Somewhere in the next few product cycles, a meaningful share of your customers will stop browsing your site altogether. They will tell an AI agent what they need, and the agent will find you, evaluate you, and buy from you, or it will not. That shift has a name: agentic commerce. And based on a conversation I had this week with a conference producer who covers customer experience for a living, most CX and IT leaders are further behind on it than they realize, mostly because almost nobody is willing to say so out loud.
This is a practical look at what agentic commerce actually requires from your organization, drawn from what I am seeing in my own advisory work and from preparing my own business as a reference case.
What is agentic commerce?
Agentic commerce is what happens when AI agents, not humans, do the shopping and transacting. It applies to both B2C and B2B. On the consumer side, instead of comparing products across Amazon, Walmart, or Tesco, a person tells their agent what they need, and the agent finds it, evaluates it, and checks out, possibly with a delivery service completing the last mile. On the B2B side, it is your customer’s procurement agent evaluating your product against a competitor’s, without a human ever opening your website.
That second point is the one most leaders underestimate. If an AI agent is the one deciding whether your company is even considered, your entire go-to-market motion has a new front door, and most companies have not measured whether they can be found through it.
Four infrastructure questions agentic commerce forces you to answer
Agentic commerce is not primarily a marketing question. It is an infrastructure and governance question. Four in particular come up immediately once you take it seriously:
1. Are you discoverable to your customer’s AI agent?
This is the search, engine optimization, generative engine optimization, and answer engine optimization question, but pointed at a new audience: agents, not people. If your content, structure, and metadata are not built to be read and cited by AI systems, you are invisible to a growing share of B2B evaluation and B2C shopping activity, regardless of how good your product is.
2. How do you authenticate an agent acting on someone’s behalf?
A human customer is easy to verify. An AI agent claiming to act for that customer is not, yet. Your authentication and identity infrastructure was not built for this, and most companies have not started rebuilding it.
3. Who is responsible when the agent gets it wrong?
If an agent purchases a one thousand dollar item when the customer only wanted the hundred dollar version, who owns that outcome? Is it the agent, the platform that runs it, or the customer? Can the order be returned or canceled, and under what terms? These are not hypothetical edge cases. They are contract, liability, and customer service policy questions your legal and CX teams need to work through now, before volume forces the issue. Visa is among the players doing early work on transaction authentication and approval for agent-initiated purchases, which is a signal the payment rails side of this is moving faster than most retailers’ policy frameworks.
4. Can your backend actually handle the volume and fulfill?
If agents can query and transact around the clock, at a pace and pattern no human shopper matches, are your systems built to respond fast enough? And once the order comes in, can you actually fulfill it? Agentic commerce turns supply chain and backend readiness into a customer experience metric, not just an operations one.
Why so few companies are talking about this openly
Adoption of agentic AI in customer experience and commerce is genuinely hard to gauge right now, and there is a specific reason for that: companies that are experimenting, or already implementing, are largely staying quiet about it. I hear two explanations consistently from leaders.
The first is fear of backlash. When a company signals that AI is reducing headcount needs in customer service or other functions, the public reaction can be sharp, and several well known consumer brands have learned that the hard way this year. The second is a belief that staying quiet protects competitive advantage: if competitors do not know what you are doing, they cannot catch up as fast.
I think that second belief is partly a trap. If nobody in your industry is willing to benchmark openly, you have no reliable way to know whether your agentic AI implementation is actually good, mediocre, or behind. Everyone assumes everyone else is further along, and that assumption becomes the only benchmark anyone has, which is not a benchmark at all.
A cautionary tale: what happens when the CX gap is already showing
Before any organization builds toward agentic commerce, it is worth asking a harder question: is your current AI-driven customer experience actually working today? I had a personal experience recently that is a useful test case. I called a large technology company’s support line after losing access to my multi-factor authentication app. Within a few sentences, it was clear I was talking to an AI voice agent, not a person. It could not correctly capture my last name after multiple attempts, and the only way to reach a human was to sound audibly frustrated, at which point the system detected that and offered to transfer me.
Once transferred, the human agent, in a different location entirely, asked me to repeat information I had already given the AI agent. After the ticket was logged, I received a follow up email asking me to confirm the issue again, in writing. That is the same information communicated three separate times across three disconnected systems.
Somewhere internally, a dashboard almost certainly shows this as a success: an AI agent handling a large share of call volume, a strong deflection rate, a lower cost per contact. None of that reflects what the customer actually experienced. If a company’s own AI customer service still cannot pass information between systems reliably, that same company is not ready for AI agents transacting on its customers’ behalf. Agentic commerce does not fix a broken CX stack. It automates whatever is already there, good or bad, at a scale humans cannot match.
A practical starting point: get found before you get transacted with
The most tractable place to start is discoverability, because you can measure it and act on it now, independent of the harder authentication and liability questions. I ran this as an experiment on my own site over the past three to four months: restructuring content, tightening keyword and entity clarity, and rebuilding pages specifically to be cited well by AI systems, not just ranked by traditional search.
Checking the analytics after roughly three months, my site had been cited by AI systems more than 1,200 times, with two or three articles consistently appearing at the top and driving two to three hundred page views each. That is a directly measurable outcome of treating generative engine optimization as core infrastructure rather than a marketing afterthought, and it is the same groundwork that determines whether your business shows up when a customer’s, or a business buyer’s, AI agent goes looking.
Where to start as a leader
- Audit discoverability first. Can AI systems find, correctly interpret, and cite your product and pricing information today? Most companies have never checked.
- Assign ownership of the four infrastructure questions. Discoverability, authentication, liability, and fulfillment volume each need an accountable owner, not a shared assumption that someone else is handling it.
- Fix your current AI-driven CX before you extend it. If your existing AI customer service creates friction, agentic commerce will scale that friction, not remove it.
- Benchmark honestly, even privately. Silence about your own agentic AI progress is not the same as competitive advantage. Without a real benchmark, you cannot tell the difference between ahead and behind.
Agentic commerce is not a distant, speculative trend. The infrastructure questions above are answerable today, and the companies that answer them deliberately, rather than reactively, will be the ones agents actually choose to transact with.