The AI Economy: How AI Is Reshaping Commerce in 2026

The AI economy became measurable in 2025. IDC counted $318 billion in AI infrastructure spending for the year — more than double 2024's $153 billion — and forecasts $487 billion for 2026, on a path that crosses $1 trillion annually by 2029. Few enterprise technology cycles have compounded that fast.

The composition of the spending matters more than its size. Gartner expects 40% of enterprise applications to include task-specific AI agents by the end of 2026, up from under 5% in 2025. That is a shift from software that answers to software that acts — agents that hold a task, evaluate options, and execute.

An economy is defined by its actors, and AI systems are becoming actors rather than tools. When software can compare offers and complete a purchase, it stops being a productivity layer and becomes a participant in markets. Merchants meet that participant earlier than almost anyone else, because shopping is the most natural first delegation.

The 2025 holiday season was the first real measurement of AI's demand side. Salesforce tied $67 billion of a record $336.6 billion in Cyber Week spend to AI and agents — 20% of all global orders — while Adobe recorded $257.8 billion in U.S. holiday online spend with generative-AI-driven traffic up 693.4% year over year.

That layer sits on top of the fastest-growing slice of retail. The U.S. Census Bureau put e-commerce at 16.9% of total retail sales in Q1 2026 — $326.7 billion, growing 9.8% year over year against 3.9% for retail overall. AI is not creating a new market so much as re-routing discovery inside the one that already wins.

The re-routing is the point. Shoppers increasingly ask an assistant instead of typing into a search box; the assistant compares products across merchants and returns a shortlist. Your product gets evaluated in that comparison whether or not anyone ever opens your storefront — which means visibility now happens in a place your analytics were never built to see.

For a year, the safe dismissal was that AI traffic browses but does not buy. That flipped. Adobe data reported by TechCrunch shows AI-referred visitors converted 38% worse than average in March 2025 — and 42% better by March 2026, with 37% higher revenue per visit, while Q1 AI-sourced retail traffic grew another 393% year over year.

Consumers explain the flip. A survey of 1,009 U.S. consumers by Exploding Topics found 77.6% used AI for shopping in the past six months and 43.2% use AI shopping tools at least weekly. Adobe's own consumer survey adds that 85% of consumers who shopped with AI said it improved the experience.

The mechanics are simple: the assistant does the comparison before the click. A visitor who arrives from an AI referral has already narrowed the field, checked the specs, and often compared prices — so they convert like someone at the end of a funnel, because they are. The uncomfortable corollary is that losing happens upstream and invisibly: the merchant who loses the comparison never sees the session at all.

Commerce shifts look slow year to year and decisive in hindsight. U.S. e-commerce took roughly nine years to go from about 8% of retail in early 2017 to the Census Bureau's 16.9% in Q1 2026. Mobile crossed quietly too: Adobe counted 56.4% of holiday 2025 online transactions on smartphones, up from 54.5% a year earlier.

No merchant experienced either shift as a single dramatic quarter. Share moved a point or two a year, the laggards lost gradually, and by the time the trend was undeniable, the winners had already spent years restructuring — mobile-first sites, warehouse networks, marketplace operations. The advantage never came from reacting to the share number; it came from rebuilding before the share number made the case.

AI-mediated commerce is at the start of that curve with one difference worth respecting: switching costs. Moving from stores to websites required trust in online payment; moving to mobile required new devices. Moving from search to an assistant requires one sentence. Adoption curves built on zero switching cost can outrun their historical analogies — 693% growth in a single holiday season suggests this one might.

In the four months between September 2025 and January 2026, every major AI surface gained a path to checkout. Stripe and OpenAI put Instant Checkout inside ChatGPT, Google shipped its Agent Payments Protocol, and Shopify and Google launched the Universal Commerce Protocol at NRF 2026 — letting merchants sell inside ChatGPT, Microsoft Copilot, Google AI Mode, and the Gemini app. Perplexity relaunched agentic shopping free for U.S. users with PayPal handling settlement.

What none of these announcements included is a transaction-volume figure — Forrester notes that native agentic transactions remain largely unquantified. Rails before volume is exactly how infrastructure shifts look from the inside, and it is the reason the merchant window exists at all. For the protocol-by-protocol breakdown — identity, mandate, checkout, settlement, and where stablecoins fit — see our , and test your own agent surface with the .

The sharpest skepticism comes from people building the infrastructure. Adyen's head of agentic commerce told PYMNTS the market sits at "version 0.5" on a five-point maturity scale. Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027 and estimates only about 130 of the thousands of self-described agentic vendors are real.

Consumers draw the same line. In the Exploding Topics survey, 31.2% of consumers would not authorize an AI agent to spend any money autonomously, and only 11.7% would approve purchases over $100. Forrester's mid-2026 assessment matches: true autonomy is rare, and most value from answer engines today comes from comparison and guidance rather than native transactions.

A shift this early rewards preparation over spend. The demand side is measurable today — AI-referred traffic converting 42% better — while autonomous checkout matures, which leaves a window where becoming machine-readable is cheap and losing comparisons is still invisible in your dashboards.

Preparation is three moves. First, measure: add server-side detection so AI agent visits and AI-referred sessions show up as their own channel — the tracking guide walks through it. Second, structure: Adyen estimates agents need roughly three times more product attributes than a traditional listing, so audit titles, variants, offers, and policies against the structured data checklist. Third, simulate: run your catalog through the models doing the comparing — GPT-4, Claude, Gemini, and Perplexity — and see whether you win or lose against competitors before real agent volume decides it for you. A shows exactly that.

What is the AI economy? The measurable economic activity created by AI systems acting as market participants — the infrastructure buildout (IDC counted $318 billion in 2025 spending), the enterprise agents executing tasks, and the commerce those agents increasingly influence.

How much commerce does AI actually influence today? Salesforce tied $67 billion — 20% of global orders — to AI and agents during Cyber Week 2025, and Adobe data showed AI-referred retail traffic converting 42% better than other traffic by March 2026. Native agent-completed checkout volume remains unquantified.

As autonomous checkout, largely yes — expects over 40% of agentic AI projects to be canceled by end of 2027 and calls the market "version 0.5." As a discovery and comparison shift, no: the traffic, conversion, and order-influence numbers are already measured.