Skyfire

Why AI shopping agents only work if they outperform humans The entire value proposition of an AI shopping agent hinges on one thing: it needs to outperform what you would do yourself, otherwise there is no reason to use it over just buying directly from a retailer like Walmart. Amir's example is a set of headphones: you tell the agent your preference, and it scours the internet for the best price, applies available discounts, and completes the purchase regardless of whether you have any existing relationship with that merchant. That last part is the key unlock, agents acting independently of brand loyalty or an existing account, but it only becomes genuinely useful once this kind of experience is scaled broadly across the internet rather than limited to a handful of partnered retailers. The bigger implication is that AI agents could fundamentally reshape how merchant relationships and customer loyalty work in e-commerce. Skyfire

Opening a bank account, getting a credit card, and making a purchase online means proving you're you. There's decades of KYC infrastructure behind those moments, and none of it applies to an agent. 🚨 A trillion dollars of commerce is projected to run through agents by 2030. Most merchants still can't tell you how many agents hit their site yesterday. Skyfire CEO and co-founder Amir Sarhangi has been calling the missing piece Know Your Agent (KYA) since the day we started. Verified information on the developer that built the agent and on the human it's working for, plus a record of what that human actually approved. Amir joined Sam Boboev on the WRAP UP Podcast (Fintech Wrap Up) to get into how that credential travels, and what merchants and card issuers can be doing now while the volume is still early. The conversation also covers: ➡️ Why the token goes to web security providers rather than to merchants one at a time ➡️ What has to change before micropayments can unbundle SaaS pricing ➡️ The slice of merchant traffic nobody can currently account for ➡️ Why agent commerce, built right, ends up with less fraud than what we have now Link to full episode in the comments

What happens when AI agents start buying for you? As AI agents evolve from recommending products to autonomously searching, comparing, and completing purchases, an entirely new set of questions emerges around identity, trust, and fraud in payments. Amir Sarhangi, CEO of Skyfire, introduces the concept of Know Your Agent, explaining why AI agents need their own digital identity so merchants can distinguish legitimate agents from malicious bots. Stablecoins and micropayments are emerging as the preferred rails for this machine economy, since traditional payment infrastructure was never built for the volume and speed of AI-to-AI transactions. For anyone in fintech, payments, or e-commerce, this points to a future where software increasingly becomes the customer, and the infrastructure to support that shift is being built right now.

An agent that can build agents that can buy. That's the power of Skyfire x Replit. The Replit team wired a Replit-built shopping agent to Skyfire and let it loose on a real merchant site. You ask for a gift, a mini fragrance sampler for mom. Skyfire issues a Know Your Agent (KYA) token on the spot. The agent shops live, pulls real options at real-time prices (the set had jumped $10 since that morning), you approve with a passkey, and a sub-agent runs the entire checkout on its own. No credentials exposed to the model, no human clicking through checkout. Receipt, and confetti. The part we love: Replit is out to empower the next billion software creators, and it built this whole thing with its own AI agent. Keeping those builders at the frontier means letting whatever they ship actually transact. With Skyfire underneath, it does. Agents building agents that buy. Great work, David Jakubowicz and the Replit team. Full demo video linked in comments #agenticcommerce #AIagents #payments

This is what agentic commerce looks like when identity is built in from the start. In the live demo linked below Nash Ali, VP of Strategy at Experian, walks through an end-to-end purchase made by an AI agent. A consumer asks an agent to find noise-cancelling headphones from Bose Corporation. Behind the scenes, Experian runs passive fraud checks on the human, Skyfire issues a KYA token that the agent carries as it shops, and Akamai Technologies lets the agent through because it can unpack and verify that token in real time. When money enters the picture, the consumer verifies their identity with a single tap on their phone. Skyfire creates a pay token, the transaction routes through the card network, and Bose calls an Experian trust score before approving checkout. No guesswork about who is behind the agent. No blocked bots. A real transaction, completed in minutes, with the human explicitly in the loop at the moment that matters. Watch the full, live demo on YouTube here: https://lnkd.in/gxp2pfb7

I wrote a blog post for Fastly on agentic commerce, payments, and where the edge fits in. AI agents are moving closer to real commerce flows: discovery, checkout, payments, and fulfillment. That shift brings a new set of infrastructure questions around trust, authorization, payment verification, and traffic control. The post looks at the protocol landscape taking shape and why the edge is a practical place to enforce policy before agent traffic reaches the origin. https://lnkd.in/e5TRXmd5

18 years since we started operations. 8 years of our Center of Excellence in India. Two milestones worth pausing on. This year Sequentum marks 18 years of web data collection infrastructure and automation at enterprise scale. It also marks 8 years since we opened our Center of Excellence in India, the engineering team behind so much of what we ship. A lot has changed since 2008. The web got harder to collect from. Compliance got serious. And now AI agents need clean, governed data to learn from, trust and to act on. That last shift is the one we were built for. Web data you can trust for the AI era is the same discipline we have practiced for 18 years, now feeding the models and agents that run on it. We are proud of our recent launch of Agent Builder, an AI-powered tool that turns a natural-language prompt into a production-grade web scraping agent in minutes. It works on its own, or with our Visual Editor for customization down to the atomic level. That discipline now reaches past data collection into transactions and agentic commerce. Our Checkout Agent lets AI agents complete real purchases with identity verification, spending controls, and audit trails. We built it with our strategic partner Skyfire, and we are a member of KYAPay.org, the open protocol for transparent and trustworthy agentic commerce. Same standard, new frontier. What has not changed is that standard: precision, compliance, traceability and reliability at scale. To our teams in the US, Australia and India, and to everyone who has built alongside us, thank you. We take this work seriously because our clients count on it. We also know how to enjoy the ride, and we look forward to celebrating this milestone together. Here's to the next chapter. Web data you can trust for the AI era.

At our event last week, The Trust Layer, co-hosted by Okta and Experian, every single demo ended with an agent buying something. So, for our closing panel, we stopped asking *whether* agents can buy - they clearly can and do. Instead, we asked four of the people building the stack what it takes for every merchant to let one. More importantly: when will the dam break and we see millions of agents treated as first class consumers across the web? Kathleen Peters Chief Innovation officer at Experian, Craig Gilbert VP of Identity Solutions at Mastercard, Gareth Davies CPO of Auth0 from Okta, and our CEO Amir Sarhangi, moderated by Danny Fortson of The Sunday Times. Their insights are worth a read and listen for anyone building agents, but struggling with identity and payment. The panel touched on not just the opportunity, but the roadblocks to agentic commerce - while all agreed on the end goal, there was also healthy debate on the finer points of how we get there: - Agent traffic has crossed 57% of what the largest networks see, and most of it isn't fraud. It's pent-up demand. - The real debate: do agents need new protocols, or do they just use the web that already exists? - Why liability still traces back to a human, because there's nothing at stake for an LLM but the next token. Panel video link in the comments. #agenticcommerce #AIagents #payments #digitalidentity