Agent-ready: what agentic commerce means for brands, distributors and retailers
AI agents have started shopping. Not at scale yet: UK and US merchants surveyed by Checkout.com say about 3% of their transactions involve an agent. But the infrastructure arrived in a rush this year, and the businesses agents recommend in 2027 will be the ones fixing their data and operations now. Here is what changed, who it affects, and what to do in the next 90 days.
What happened in 2026
On 15 October, Groupon relaunched LivingSocial as an agent-first marketplace. Any AI agent can search and compare more than 50,000 US local experiences, build a cart and hand it to its owner, who pays. Payment networks, wallets and AI platforms spent the year building the same plumbing from the other side. It comes in four layers.
- Discovery: agents find, compare and recommend. Built by Google's Universal Commerce Protocol, the Agentic Commerce Protocol from OpenAI and Stripe, and marketplaces like LivingSocial. It asks you for accurate, structured product, price and stock data.
- Payment: agents pay with scoped, revocable credentials. Built by Visa, Mastercard and Amex agentic tokens, and by Stripe Link, Shop Pay and PayPal. It asks you for a payment gateway that accepts tokenised agent payments.
- Trust: identifying agents and settling their mistakes. Built by Visa's agent scoring, Amex Agent Purchase Protection and fraud vendors. It asks you for clear policies, and fraud rules that let legitimate agents through.
- Conversation: agents serving customers where they already are. Meta Business Agent on WhatsApp, Messenger and Instagram. It asks you for your own agent on the channels your customers use.
The gaps matter as much as the launches. Nobody has settled who pays when an agent buys the wrong thing. Some large merchants are blocking agents in the meantime: Amazon shut out Meta's new Muse assistant in September. Shoppers are cautious too: Worldpay finds 59% of those open to AI shopping still want to approve every purchase. And only 11% of small businesses are agent-ready, according to PYMNTS and Visa.
What changes for each player
- Retailers: the shelf becomes a data feed. Banners, end-caps and retail media are invisible to an agent. First move: ask four AI assistants to recommend and buy your top 20 products, and log every error.
- Brands: loyalty shifts to agent preference. Specs, reviews and price history decide, not the TV ad. First move: keep one source of product truth and syndicate it everywhere agents read.
- Distributors: reorder agents compare distributors instantly, squeezing anyone who only moves boxes. First move: put an agent on order capture, stock and credit checks on WhatsApp.
- Suppliers: procurement agents ask for specs, minimum orders and lead times as data. First move: publish specs, certifications and lead times in structured form, not PDFs.
For customers, the shift is from searching to delegating. Expect "agent finds, human approves" first, then standing mandates for repeat purchases such as groceries, refills and subscriptions.
Why Africa will move differently
South African online retail is heading for about R159 billion this year, roughly 10% of all retail, and 57.9% of online shoppers buy on a smartphone (World Wide Worx). Commerce here already runs through WhatsApp, and across much of the continent through mobile money, so agentic commerce will arrive through chat before it arrives through a browser agent.
Payments will differ too. Card networks' agentic tokens assume a card, while much of the continent pays by mobile money, instant EFT or cash on delivery. And the biggest early prize is probably B2B: spaza shops, taverns and independents reordering from distributors, where an agent can take the order, check credit and book the delivery.
The 10-point agent-readiness checklist
- Ask ChatGPT, Gemini, Claude and Perplexity to recommend and buy your best-selling product. Record every error.
- Complete your product data: titles, specs, sizes, images, barcodes, price and stock.
- Publish live stock and real delivery dates, not "usually in stock".
- Write returns, refunds and cancellations in plain language an agent can quote back.
- Keep reviews current and verifiable.
- Ask your payment gateway which agent payment options it supports today.
- Decide who carries the cost of an agent's mistake and put it in your terms.
- Tag agent traffic in analytics and fraud rules so good agents aren't blocked.
- Put an agent on your own front line: WhatsApp orders, service queries and reorders, with a person approving what goes out.
- Track the share of enquiries and orders that arrive through agents, every month.
Points 2, 3, 4 and 9 pay off now with human customers, whether or not agents arrive on schedule.
How we help
We install working AI systems inside founder-led businesses: order intake, customer service, reporting. Every system has a person approving what goes out, and the number it has to move is written down before day one. Our first deployment runs inside a multi-million-dollar e-commerce marketplace and went from spec to live in under seven days.
On Proof Sprints and Core Builds: measurable ROI in 90 days — or 50% of your project fee back.
If you want to know where you stand, the free 15-minute audit is where every Automaiq engagement starts. No pitch.
Sources
- Groupon: LivingSocial relaunch, 15 Oct 2026
- Checkout.com: Agentic Commerce 2026
- Worldpay: Agentic Commerce Report 2026
- PYMNTS and Visa: small merchants and AI shopping
- World Wide Worx: Online Retail in South Africa 2026
- Retail Gazette: Amazon blocks Meta's Muse
- Google: Universal Commerce Protocol
- Visa: agentic commerce announcements, Jun 2026
- American Express: Business Playbook for Agentic Commerce
- Bloomberg Law: Meta Business Agent