On Shopping Super Agents with Li Haslett Chen // BRXND Dispatch vol 118

An interview with Li about how shopping will be the use case for the first scaled consumer “super” agent, why creators are the PhDs of commerce, and how ads will enrich agentic experiences

As we build towards this year’s BRXND NYC conference, I’ll be doing a series of interviews with founders, investors and marketing leaders building in the problem space around how AI is fundamentally reshaping the way consumers build preference and purchase brands. If there’s someone you think I should get to know, please drop me a line at mike@brxnd.ai.

Speaking of BRXND NYC 2026, tickets to the show are $749 through July 15 before we hit the GA price of $1,000+.

If you’d like to attend the show, please add your name to the early bird waitlist here and purchase before July 15 to get the best possible pricing.

---

On Shopping Super Agents

Two things are true right now about the nebulous world of agentic commerce:

1) Finding serendipitous inspiration on what to buy remains a trillion dollar unsolved problem, arguably the most lucrative on the consumer web.

2) Even as Anthropic and OpenAI race to IPO at trillion-dollar valuations, they do so in a world without a mass market consumer agent.

There’s a frenetic race to change that as OpenAI gets set to effectively relaunch a version of ChatGPT as a “super app” that prioritizes agents over chat-based answers. In any semblance of first principles thinking, chat has never really made sense as a plausible medium for something as visual as shopping. Neither for that matter have rectangle search bars and PDP squares but historically it’s been the best bad interface we got.

Few have obsessed about this problem space more in the last decade than Li Haslett Chen, founder and CEO of Howl, the leading creator commerce platform for consumer tech, gaming and wellness.

Li is a Forbes AI 50 winner, Financial Times Retail Disruptor, served on the board of Warner Brothers Discovery, and has written for The Information about how AI earns its place in commerce.

For its part, ChatGPT certainly isn’t giving up on commerce, even as it pursues ads as the core monetization model. Here’s an email I (and presumably tens to hundreds of millions of other ChatGPT users) got this morning 20 minutes before I was set to hit send on this piece.

Here I must confess, this interview is a touch more personal as I too obsessed about this problem for several years alongside Li, first as her head of marketing and then as a sales leader for her previous company, Narrativ. We spent much of 2019 working on solving semantic search for commerce when OpenAI was still a plucky non-profit, concerned that GPT-2 possessed too strong an ability to generate fake news.

Li and I caught up for a whirlwind discussion around how shopping will be the use case for the first scaled consumer “super” agent, why creators are the PhDs of commerce, and how ads will enrich rather than enshittify agentic experiences.

A lightly edited version of our conversation follows below:

---

Mike: While agents voraciously devour more and more B2B workflows on top of coding, we’re still waiting on the first truly transformative consumer agent.

You’ve made the case that this will ultimately come in shopping/commerce. Can you lay out why shopping is the first major consumer category where agents will offer a transformative UX?

Li: There’s a version of this story where the first transformative consumer agent shows up in something “important” - your health, your money, your relationships.

But agents don’t just go where the stakes are the highest, they go where friction is highest and the infrastructure already exists. That’s shopping! Payments, feeds, reviews, returns - the plumbing already exists.

Shopping is the most ordinary thing we all do - it’s high frequency, relatively low stakes. The world doesn’t have to be rebuilt for it.

What’s missing to solve shopping isn’t a technology problem, it’s a strategy and conviction. Good software is opinionated software and opinionated design.

We have Codex for developers, we have Codex for marketing for and sales teams, but we don’t have Codex for consumers - and I want to see Codex for shopping.

Shopping is continuous, it’s complicated, it’s massively operational. That’s the perfect job for an agent.

Here’s my opinionated, “contrarian” take on the agentic UX: It’s time to take shopping seriously. Don’t treat it as entertainment and build the UX on impulse buys. Ads have that covered. Shopping is how we spend our money and time. It’s identity - Your home, how you show up for yourself and your kids, what you give the people you love. That deserves a purpose-built surface, your personal command center. It can still be visual and fun, but it also needs permanence and structure.

Mike: To date, much of the agentic commerce discussion has focused on solving the last mile problem of an agent purchasing an item on a user’s behalf. There’s obviously a hell of a lot more TAM in an agent that could accelerate discovery for a shopper. What would need to happen to cross that chasm?

Li: Checkout and discovery are two different kinds of problems. For an agent to actually help you discover new things, it has to know you - your taste, your budget, the difference between the thing you need and the thing you’ll cave on. No one has taught the machine how to get to know you.

The last mile is a bounded problem. I don’t want to downplay the protocols - UCP, ACP, that work is very real - but when the agent knows the SKU and has your card on file, it can execute. So everyone built the checkout robot first. Fine.

Mike: So…what kind of signal can an agent realistically get quickly to create an experience that quickly feels magical rather than arduously teaching a machine what you like?

Li: Here’s the interesting part: the best machine we’ve built to date for getting to know a shopper is advertising. Every ad is a tiny experiment - platforms show you something, watch what you do, learn. Meta and Google already understand us incredibly well.

What’s amazing is that hardware-embedded agents will have a whole new generation of inputs coming - complementary but also incremental to ads. Imagine an agent that, with your permission, can see what’s on your calendar, what’s in your inbox, the conversations you’re actually having. The good version of this isn’t creepy. It’s invited. You let it in because it works for you, not the advertiser.

There’s another simple way to learn about someone: you ask them. Every good B2B agent is prompted to interview you.

A great personal shopper observes you…AND asks you questions. It’s a no-brainer that your shopping agent should ask questions that a great personal shopper would. Why do you keep buying and returning slim fit pants? Whose taste do you trust now versus 2 years ago? People love being asked about themselves. Done right, that’s not friction.

Once agents have access to better data than ads engines do, they can just focus on being useful.

Mike: Isn’t this what OpenAI was trying to pull off with Deep Shopping Research? And to a large degree, the old principles of online commerce applied where slow, cumbersome experiences frustrated users.

Li: I love this question - yes, there’s a surface resemblance but the mechanics are very different.

Deep Shopping Research interviews you because it has to. It’s starting from zero and the questions are very basic: what’s your budget, what’s the use case, what have you tried. That’s a cold start dressed up as a conversation.

An agent with persistent context asks questions to refine what it already knows. This is a relationship, instead of ‘what’s your budget?’ Your agent can say “the last two months you didn’t spend your full budget, I think we can stretch it on this couch since you keep going back to it - what do you think?”

OpenAI saw the right behavior - agents should interview - but bolted it onto the wrong architecture.

Mike: What kind of context does a consumer shopping agent ultimately need to be successful that is hard to access today? Why is that information so hard to gather and present in a structured way to an agent?

Li: I’d start at a more foundational level. Every other category of AI has a benchmark - coding has eval suites, math and science have clear tests engineers race to beat.

Shopping doesn’t have its own Turing Test. We have no standardized way to measure whether a shopping agent is any good.

Mike: Oh…I like where this is going

Li: You can’t improve what you can’t measure - and I think that’s exactly why coding agents feel like magic while shopping still feels primitive.

The decision-making axes I’d measure, in today’s vernacular: price-maxxing, the best deal; quality-maxxing, the best thing; trend-maxxing, what’s having a moment; identity-maxxing, what’s most you.

Mike: We’re a little bit older these days than we were when we worked together. Are we still allowed to drop much gen alpha lingo?

Li: A great agent doesn’t just optimize one axis. It demonstrates judgement and knows which of these matter right now and it weighs all of them against what it knows about you. That’s where the continual access to data is so critical. Whoever delivers judgment best will dominate the category.

Mike: I’d like to go deeper on ads since OpenAI is all-in on ads as their free consumer monetization model and anything Meta does has to be grounded in ads yield.

One of the major innovator’s dilemma challenges that has plagued early agentic shopping experience is that all companies building agents (i.e. Amazon) are existentially dependent on ads.

So……how could ads enhance a consumer shopping agent rather than make it worse?

Li: I don’t think ads and agents are at odds. People frame it as adversarial - if the agent works for you, ads must lose - but I think that’s backwards. Ads are one of the best on-ramps we have for an agent to learn about you, and they can be the doorway into the agentic experience itself.

Say you see an ad that catches your eye, but you don’t trust what it’s telling you. Today that’s a dead end - you either take the bait or scroll past. With an agent, you can hand it off: “look into this for me.” It runs an ad-research skill that does the digging you’d never do yourself - whether the thing is actually good, who else makes it, whether the price is real, and how it squares with the budget and taste it already knows you have. The ad did its job: it surfaced something. Your agent does its job: it cuts through the pitch and tells you what’s true and whether it’s actually right for you.

Mike: This is something big I think many people are missing when they think about “marketing to agents” and how AI will mediate preference. Agents will be extensions of real users and will need heuristic help to make the best decisions!

Li: Yes! That’s when ads and agents stop competing and maybe even start accelerating each other. The advertiser gets a genuinely interested shopper and you get a researcher and an assistant in your corner.

The one non-negotiable is that the agent works for you and only you, the shopper. Keep that line clean and ads and agents make each other better.

Mike: By all accounts, Meta looks poised to take the first major crack at a large-scale consumer agent that “works for you” as the core of the UX in the coming weeks with Hatch. Initial thoughts here?

I agree Hatch is one to watch. I’ve said that creators are the PhDs of commerce - and Meta has the strongest starting position - Instagram sits on billions of hours of organic creator content oriented around shopping.

Try the same prompt in meta.ai versus gemini or openai, meta is already better - and it’s because of creator content.

AI can teach itself how to code, it cannot teach itself how a couch feels. Product experience must come from qualified creators and the best shopping AI will be the one with access to the most lived experience.

Creators are the obvious entry point into an agent-led experience. If you’re watching a haul video from a creator you follow, imagine just saying: “put everything in this video in an excel, research it and add links, notify me if anything goes on sale.” This is something people are already trying to do with screenshots and twelve open tabs.

From there, your agent can earn the right to do more: pull in the savings codes, handle the returns, save shoppers time and money every week.

Mike: So of course, I have to ask the question that looms over any conversation around content owners and LLMs.

How do creators get fairly compensated for the existential role they’ll play in any true agentic shopping experience?

Li: Creators and publishers provide the commerce data these AI systems can’t generate for themselves. The creators whose content trains and feeds these experiences have earned a cut of the action. The discovery has always run through them and the economics should too.

The platforms know this - in March, Meta relaunched creator affiliate commerce on Instagram and Facebook after sitting out that market for three years - and six weeks later, it announced a consumer agent. I don’t think that’s a coincidence.

Gemini has Youtube and it already has its own creator rails. Both Meta and Gemini have the creators, the feed, the affiliate rails, the agentic ambition - that’s the entire chain for agentic shopping.

Mike: So how crazy are things going to get once the first consumer shopping agent gains scale? Will I recognize online commerce as I currently know it?

There’s a familiar pattern here. Every new platform starts by skeuomorphing the old one - the first websites looked like brochures, the first mobile apps looked like websites, and AI shopping today looks like Google circa 2005: a text box and blue links. That’s always temporary.

I agree that “chat is dead” but “super app” is the wrong noun. An app - even a super one - waits for you to open it. The whole premise of an agent is that the destination disappears.

What’s actually coming is the super agent. It’s a pin, an ear cuff, a watch - you never open it because it’s never closed. Your wearable syncs with the nearest screen - and when you need to see something, it doesn’t send you to a website, it spins up a new one for you.

Mike: Thanks so much for the great conversation. Until we get the second coming of Tropical 128…IYKYK

---

Li Haslett Chen is a Forbes AI 50 winner, Financial Times Retail Disruptor, served on the board of Warner Brothers Discovery, and has written for The Information about how AI earns its place in commerce.

To suggest additional builders that I should feature in this series, please get in touch at mike@brxnd.ai

If you have any questions, please be in touch. As always, thanks for reading.

— Mike