What Agentic Commerce Means for Online Sellers

Show notes

At Cross Border Summit 2026 in Chiang Mai, Janette Toral asks Mike Michelini where agentic commerce could help online sellers, how to tell whether an AI agent is worth the investment, and which decisions should stay with people.

In this episode

1. What stood out at the summit — Janette reflects on seller resilience, quality inspection and TikTok as a discovery channel. Mike describes a market where experienced sellers are still upgrading their skills.

2. What Mike would work on over the next 90 days — Give each agent a specific role, then improve how its work is measured and how the team communicates with it.

3. Where agents might fit in product discovery — Janette asks where agentic commerce could improve profitability. Mike shares his early view on how shoppers and their agents might research products and use a seller’s own website.

4. How to decide whether an AI agent is worth it — Mike weighs sales, operating costs and avoided losses, with time saved as his clearest example.

5. Which decisions still need a person — The conversation turns to limiting agent access and keeping sensitive actions, such as sending a wire transfer, with authorized people.

6. Who owns cybersecurity — Janette asks who secures automated workflows. Mike describes the system administrator’s role in updates, credential rotation and permissions.

Full interview transcript

Thanks Jeanette.

Came from the Philippines.

Manila.

And you hosted a great panel.

As always, you always, I've always flexed, you know, you can always make things happen.

We had some adjustments.

End of day one.

Cross-border summit 2026.

Yes, and congratulations, Michael.

It's a wonderful event.

I mean, of course, having seen the one in Shenzhen.

Yeah, we were in Shenzhen.

And now here in Chiang Mai, it's amazing how the community has grown and matured also because a lot of them are repeat attendees, anyway, we learned so much from the experiences of the panel today.

I mean, of course, everyone is recognizing the various challenge that entrepreneurs are facing, especially e-commerce centers.

But despite that, of course, we all need to continue doing our business, we all need to continue serving our customers, and it's amazing how flexible and resilient our entrepreneurs are.

Also adjusting to all the latest technologies that are available, improving their operations with AI and agentic AI as discussed by Michael extensively, and we've also seen how others focus on sustainability, differentiation, and not fall into the price trap type of competition.

We also learned how automation can be used to streamline operations and really improve the transparency of the data that we have, what are the things that we should know about the data that we collect or the data that we process in order to make better decisions.

And one thing that stand out to me also was the quality inspection.

Those are behind the scenes that are often not discussed.

Usually entrepreneurs are just worried where their products will be put on hold by customs.

But at the end of the day, if your product went through proper quality inspection, the likelihood of that happening won't be at all.

And there was also an interesting discussion as well about TikTok, and we learned a lot.

I think even if it's simple, I think it has to be simply explained, but it has to be repeated.

The product must be demonstrable, and then you have to be able to be very, how do you call this, repetitive, if you have to, until you get the right formula and then you get the right angle.

It's true.

Yeah.

And if brands will be able to do that.

And he also emphasized that even if you're on TikTok, if you don't have an Amazon, it kind of disappoints the buyer.

So being present to where you need to be present and creating awareness, discovery through TikTok.

It's a wonderful strategy.

How about for you, Michael?

Yeah.

What, I mean, you've been doing this every year, but what stood out for you this year?

I mean, I don't want to, I don't want to say harder, but it's getting harder.

And like you said, you have to be resilient, upgrading the technology skills, logistics, transparency, like everybody's maturing and the market's maturing.

I think as much as I hope there's new sellers, I think it's more like experienced sellers, they're here.

And, and, and maybe I can also throw the question to you, I mean, what advice can you give to entrepreneurs to increase their profitability, especially with the use of AI, agentic commerce, how do you see that coming into play?

So, you know, agentic commerce is still such a new thing.

And I, you know, I mean, I know, I know you said profitability and others here are more profit centered.

I think, you know, me, I'm more of the,

you're doing it for the agents, doing it for the agents.

I'm a, I tried to do things early and it might, might be hard to know about the profitability now because it's so, it's an emerging technology.

but like, but in your case now, if you're going to focus on the next 90 days, what are you going to work on? Based on what, what you've been thinking about lately?

For me, like you said, training agents, better, just like people, you need agents that have different skills and knowledge and abilities to do things.

So I'm going to be working even harder on refining which agent does which tasks and, and how they do that and how I measure and communicate with them.

Actually, Michael, I thought you were going to join the panel earlier.

Oh, okay.

I don't know if we put it on here.

Yeah.

Okay.

Let's make, let's make this vlog seriously ready.

All right.

All right.

Let's do it.

Maybe we can make it a podcast.

Yeah.

Maybe we can make it a podcast.

Uh, I actually, because I, I have, okay.

So where do you see the most immediate opportunities for agentic commerce to improve profitability rather than simply introduce another technology expense?

So, you know, as J.P. said on the panel, it's hard to really measure people search TikTok, maybe then they go to Google, then they go to their AI LLM.

But I believe they're going to, I believe agents will be more bottom of funnel.

The human will speak to their agents to find product suggestions for a solution.

The agent will do maybe the top of funnel.

So they need to maybe read the content, to read your content.

I think me, like I said earlier, I'm making blogs not for humans, but for agents.

They're going to want to verify your, you have, even though they know the answer, they want to see that your website or your business talks about those things to be relevant and has the educational information.

So, um, I still have trouble measuring it, but they said Shopify says from what I learned from Tomer is about 30% or more of transactions are coming from LLM.

So how, so following that, so what should business measure to determine whether implementing an AI agent is worth the investment?

Sure.

So I think there's, there's two different, two different ways.

I mean, there's three ways to make money or monetize, of course, make more sales, but the other is save money and the third is protect losing money like insurance.

So I don't know if this really protects you, but I think it can help you make more money, but also save money because you can do more, you can scale more work.

What used to take me so much more time manually I can do in a couple of prompts.

So I think it's going to help you further save time and save money to focus on the hard questions.

So somebody said, it's like everybody now has the ability of say Jeff Bezos as like a huge CEO because he only gets the hardest questions now.

I mean, he, I think what they say as a CEO, what they get is the hardest questions, the easy stuff, or even the medium stuff is done by people or agents in today's world.

Their job now is just to deal with the hard questions.

So we've seen AI and automation being used to improve inventory forecasting, advertising decisions, and even operational efficiency.

But there were also concerns about allowing AI to make decisions.

So you said that, okay, you can now make the harder decisions.

So in your case, where should human oversight remain essential or where, I mean, in your experience, where has it remained essential for you despite all the automation initiatives that you have done?

Say the last part again?

The human oversight.

I mean, we've done a lot of automation, we've done a lot of AI experiments, but where did you realize that, okay, this still has to be me deciding even as you thinker, you thought it can be automated?

I think, like I give my agents human names and I think of them somewhat like an extension of a person.

And just like a person on my team or in my network or in an agency, I give them the knowledge and the access and the tools, or we call it like permissions to do those things.

So some people think, well, wouldn't the AI just steal your money from your bank, but I'm not giving that agent, maybe they can read my bank if they were doing my accounting, I'm not going to give them the ability to send the wire.

The same with a human on the team, you're not going to give them the ability to transfer money.

So you would think of it in that way, but I still think you should trust the agents a bit more to make decisions.

Just like with humans, there's bottlenecks, if you can't delegate or empower agents or humans to do things, you're not going to be able to do as much.

But I protect myself a little bit because I have multiple different agents on different, it's not like one agent has access to everything and can do anything.

You know, there was an interesting discussion in our mastermind table yesterday in automation, because they've been talking about how can you do automation on this and on that.

And then I asked about how do you put in cybersecurity in place in automation and in AI agents?

Because at the end of the day, of course, we've seen a lot of rogue or unexpected activities.

You cannot say that AI systems can't be hacked, you cannot say that because AI systems can hack another AI system.

So what are your thoughts on cybersecurity, how do you implement cybersecurity as you do all of these automation and AI agent experiments?

So I think, my team, we've worked together.

So I have, I can't say his name but the title...

Or do you also have a cybersecurity agent?

You can have an agent, and I actually learned from some here that they have an agent for that.

It's called human, and it's, I would call it system administrator.

I think that's what I've been telling people, they're maybe not, of course, they could be a programmer, but their more job is permissions, access, upgrades.

So you need to make sure your softwares are up to date.

I mean, of course, like even example, WordPress, right?

If you don't update a plugin or update WordPress, for example, you can get hacked.

It's not like 100% that you'll never get hacked, but you have to maintain.

You have to maintain, of course, passwords, rotating API keys or passwords, limiting the amount of users or what users have what permission.

But I would say it's a system administrator.

And it shouldn't be you, of course, you as the boss or owner can do it.

But I would say you should have a system admin on your team that, of course, you have to trust them.

They're going to have the access.

Of course, agents can help you with updating, rotating keys and things like this.

But it's honestly, we're in the end of 2026, there's been a lot more hacks.

I think it's going to get even worse.

Some people think it's just ripping the bandaid off.

We're going to get hacked.

Like the Binance hack.

Yeah, Binance is so scary.

There's been other big ones, like even the Linux kernel.

I'm not so technical to speak for it, but everything's strengthening and hardening.

So more interesting developments and more interesting challenges to come.

And we hope if you are not able to join Cross Border Summit this year, you should join next year.

Yeah.

And share your experiences.

Be part of the discussion.

And even help shape the topics that will be presented at Summit.

We'd love to have you.

We tried our best to involve everybody attending, we brought people up on stage, we learned about what people are doing, and we would love to have you there next year.

And it's amazing how the AI discussion last year has changed this year because of the new changes happening.

So much.

It will be interesting to see how maybe agents will play a role next year, maybe we'll have robots inside the contest next year.

Yeah.

Who knows, right?

We'll see.

I would like that.

But maybe, we'll see.

Who knows?

Maybe we'll have a robot next year.

I hope so.

I would like that.

Fingers crossed.

All right.

Thanks, Janette.

Thank you so much.

Thank you.