Agentic Procurement Reality Check: The $15T Forecasts vs. the "Agent-Assisted" Present
Update (October 5, 2026): The prior revision covered Gartner's agentic commerce framework (90% of B2B buying AI-agent-intermediated by 2028) and Vertice's buy-side negotiation agent Ana. This cycle adds the strongest reality-check evidence yet against the aggressive end of that trajectory — useful counterweight for founders calibrating how much roadmap to bet on autonomous buying.
The forecast side (what the bulls project)
Mastercard's agentic commerce trust white paper (published Sept 30, 2026) repeats the headline numbers: "ICSC and McKinsey project that agentic commerce in US consumer retail could reach $1 trillion in revenue by 2030, and Gartner expects AI agents to intermediate more than $15 trillion in B2B spending by 2028." McKinsey separately sizes agentic commerce as a $3T–$5T global market by 2030. Google announced its Universal Commerce Protocol in January 2026 with Walmart, Target, and Shopify aboard — part of what CMI calls "a growing alphabet soup of agent plumbing (UCP, AP2, A2A, MCP)."
The reality side (what actually happened this quarter)
- The Register (Oct 1, 2026) — "The tech industry is hot for shopping bots, but they're years away." Amazon blocked Meta's Muse shopping bot (after previously blocking Perplexity's bot and winning — then losing on appeal — a preliminary injunction). At The AI Conference in San Francisco, Lindsay Walker (product manager, Hedera AI Studio) was blunt: "Agentic commerce is not a reality yet... There are a lot of products out there claiming that they're creating agent wallets or agent cards. But when it comes down to it, it's not functional yet and hasn't truly emerged." Her demo failed on mundane mechanics — the agent couldn't log in, couldn't access the card-number DOM element, and tax/shipping logic broke on location data. Her thesis: "Money is too important. Funds are too important. Value is too important. You need to have deterministic gates that can't be bypassed by a probabilistic model." And the kicker: "The reality right now is we're simply working with agent-assisted shopping. There's no true agentic autonomy going on. And it's very different than autonomous agentic commerce. The human is still there at the moment of purchase."
- Content Marketing Institute (Sept 29, 2026) — "The Robot Buyer Isn't Here Yet." OpenAI quietly shut down ChatGPT's "Buy It" feature after "fewer than 30 of Shopify's millions of merchants had ever gone live with it" (per Fast Company). On the B2B side, CMI argues the agentic-buyer-meets-agentic-seller vision rests on assumptions nobody in B2B has observed: "That's not a throughput problem, and no AI tool fixes that because an agent cannot get a CFO to consensus." Per McKinsey's 2026 B2B Pulse Survey (cited in the piece), the #1 reason buyers switch suppliers is inconsistent information across the vendor's teams — a message-consistency problem, not a checkout-speed problem. And the trust math: "Though just 6% of marketers consistently accept their ad platform's AI recommendations, the vision assumes buyers will let an AI agent shortlist a six-figure software vendor." StackAdapt's AI Delegation Gap study (Aug 2026) found 90% of marketers use AI but only about half are comfortable letting it act autonomously.
- Regulatory friction — the FTC sought comment in August 2026 on an enforcement policy statement regarding AI-driven personalized pricing, a live constraint on agent-mediated price discovery.
What it means
The honest 2026 synthesis for the Agentic Procurement Reality Check: The $15T Forecasts vs. the "Agent-Assisted" Present throughline: protocols and money are real; autonomy is not. The near-term state is agent-assisted research and shortlisting (see AEO → AXO → Agentic Shortlisting: Machine-Readable Vendor Data Is Now a Revenue Variable) with humans making purchase decisions — especially for six-figure enterprise contracts where, as CMI notes, the binding constraint is 13 humans reaching consensus, not checkout throughput. Founders should build for agent readability now (structured data, citations, deterministic policy gates) without rebuilding GTM around a robot buyer that, per Walker, is ~five years from functional. This also explains why the governance-first evaluation criteria in Agentic-Washing Meets the Governability Gap: RFPs Now Score Policy Enforcement, Audit Trails, and Cost Predictability are winning in RFPs: buyers are being asked to trust agents they demonstrably don't yet trust.