The Last Transaction With a Human in It
What happens when neither party is human anymore?
D2C Is Dead. Welcome to A2A.
Every letter in this naming convention is a human leaving the room.
D2C cut out the retailer. B2C widened the tent. B2B served the enterprise. B2A — the shift I wrote about in January — meant your customer sent an AI agent instead of showing up themselves.
Now: A2A. Agent-to-agent commerce. Your brand’s AI negotiating with your customer’s AI. No human on either end — not in the browsing, not in the negotiation, not in the checkout.
The sale closes and neither party is a person. Sit with that.
Last week I wrote about what brand looks like when it becomes a spreadsheet. This is final essay in the Agentic Shift series is about what happens on the other side of the table — when the buyer isn’t there either.
“Agents negotiating with agents” sounds abstract until you walk through it, and it’s something I find both awe-inspiring and terrifying.
What A2A Actually Looks Like
Someone — let’s call her Maya — wants to find a birthday gift for her friend. She opens her phone and tells her personal assistant: “Find me something good under $50. She likes ceramics and sustainable brands.” Maya doesn’t open a browser. She doesn’t open Instagram or Pinterest. She doesn’t visit your website. She doesn’t see your ad. She gives intent to her agent and goes back to her coffee. (I picture her at a kitchen counter, still in yesterday’s socks, not thinking about your brand at all.)
Her agent knows her preferences, her purchase history, her friend’s publicly available wishlist. It uses an MCP — the protocol that standardizes how agents connect to external data — to query merchant catalogs. Not typing keywords into a search bar. Issuing a structured request: category, price ceiling, brand criteria, available inventory. Dozens of catalogs, queried simultaneously, in the time it takes Maya to pour a second cup.
The reason this works now and didn’t three years ago is a protocol problem that finally got solved. Before MCP, every agent needed a custom integration for every data source — the complexity scaled as N-times-M and made the whole thing slow, expensive, and impractical. MCP collapsed it to M+N. Web MCP means doesn’t even need to take screenshots to read a page anymore. One integration, every agent ecosystem. That’s the unlock.
Your brand’s agent — the one Klaviyo calls a Customer Agent — receives Maya’s request. It doesn’t return a generic product page. It pulls Maya’s friend’s profile data from your SSOT (if she’s a past customer), surfaces the ceramics that fit her aesthetic, and packages an offer: a $44 ceramic planter, gift-wrapped, with expedited shipping. Maybe a 10% first-purchase discount because the query came from a new household. Maybe a bundle. It negotiates.
Maya’s agent evaluates the response against other offers it received in parallel. It factors in reviews, return policy, delivery window, brand trust signals. It ranks. It selects.
Then the plumbing kicks in. ACP — the Agentic Commerce Protocol, developed by OpenAI and Stripe as an open standard — manages the cart and checkout sequence.
The money moves through AP2, Google’s Agent Payments Protocol, built with PayPal, Mastercard, and 60+ partners. AP2’s key innovation is the mandate: a cryptographically signed digital contract proving Maya authorized her agent to make this purchase on her behalf. Tamper-proof. Verifiable. No fraud ambiguity. No disputed charge. No human hand on the keyboard at any point.
The planter ships. Maya gets a notification. Her friend gets a gift she actually wanted.
Your brand made a sale without a single impression, click, or email open.
What “Selling” Becomes
Here’s where it gets uncomfortable for people who built careers on persuasion. (People like me.)
The whole infrastructure of marketing was built on the assumption that there was a person at the end of it who needed to be *moved*. Copywriters wrote headlines because humans respond to language. Art directors chose images because humans respond to visual cues. CRM teams built journeys because humans respond to timing and relevance.
In A2A, there isn’t a person at the end. Maya’s agent isn’t moved by your brand story. It doesn’t feel the pull of aspirational lifestyle photography. It doesn’t respond to urgency copy (”Only 3 left!”). It evaluates:
Does this product match the stated parameters?
Is the brand’s data clean and structured?
Does the trust signal — reviews, return policy, fulfillment reliability — meet threshold?
It’s not persuasion. It’s evaluation. And the difference changes everything about where you put your effort.
The creative brief doesn’t die. It becomes the agent’s personality — the training data that determines how your brand agent answers questions, what it prioritizes, how it negotiates. Here’s what that actually looks like: where a creative brief used to say “Our voice is warm, knowledgeable, and slightly irreverent — think the friend who knows wine but won’t be weird about it,” a model specification says “When price queries fall within 15% of ceiling, lead with product-match confidence score, then social-proof signal, then offer a curated bundle. Tone: consultative. Never open with discount.” The brief stops being a document for a copywriter and becomes a specification for a model. Same brand DNA, different container.
Customer relationships become agent relationships. Loyalty isn’t emotional anymore — it’s contextual. Does your agent remember that this customer prefers matte finishes and hates bubble wrap? (The bubble wrap thing is real. I have opinions about unnecessary packaging and I want my future AI to know them.) Does it know she shops for her mother-in-law every December and her price ceiling goes up 30% for those orders? Does it negotiate well, or does it return generic offers that Maya’s agent deprioritizes immediately?
Clean data and a well-trained agent aren’t nice-to-haves.
They are the product.
The marketer’s job shifts from “create campaigns that persuade humans” to “train agents that negotiate competently on behalf of the brand.” That’s closer to product management than copywriting. It’s closer to data architecture than brand strategy. It requires caring deeply about things that used to live in the engineering org: schema consistency, API reliability, protocol compliance. I think the people who are slightly annoyed right now — the ones thinking *I didn’t get into marketing to care about schema* — are the ones who most need to hear this.
This is the part where I’m supposed to reassure you that creativity still matters. It does. But the creativity that matters in A2A is the creativity of structuring information so that an agent can use it — not the creativity of making a human feel something. Those are genuinely different disciplines. The people who understand both will be worth a lot.
The Through-Line
D2C was about getting closer to the customer. We spent a decade celebrating that proximity — the customer relationship as competitive moat, the first-party data as the thing that made you defensible.
A2A is the inverse. The customer isn’t closer.
The customer isn’t even there.
What’s there instead is a system that acts on her behalf, makes decisions according to her stated and inferred preferences, and executes without her involvement. The relationship isn’t gone — her preferences, her history, her trust in your brand all still feed into the outcome. But the moment of transaction has been delegated entirely.
For people who built careers on the belief that great marketing is fundamentally about human connection — this is a genuinely disorienting thing to sit with. I don’t think it means human connection stops mattering. I think it means the place where that connection gets established moves earlier in the funnel, further from the transaction, deeper into brand infrastructure. Changes how brands show up and add value away from the purchase cycle.
You earn the relationship before the agent ever shows up.
The agent executes it.
Which means the question isn’t whether you believe in A2A. The question is whether your brand data is clean enough, your agent is trained well enough, and your infrastructure is structured enough for someone else’s AI to choose you when no human is watching.
That’s a harder question than “is our creative good?” And most marketing teams aren’t asking it yet. I’d bet dollars to donuts they will be by next quarter.
p.s. There’s a lot of snake oil salesmen in the AI space right now. I’ve been working on a few things and will be putting my money where my mouth is with a few releases coming to an open source repo near you. Stay tuned for more
.
