Llama Talk – Sept. ’26

1702059841openai-icon

The Agent Correction

Agentic checkout failed its first real test. Your buyers adopted AI anyway, they just did it somewhere else. Here is where the competition actually moved, and what to fix this quarter.

The prediction everyone repeated has now been tested.

For most of 2025 the pitch was the same: AI agents would soon buy on your customers’ behalf, and you needed to be ready for the checkout. That prediction has now run its first real experiment, and it did not go the way the slide decks promised.

In March 2026, OpenAI discontinued Instant Checkout (the flagship agentic-checkout product it had launched with Stripe six months earlier) and moved purchasing toward apps, with merchants completing the transaction in their own environment. Walmart had already withdrawn, and the reason it gave is important: agent-driven traffic converted at roughly one-third the rate of ordinary website visits.

The rest of the market did not rush in behind it. Flagship Advisory Partners tracks the ten largest companies by revenue across verticals in the US and EU – which is 80 merchants in total. In its August 2026 update, 4 of those 80 supported agent-led checkout, all US-based, and not one additional brand had declared support since December. AI-referred traffic is at record highs and still accounts for under 1% of total retail visits.

So the practical read for a B2B seller is this: if someone is selling you an agentic commerce project on urgency, they are selling you last year’s story. Agent-executed checkout is not your near-term problem.

It is also not the thing that changed.

Your buyers moved – upstream, where you are not measuring.

The adoption absolutely happened. It happened on the buy side, and it happened before the transaction. Deloitte Digital surveyed 530 US B2B buyers and 530 US B2B suppliers (published February 2026): 61% of buyers report using AI in purchasing and 38% report using agentic AI. Among suppliers, it is 45% and 24%. Your customers are meaningfully ahead of you, and the gap is widest on autonomous tooling.

If those agents are not checking out, what are they doing? Research. Specification matching. Comparison against approved alternates. Assembling the requisition. Then a human places the order in your portal, through your rep, or over EDI. They do it inside the systems where negotiated pricing, credit terms, tax exemption, freight, and approval workflow really live.

That division of labor is not a transitional phase on the way to full automation. For B2B it is the reasonable end state, because those financial and operational controls are in your systems deliberately, and no serious procurement organization wants an external chatbot routing around them.

But it does relocate the competition, and that should get your attention. You do not lose an agent-assisted deal at checkout. You lose it at the shortlist – before there is a session, a quote request, or any analytics event that tells you it happened. If a machine cannot read your catalog, cannot match a manufacturer part number, cannot retrieve a current price and stock position, or cannot parse a specification that exists only inside a PDF, you are not in the consideration set. The competitor whose data is legible is.

One encouraging number, from a reliable source: Adobe Digital Insights finds AI-referred retail visitors now generate 53% more revenue than visitors from other sources – which is a reversal from last year, when they were worth about half as much. Small volume with high intent. The buyer arrives already qualified, or does not arrive at all.

The Investment Case that Requires No Prediction

Here is why we are raising this now rather than waiting for agentic commerce to prove itself: every item on the agent-readiness list is work you need for reasons that have nothing to do with agents.

Canonical identifiers and complete structured attributes fix on-site search and faceted navigation – which is where most of your existing revenue is quietly lost today. Real-time price and availability APIs are the same endpoints your reps, your punchout customers, and your marketplace listings already need. Entitlement resolution – one authoritative answer to “what can this account buy, at what price, on what terms” – is the hardest of these and the most valuable, and it is what makes self-service genuinely work for contract customers. None of that investment strands if agents underdeliver.

The timing is also unusually favorable. 87% of suppliers in the Deloitte survey are upgrading or planning to upgrade their ERP. If you are one of them, the data-quality and integration work is already funded and scheduled, and the marginal cost of doing it in a machine-legible way is small. Doing it as a standalone project in 2028 will not be small.

What it is worth, in Deloitte’s terms: suppliers it classed as digitally mature beat their annual sales goals by a 110% greater margin than low-maturity peers, averaging 6.1% revenue growth against 2.9%. And the channel shift is not a rising-tide story – US B2B ecommerce grew 13% to $2.93 trillion in 2025 while total US B2B sales grew 0.4% (Digital Commerce 360). Digital is taking share from other channels, not riding growth.

What to Do This Quarter

Next 30 days – find out what machines say about you

  • Ask the major assistants your top ten category and specification questions, the way a buyer would phrase them. Record whether you appear, whether the specs are right, and who appears instead. Treat wrong specifications as a higher-priority defect than absence.
  • Check whether you are blocking AI crawlers. Many teams added those blocks in 2024 and 2025 and never revisited them. Review robots.txt and any bot-management rules at the CDN or WAF.
  • Segment AI referrals in analytics so you have a baseline before you change anything. You cannot show progress on a channel you are not isolating.

 

Next 60 days – fix the machine-readable layer

  • Put canonical identifiers on every SKU: manufacturer part number, GTIN where one exists, and brand. Agents match on identifiers before they will trust a product record.
  • Get specifications out of PDFs and into structured attributes. A datasheet a human can read is invisible to everything else in the stack, including your own search.
  • Add product structured data with price and availability, and make availability resolve in real time rather than from a nightly cache.
  • Separate two kinds of truth: the public catalog (what exists, at list price) and the entitled view (what this account may buy, at its contract price, under its terms). An authenticated request should resolve to one authoritative commercial state and not a generic list price the buyer will never pay.

 

Next 90 days – set your posture in writing

 

Tariffs: Your Landed-cost Model is Probably Out of Date

Current as of September 3, 2026. The 10% blanket tariff imposed in February under Section 122 expired on July 24 by operation of law, after the Supreme Court struck down the IEEPA tariffs in February. What replaced it the same day is a set of Section 301 forced-labor duties covering 60 economies at 10% or 12.5%, stacking on top of MFN rates and existing Section 232 and Section 301 duties, with large product-level exemption annexes.

Separately, the $800 de minimis exemption remains suspended — and since a CBP rule effective June 24, it is suspended indefinitely for every mode except international post. Statutory repeal lands July 1, 2027 regardless of how the litigation resolves.

Two actions: confirm your landed-cost model reflects the post-July structure rather than February’s, and check whether your contract pricing carries a duty pass-through clause. Given the pace of the litigation, re-verify before you quote anything long-dated.

Make Shipping the Easy Part of Your Migration with ShipperHQ

If you’re migrating, you don’t have to rebuild your shipping setup from scratch. ShipperHQ can help you set up or migrate your shipping and checkout logic, so you can replatform without having to rebuild it.

Platform migration is enough of a headache on its own. If a move is on your radar, even six months out, get ahead of it: schedule a free call with a ShipperHQ migration specialist and let shipping be the easy part.

Learn how ShipperHQ can support your migration.

Share it

Related Posts

Adobe Commerce

Other Industries

Llama Talk – Sept. ’26

Adobe Commerce

Other Industries

Llama Talk – June ’26

Adobe Commerce

Other Industries

Llama Talk – May ’26
See All Posts