---
title: "When the brand stops being a destination: welcome to the agentic channel"
date: 2026-06-05T14:31:06.000Z
canonical: https://www.ekino.com/publications/when-the-brand-stops-being-a-destination-welcome-to-the-agentic-channel/
---

## Three places, then none at all

For thirty years, the relationship between a brand and its customer was defined by a place: first the shop, then the website, finally the mobile app. At every stage it was up to the customer to make the move — to **come** to the brand.

The next point of contact will not be a place. It will be an agent.

Three building blocks are falling into place at the same time, and it is their convergence that changes everything: the user’s **personal agent**, the **brand agent** that companies will have to field in response, and the **agentic channel** where those two agents talk to each other directly.

But there is a fourth block, less visible and yet decisive — the one the other three depend on. I will come back to it later: it is what separates an appealing demo from a system a company can actually deploy.

## 1. The personal agent is no longer a hypothesis

Tomorrow’s customer will no longer type “changing table” into a search engine. They will tell their agent: *“We’re expecting a baby, help me turn the spare room into a nursery, we like light wood and we do the work ourselves.”* Or, the following month: *“Find me two weeks in Thailand in November, and book a car for when I arrive.”*

This agent does not know everything from the outset: it **learns gradually**, conversation after conversation — your tastes, your budget, your constraints, your history. It no longer answers questions, it carries an intention and acts — booking, comparing, and increasingly buying.

If Apple confirms this direction at WWDC, it will be the mainstream illustration of a movement already under way: as our devices evolve, we will no longer be offered a mosaic of applications but a **consolidated interface** that connects us to the relevant services and brands. The important point for brands: **this agent becomes the interface**. Between you and your customer sits a piece of software that filters, recommends and chooses — as Google did for the web, Amazon for products, or Booking for travel.

## 2. The brand agent: the brand becomes a service

Faced with that personal agent, what does the brand do? At first, the customer’s agent will manage as best it can: it will go and read (“scrape”) the pages of your site, the way a hurried visitor would. But that is laborious, expensive in compute and error-prone. Very quickly it will prefer another route: **talking directly to a brand agent** that already holds the information, structured and reliable. Between two agents, the exchange is faster, cheaper in *tokens*, less prone to error — they are cut from the same cloth.

That is why a brand that wants to be well served by its customers’ agents has every interest in giving them someone to talk to: its **brand agent**. An intelligent point of contact that does nothing on its own, but orchestrates an ecosystem of specialised agents. In practice, each brand will not have *one* brand agent, but an **orchestrator** that receives the request and distributes it to specialised agents, each lightweight and expert in its own domain — a single agent doing everything would quickly become unmanageable and ruinously expensive in compute.

Take the nursery example again. The customer’s personal agent connects to the retailer’s brand agent. That agent plays conductor: one agent extracts the room’s dimensions from a video, an architect agent generates a 3D model, a decorator agent composes a proposal consistent with the tastes it has been given, an agent checks stock, a quoting agent prices two scenarios (“100% do it yourself” or “with a tradesperson”). A few minutes later, the customer receives not a list of products, but a project they can rotate with a finger and refine by voice.

The brand then stops being a catalogue. It becomes a **project partner**: inspiration, planning, coaching, introductions, maintenance. It sells an end-to-end solution, not an individual product. That is where loyalty and basket size are decided.

For that, the customer’s agent and the brand’s agent still have to speak the same language. That is the whole point of the channel that has just come into being.

## 3. A new channel: the agentic one

If every user has an agent, and every brand does too, then a new exchange layer appears: **a space that lets agents communicate with each other**. That is the agentic channel, and in a year it has acquired its basic protocols.

- **A2A (Agent2Agent)**, launched by Google in April 2025 then handed over to the Linux Foundation, standardises the “handshake” between agents: each agent publishes an *Agent Card* declaring its identity, its capabilities and the way it authenticates.
- **MCP (Model Context Protocol)**, from Anthropic, has become the de facto standard for plugging an agent into tools and data — more than 13,000 MCP servers published on GitHub in 2025. It is the layer that lets a brand agent genuinely query the information system (PIM, DAM, OMS…).
- On the transaction side, three agentic commerce standards appeared almost simultaneously: **ACP** from OpenAI and Stripe (the *Instant Checkout* launched in ChatGPT in September 2025), **AP2** from Google (Agent Payments Protocol, 60+ partners including Mastercard, PayPal and American Express) and **UCP** from Google and Shopify, presented at NRF in January 2026.

The consequence is simple to state: tomorrow, when you hand a purchase mission to your agent, it will turn your basket into an order **through these protocols** — and no longer by filling in a web page field by field. Checkout becomes a protocol, not a page. And a new channel means a new battle for visibility: in early 2026, AI-referred traffic was already converting 42% better than conventional traffic (Adobe) — a complete reversal on a year earlier. This is not a gimmick, it is an acquisition tap being turned on.

One question remains that the protocols do not settle: where, and on whose infrastructure, all this runs.

## 4. The hidden block: the platform you have chosen

Here is the fourth block announced above, and the least visible. **The quality of a brand agent depends entirely on the platform that runs it.** It all comes down to one word: **sovereignty**. Not in the flag sense, but in the very concrete sense of *who controls what*:

- **Where do your agents run?** On an OpenAI or a Google infrastructure — or on your own?
- **Where does your customers’ data live?** Does it pass through a model hosted by a third party, or does it stay within your perimeter?
- **Who holds the memory of the relationship?** The notes, the history, the context that make an agent “know” your customer — yours, or the platform’s?
- **Who can prove what happened?** The day a regulator or a customer asks for an account, do you have a tamper-proof log of the agent’s actions?

The real decision is therefore not “which brand agent to build”, but **which part of that chain you are going to own and control**. As long as the answers all point to a third party, your brand agent remains a *rented agent*: it impresses in a demo, but you are not at the controls. The good news: the issues that remain open are not a list of threats. Taken together, they are the specification for that platform.

## What remains to be settled

The vision is technically credible; it is the open issues that will decide the winners. None of them is inevitable — each of them has a known answer. The real question is not how to solve it, but who holds the keys.

**1. Identity and trust, in both directions.** How can each side know it is talking to the right agent and not to an impostor? On the brand side, certificates backed by a PKI (Public Key Infrastructure) and universal directory projects — the *Agent Name Service*, a sort of “DNS of agents” — are still under construction. On the customer side, delegation through cryptographic passes, limited in time and in scope. The essential point: this access control must remain in the brand’s hands, not in those of the personal agent’s host.

**2. Discoverability: the new search optimisation.** If the agent chooses for the customer, the marketing question becomes: **how do you get chosen by the agent?** Yesterday’s SEO turns into optimisation for agents. A brand whose data is not properly exposed will simply be invisible — or replaced by a better-structured competitor.

**3. Control of data — and of its memory.** This is the least glamorous and the most decisive of these issues. A brand must be able to make an AI talk to its information estate, and not only in read mode: the agent has to **remember** each customer, pick a conversation up where it left off, within watertight perimeters where each party sees only what their role allows. This is also where to start: an agent working internally delivers value immediately, at controlled risk. The spectacular nursery scenario comes afterwards.

**4. Who owns the customer relationship?** The platform hosting the personal agent — OpenAI, Google — becomes the new *gatekeeper*. The proof: OpenAI stopped its *Instant Checkout* in early 2026 to move to “apps” operated by the retailers themselves (Walmart, Target, Etsy, Instacart…). Whoever does not own their own agent risks being disintermediated: visible only through the platform’s terms — and its commission.

**5. Governance, compliance and liability.** Who answers when an agent buys the wrong product or goes over budget? The GDPR and the AI Act strictly frame profiling, consent and the right to be forgotten, across a chain of agents that is hard to audit — a good deal less hard, precisely, when you keep the register yourself and mask sensitive data before it reaches an external model. The Amazon v. Perplexity dispute gives a foretaste of the legal turbulence ahead.

**6. Brand and quality: the human in the loop.** A single bad automated answer can damage an image built over years. Brand voice, visual guidelines and the risk of hallucination call for human supervision — not as an option, but as a pillar: a confidence score on every proposal, sensitive cases put in a queue, and experts who train the AI rather than submit to it.

## What to do today

The window is narrow and most of the solutions already exist. The hardest part is not technical: it all starts from the **use cases you have already identified as the most unavoidable for your brand** — most leadership teams now know what they are. From there, you have several steps:

1. **Define the mission of your brand agents.** Framing before technology: what value proposition for each agent? You will not have just one — an agent for your subscribers, that knows them and suggests the right content; another for your prospects, in a different tone; a “corporate” agent that tells the brand story and carries the loyalty relationship, where yesterday you had a corporate website.
2. **Decide what you own vs. what you rent.** This is the decision that governs all the others: choosing an orchestration layer you control — your models, your data, your audit log — rather than handing the heart of the relationship to a third party. Everything else follows from it.
3. **Structure your data and give it a memory.** Expose catalogue, stock and specifications properly, but also let the agent retain and organise what it learns.
4. **Start internally.** Prototype an agent on a high-value use case, end to end, where the risk is controlled — before aiming at the general public.
5. **Adopt the protocols rather than ignore them** — A2A and MCP on the technical side; AP2, ACP and UCP on the transaction side. The cost of waiting is invisibility.
6. **Put the human loop at the centre**, from the first prototype, not as a patch.

## Conclusion

Only yesterday, all this belonged to futurology. Today the protocols are published, the partners lined up, the first journeys in production. The move to the agentic channel is no longer one technological evolution among others: it is a paradigm shift that redefines **who owns the customer relationship**.

And it is indeed the verb that matters: *to own*. The question is no longer *whether*, but *who* will lead — and *who* will remain the owner of their customer relationship, their data and their agents, rather than a tenant of a platform’s. Those who build their own orchestration layer, sovereign and auditable, will not merely exist in the agentic channel: they will keep its keys.

**Note - this conviction is not only theoretical. At ekino, we have built exactly that layer: a sovereign multi-agent orchestration platform, already in production in the public sector and at large accounts, where sensitive data need never leave the client’s network, where each agent only accesses its own perimeter, and where every action is logged in a tamper-proof way. The issues described above are not ones we merely contemplate: we build the tooling that turns them into concrete results.**

### Sources and reference points (last 12 months)

- ***A2A (Agent2Agent)** - Google, April 2025; transferred to the Linux Foundation (June 2025). Identity via Agent Cards. [Google announcement](https://developers.googleblog.com/en/a2a-a-new-era-of-agent-interoperability/) · [Donation to the Linux Foundation](https://www.linuxfoundation.org/press/linux-foundation-launches-the-agent2agent-protocol-project-to-enable-secure-intelligent-communication-between-ai-agents) *

- ***MCP (Model Context Protocol)** - Anthropic, November 2024; de facto agent-to-tools standard. [Anthropic announcement](https://www.anthropic.com/news/model-context-protocol) *

- ***ACP (Agentic Commerce Protocol)** - OpenAI + Stripe, September 2025; Instant Checkout in ChatGPT, discontinued in March 2026 in favour of discovery and retailer-operated apps. [OpenAI](https://openai.com/index/buy-it-in-chatgpt/) · [The reversal (CNBC, March 2026)](https://www.cnbc.com/2026/03/24/openai-revamps-shopping-experience-in-chatgpt-after-instant-checkout.html) *

- ***AP2 (Agent Payments Protocol)** - Google, September 2025; 60+ partners; mandates signed as verifiable credentials; donated to the FIDO Alliance in April 2026. [Google Cloud announcement](https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol) · [Donation to the FIDO Alliance](https://blog.google/products-and-platforms/platforms/google-pay/agent-payments-protocol-fido-alliance/) *

- ***UCP (Universal Commerce Protocol)** - Google + Shopify, NRF January 2026. [Sundar Pichai’s remarks (NRF 2026)](https://blog.google/company-news/inside-google/message-ceo/nrf-2026-remarks/) · [Under the hood (Google Developers)](https://developers.googleblog.com/under-the-hood-universal-commerce-protocol-ucp/) *

- ***Agent Name Service (ANS)** - an OWASP proposal for a universal agent directory backed by a PKI (the “DNS of agents”). [OWASP GenAI Security Project](https://genai.owasp.org/resource/agent-name-service-ans-for-secure-al-agent-discovery-v1-0/) *

- ***NIST AI Agent Standards Initiative** - launched on 17 February 2026 (agent identity and security). [NIST announcement](https://www.nist.gov/news-events/news/2026/02/announcing-ai-agent-standards-initiative-interoperable-and-secure) *

- ***Adobe** - in early 2026, AI-referred traffic converts 42% better than conventional traffic. [Adobe Blog](https://business.adobe.com/blog/ai-driven-traffic-surges-across-industries) · [Analysis (TechCrunch, April 2026)](https://techcrunch.com/2026/04/16/ai-traffic-to-us-retailers-rose-393-in-q1-and-its-boosting-their-revenue-too/) *

- ***Amazon v. Perplexity** - in March 2026 Amazon obtained an injunction blocking the Comet agent on its marketplace. [CNBC](https://www.cnbc.com/2026/03/10/amazon-wins-court-order-to-block-perplexitys-ai-shopping-agent.html) *

- ***WWDC 2026** - Apple keynote on 8 June 2026, strongly AI-oriented (redesigned Siri, agents). [Apple Newsroom](https://www.apple.com/newsroom/2026/03/apples-worldwide-developers-conference-returns-the-week-of-june-8/) *