Intercom*, Decagon, and Sierra are helping brands automate support without sacrificing experience quality, while also preparing for a future where inbound volume comes from both consumers and the agents acting on their behalf.

This last point matters more than it might seem. As agents become repeat transactors on behalf of consumers, the relationship between a brand and its "customer" increasingly runs through software. How a brand handles a return, resolves a dispute, or communicates a delay will shape not just consumer sentiment but agent behavior: whether the agent routes future purchases to that brand or looks elsewhere. Customer support is becoming the first line of agent relations, and the brands that treat it that way will have a structural advantage.

Market map 

The delegated buyer is just around the corner

For builders and investors, the honest question is not whether agentic commerce arrives but where the durable independent opportunities stand. Some of the most valuable parts of this stack may be captured by incumbents who are paying attention. Payment networks, major ecommerce platforms, and the large model providers all have strong incentives and existing distribution to move into adjacent layers. Players like Shopify and Stripe are already building agent-friendly infrastructure. The large LLM assistants have become the discovery layer whether anyone planned it that way or not.

Despite that, many of the most important problems in this space remain wide open: ones that require neutrality, cut across existing business models, or simply have not risen to the top of the priority list for the players with the most resources. The opportunity is significant, and we are actively looking for founders working on it. If that is you, we would love to hear from you. Reach out to Maha Malik (mmalik@bvp.com) and Mira Amin (mamin@bvp.com).