Tariffs are shifting. De minimis rules are under pressure. And new enforcement could change how retailers operate overnight. If you sell across borders—through eCommerce, marketplaces, or direct—these changes don’t just impact compliance. They impact your pricing, margins, and conversion rates.

Register for our Trade and Tariff webinar series for regular updates.

I spent 30 minutes with Danny Smith from Stripe talking about agentic commerce, and I walked away with three things I can't stop thinking about.

Software agents convert about four times better than humans. That's not a rounding error. That's a different species of buyer. The reason is simple once you say it out loud: by the time an agent hits your checkout, the decision is made. No browsing, no second-guessing, no abandoned cart because the kids needed dinner. The agent shows up to finish, not to shop.

So what do most merchants do? They treat that buyer like a threat. Years of bot-blocking, CAPTCHAs, and forms built for human eyes. All of it now points the wrong way. You spent a decade hardening the door, and the highest-converting customer on the internet is standing outside it.

Danny put a number on it that stuck with me. Without the right plumbing, agents fail to check out roughly two-thirds of the time. They can't read your human-friendly form. Then your fraud system looks at a buyer with no mouse movement, no hesitation, no messy human behavior, and flags it as card testing.

Think about that. The cleanest, most decisive buyer you'll ever get trips every alarm you have. We built fraud models on the assumption that real people fumble. Agents don't fumble. The model breaks, and you lose the sale, and you never even know it happened.

This was the part that reframed it for me. Danny laid out 4 models, and the useful bit is that they're not competing visions. They're rungs.

Bottom rung, a human still hits buy. Think, ask Ralph, or shop through Alexa. Next, you hand the agent a scoped, budgeted, and time-bound token so a hallucination can't drain your account. Above that, hybrid autonomy, where the agent runs inside your policy and only escalates when it steps outside. Top rung, agents paying agents directly, settling in stablecoins for transactions too small to bother a human with.

Most of you don't need rung four. You need to stop pretending rung one doesn't exist.

The bots already make up the majority of web traffic. The only question is whether yours is set up to sell to them or just to fight them.

What's your robots.txt doing right now? Be honest.

None of this requires a moonshot. Danny's near-term advice was almost boring, which is how you know it's right. Update your robots.txt so it acts like a doorman instead of a bouncer, and explicitly wave through the agents you want, Claude, ChatGPT, Gemini. Add an llms.txt so agents stop burning tokens scraping your site to figure out what you sell. Look at machine-payment protocols if you want to charge for that traffic instead of eating it.