SharkNinja's AI agents now handle about 20,000 customer chats a week, covering both pre- and post-purchase support, while Perform.AI launched a new operating system built on the premise that delivery and returns performance, not advertising, will determine which merchants AI shopping agents recommend. The moves, reported between September 25 and 27, show brands and merchants rebuilding operations to serve AI agents that increasingly open their own purchase channels.
Customer service agents move beyond the buy button
SharkNinja rebuilt its commerce platform on Salesforce in nine months, going live in the US and Canada before a global rollout. CIO Velia Carboni told diginomica that Agentforce, Salesforce's AI agent product, now handles roughly 20,000 customer chats each week. Scanning a QR code on a product opens a chat that walks the customer through setup, and the AI can look up replacement parts such as brushes and filters.
Carboni highlighted the "unboxing agent," designed to enroll customers who bought through retailers, Amazon or TikTok Shop as direct members of the SharkNinja family. "Many brands sell mostly through other companies' channels and struggle to build direct customer relationships," the company's approach shows. Salesforce's published case study puts the conversion rate of shopping-enabled chats at 11%, though it does not define how that figure is calculated.
GoKwik, an Indian e-commerce technology company, published six months of data from 30 D2C brands showing that adding AI voice calls to WhatsApp cart recovery campaigns doubled recoveries. Average call answer rates were 60 to 65 percent. Customers reached through both AI calls and WhatsApp were 2.5 times more likely to complete their purchase than those contacted through only one channel. Co-founder and CEO Chirag Taneja said "carts are not always abandoned because of weak intent or pricing," and that in many cases a shopper's question has gone unanswered. All figures are GoKwik's own data without third-party verification.
Delivery data becomes the new storefront
Delivery tracking company Parcel Perform renamed itself Perform.AI and announced an "AI Commerce Operating System" on September 22. The company's core argument: AI agents helping consumers shop assess products on price and delivery speed, and recommend only brands whose delivery record matches what they promise. "Neither loyalty nor advertising influences an agent, but consistent performance does," the company said in its framing.
The system includes modules for visibility, checkout, post-purchase, and returns, plus a logistics layer. A component called AI Decision Intelligence runs across the system, reading incoming data and acting on it with gross margin as the optimization target. Marketing, logistics and customer service teams can connect to their e-commerce operations data through the Model Context Protocol (MCP), querying it from the AI tools they already use. The practical takeaway: delivery and returns records are now something merchants need to audit if they want AI agents to pick them.
SureDone introduced Taska AI, an agent that handles marketplace listing work, on September 23. Sellers can ask it to examine products, improve titles and descriptions, prepare channel-specific fields, and create scoped listing plans for review. A distinctive feature is that it distinguishes submission from a verified marketplace result, surfacing rejections and suggesting the next step. It launches with a review-first experience, with plans to expand customer-controlled automation over time.
Payments infrastructure prepares for machines
Block, the operator of Square and Cash App, joined the x402 Foundation on September 24. The x402 standard uses the web's "402 Payment Required" status so AI agents can pay for APIs and data on the spot. Block will contribute to working groups and continue supporting Bitcoin Lightning for x402. The company has also joined the Universal Commerce Protocol (UCP), covering both product checkout and small machine-to-machine payments with separate standards.
BlackRock published research in mid-September arguing that AI agents could become a new source of demand for stablecoins as software begins paying for data, services, and computing power without direct human involvement. The paper estimates adjusted stablecoin transaction volume exceeded $11 trillion in 2025 and market capitalization passed $300 billion by September 2026. BlackRock itself cautions that AI-agent payments remain at an early stage.
Andrew Shikiar, CEO of the FIDO Alliance, told Payments Dive that verifying not only the agent but also the individual behind it is critical. With AI fueling credential theft and deepfakes, he argues services need signals that give them confidence the person is who they say they are for every transaction. He does not expect agentic commerce to happen at scale for years.
Why this matters for customer support, operations, and sales teams
The week's developments point to a concrete shift: AI agents are now handling work on both sides of a transaction. For customer support teams, SharkNinja's 20,000 chats per week and GoKwik's voice recovery data show that post-purchase touchpoints are being automated at scale, not just triaged. For operations and logistics professionals, Perform.AI's thesis means delivery and returns performance data will directly affect whether AI shopping agents surface a brand's products-making ops metrics a revenue factor, not just a cost center. For sales and marketing teams, the spread of standards like x402 and UCP signals that machine-to-machine payments and agent-driven checkout are moving from experiment to infrastructure, even if full adoption is years away.
Your membership also unlocks:
