Analytics6 min read

AI Commerce in 2025: The Year Agents Became Real

The infrastructure for agentic commerce arrived in 2025. Payment rails, checkout integrations, holiday-season data — all of it landed at once. The harder question is why consumer behaviour hasn't caught up yet.

Sarah Chen

Senior Editor

—22 December 2025

In December 2024, I wrote that the real lesson of that year was that AI in commerce isn't a capability you buy. It's something you build, incrementally, on top of foundations that take longer than anyone admits. I also said I expected the 2025 version of this piece to spend more time on agents.

Well. Here we are.

The thing that actually happened

Retail Brew called 2025 "the birth of agentic commerce", and while I have some scepticism about birth metaphors for software capabilities, the framing is broadly right. Several things that were speculative a year ago are now real.

ChatGPT Instant Checkout launched on 29 September, with Etsy first and Shopify merchants following. Visa, Mastercard, PayPal and Stripe all shipped agentic payments infrastructure (the tokenisation and fraud-detection rails that AI agents need to transact safely) across the same calendar year. Walmart made its deal with OpenAI in October. The holiday-season data landed too: AI sources drove 693% more traffic to US retail sites than a year earlier, per Adobe Analytics, converting 31% better than other traffic across the full season.

The agentic layer is present. It is not yet mainstream. That gap is worth being honest about.

The infrastructure vs. adoption problem

Something I noticed writing about these stories through the year: there is a consistent lag between when infrastructure ships and when consumer behaviour changes to use it.

ChatGPT Instant Checkout launched in late September. By March 2026, Walmart had pulled the plug on OpenAI's version entirely, switching to embedding its own Sparky chatbot directly inside ChatGPT and Google Gemini instead. The reason wasn't ideological. It was conversion data. A Walmart executive told WIRED that sales through Instant Checkout were "disappointing", with conversion rates running at roughly one-third of what Walmart sees on its own site. Customers didn't want each item triggering a separate checkout; they wanted a basket and one transaction, the same way they've always shopped online.

That's not a failure of AI. It's a discovery about consumer behaviour, and an important one. Infrastructure being available doesn't change how people act. Trust in AI agents completing purchases on your behalf requires something different from trust in clicking "buy now", and that kind of trust builds slowly, by category, one successful transaction at a time.

The retailers best positioned for 2025 were mostly those who started building in 2022 and 2023. Those who started in earnest in 2024 are largely still finding out whether their foundations were solid enough.

The UK picture is sobering

The 77% of UK ecommerce retailers admitting AI initiatives are falling short is the finding from 2025 I keep returning to. The £92 million of investment considered at risk. The enterprise/mid-market/SME adoption gap that keeps widening, as larger retailers compound their advantage through compounding data and compounding capability.

None of that contradicts the more optimistic global narrative. It suggests, rather, that the gap between what AI commerce can do and what most retailers are actually achieving with it is larger and more persistent than the trade press implies. The headline numbers come from Walmart, Amazon and Salesforce Commerce Cloud implementations. The median experience is different.

Worth noting: the Adobe holiday-season data above is US-only. Traffic and conversion figures from UK retailers remain harder to benchmark publicly. The directional story probably holds, but the specific numbers do not automatically transfer.

What changed that actually matters

The payment infrastructure. Visa, Mastercard, PayPal, Stripe and OpenAI all building agentic payment rails in the same calendar year isn't a coincidence. It's the foundations for a different kind of commerce. That infrastructure will still be there in five years. The specific products built on top of it will change considerably, as Walmart's pivot already demonstrates.

The conversion quality signal. AI-referred traffic outperforming other channels in quality metrics is a signal that will drive investment decisions. When finance teams can see AI traffic converting 31% better across a full season, or 54% better on a single peak day like Thanksgiving (US data, Adobe Analytics), the budget discussions change. The basket-building behaviour matters too: shoppers arriving from AI assistants spend 45% longer on-site, per the same Adobe data.

The consumer trust question getting surfaced clearly. The Walmart checkout pivot, the periodic backlash against algorithmic pricing, the emerging body of consumer research on attitudes to AI agents: 2025 was the year the trust question moved from theoretical to empirical. The answers so far: trust is real but category-dependent, it builds slowly, and "the AI did it" is not an acceptable answer when something goes wrong.

What I'm watching in 2026

The structural pressure Digital Commerce 360 described as a "reckoning" (AI-powered shopping, zero-click search, agentic purchasing, and tariff disruption all hitting simultaneously) feels like a genuinely different operating environment rather than an incremental one. What I don't yet know is whether that resolves into a new stable state, or whether it's the start of a longer period of structural instability.

Also: retailer pushback. Concerns about margin erosion from AI-negotiated deals, loss of direct customer relationships, and platform dependency are starting to surface publicly rather than staying in private conversations. That dynamic will be worth watching closely.

A year ago I predicted agents would dominate 2025. They did, as a topic. The question for 2026 is whether they dominate as a reality: whether consumer purchasing behaviour actually shifts in measurable ways, not just as a share of venture capital and trade press column inches.

I genuinely don't know. That's either a very honest answer or an insufficiently bold take, depending on your temperament.

See you on the other side of NRF 2026.

Data sources and further reading

AI Commerce in 2025: The Year Agents Became Real

Sarah Chen

Senior Editor

—22 December 2025

Well. Here we are.

The thing that actually happened

The agentic layer is present. It is not yet mainstream. That gap is worth being honest about.

The infrastructure vs. adoption problem

Something I noticed writing about these stories through the year: there is a consistent lag between when infrastructure ships and when consumer behaviour changes to use it.

The UK picture is sobering

What changed that actually matters

What I'm watching in 2026

I genuinely don't know. That's either a very honest answer or an insufficiently bold take, depending on your temperament.

See you on the other side of NRF 2026.