What Happens When AI Does the Shopping? Marketing to Agents Without Losing Your Brand

Introduction

Personal shopping agents went from a hobby project for technical early adopters to something any prosumer can switch on in a matter of weeks. Jamie Domenici, Chief Marketing Officer at Klaviyo, the B2C CRM serving more than 200,000 consumer brands, has a front-row seat to that shift. With 20 years in SaaS marketing, including a decade at Salesforce and a CMO run at GoTo, Domenici lays out how marketers should split their effort between humans who feel and agents who don't, how search is being rebuilt as answer engine optimization, and where autonomy should and shouldn't be handed to AI.

n

Who Owns the Customer When an Agent Buys?

Domenici's position is that neither the brand nor the agent owns the relationship outright. The agent executes the purchase, but the consumer trains it, sets its preferences, and still says yes or no. I pushed back on that, because in practice most of us now ask an LLM which brands to consider before we ever form an opinion. Her response was that the marketer's job didn't shrink, it multiplied. SEO, paid media, and direct buys still matter, and now AEO and agent discovery sit alongside them as additional surfaces to cover.

n

Humans Respond to Emotion. Agents Don't.

This is the clearest framing from the conversation. As Domenici put it, "Humans respond to emotion. Agents do not care. Agents have no emotions." Brands still need the human co ection work: positioning, creative, memorability. But the purchase increasingly runs through a system that evaluates structured signals, citations, and indexability. She describes it as a move from optimizing for attention to optimizing for decisions, where the science is understanding what trains the agent and the art is influencing the human who influences the agent.

n

What AEO Is Actually Doing to Traffic

Klaviyo's attributable AEO traffic is still small, but direct traffic has tripled. People research in an LLM, then arrive at the site already decided and wanting to see the product. That changed the website's job. Instead of a long educational journey, Klaviyo now A/B tests toward getting visitors into a trial from the first click. On the content side, the team leaned into video, optimized its YouTube cha els, spends real time on Reddit, and restructured pages so LLMs can cite them. Domenici is candid that she hasn't cracked the code and still feels at the mercy of algorithm changes, but the foundational work is moving the needle.

n

Brand Is the New Differentiator

Domenici argues that execution-layer tactics are being democratized. Keyword research, landing pages, and long-form content are now automated through tools like AirOps and Vercel, so Klaviyo runs that engine with one content engineer instead of a team of copywriters. What can't be automated is being memorable at the top of the fu el. She also pushed back on the idea that nothing is measurable anymore: multi-touch attribution, incrementality testing, and A/B testing now run at a speed and scale that wasn't possible a year ago. The insight is partial, one or two points at a time, but it's improving.

n

Earning Autonomy: Where AI Runs Email and Where It Doesn't

Klaviyo uses Composer, its own marketing agent, to audit flows, recommend campaigns, draft content, and build segmentation. Nothing ships while the team sleeps. Humans sit in the approval seat, declining, training, and enforcing brand guidelines. Domenici's rule is that autonomy is earned, not switched on. Every output gets A/B tested against results, and as confidence grows, more of the process moves to the agent. We agreed transactional, templatized sends should go out fast without a human QA step. Net-new campaigns and personalized promotional work still need a person checking tone and voice.

n

Where the Money Is Going

Domenici keeps a small testing budget, runs competing agents head-to-head, then commits to one. The real spend isn't tokens for content. It's infrastructure: Profound for AEO measurement, Vercel for building and publishing apps, and Claude for internal work. Her bigger operational problem is tech sprawl. Open access to Claude, ChatGPT, and Glean meant everyone built their own app and workflow, leaving a two-person marketing ops team with redundant systems nobody can maintain. Her answer has been to catalog every marketing workflow in a spreadsheet and automate them one at a time, starting with high-yield ones like product launches, where a code push in Linear now generates the blog post, the What's New update, and the internal Slack to customer success.

n

Conclusion

The takeaways are practical. Market to the human and the agent separately, because one responds to emotion and the other to structured signals. Treat AEO as a growing surface alongside SEO, watch direct traffic as the real signal, and redesign the site for visitors who've already done their homework. Let agents earn autonomy through tested results, automating transactional sends first and keeping humans on anything that carries your voice. And before adding more AI tools, inventory your workflows, because the governance problem arrives faster than the productivity gains. As Domenici noted, what's old is new again. Brand wins, and marketers who manage the architecture, not just the tools, will keep up.