Is Your Magento Agency Using AI Where It Actually Works?

Originally published October 30, 2024. Substantially updated July 2026 to reflect the current Adobe Commerce AI stack, agentic commerce standards, and what we’ve learned from B2B implementation work.

Every Magento agency now says it “does AI.” The claim costs nothing, so it appears on every services page.

What separates a capable Magento agency from one repeating vendor marketing is where the AI is applied. The high-value work in 2026 is not bolting a chatbot onto a storefront. It’s making your product data accurate, machine-readable, and searchable — for human buyers today and AI agents tomorrow.

This article covers what actually changed since 2024, which AI capabilities ship in Adobe Commerce right now, where AI delivers measurable results in real projects, and the questions that quickly reveal whether an agency can do this work.

What changed since 2024

The AI-in-commerce conversation moved fast. Much of what the industry (Adobe included) promoted in 2024 has been replaced, and some of it discontinued — Adobe Aero, widely recommended for AR shopping experiences, was shut down in late 2025. That turnover is the first lesson: evaluate an agency on what it builds, not on which tool logos it lists.

2024 assumption	2026 reality
AI = recommendations and chatbots on your storefront	AI = the front door: buyers start product research in ChatGPT and Gemini
Humans place every order	Orders come from humans, AI-assisted humans, and increasingly agents acting on a buyer’s behalf
“Adobe Sensei” as the umbrella brand	Named capabilities: Live Search semantic search, Product Recommendations, Commerce MCP server
AR/VR as the next commerce frontier	Adobe Aero discontinued in 2025; agentic commerce standards took its place
Catalog quality as a back-office concern	Catalog quality as the deciding factor in whether AI can sell your products at all

The shift that matters most to a merchant’s P&L: discovery is leaving the storefront. McKinsey’s 2026 Global B2B Pulse survey of nearly 4,000 decision-makers found that generative AI has risen into the top five channels B2B buyers use for supplier discovery and evaluation. And Adobe’s January 2026 data shows AI-referred shoppers convert 31% higher and generate 254% more revenue per visit than traditional traffic. If AI assistants can’t read your catalog, you’re not in the consideration set.

The AI that ships in Adobe Commerce today

Here is the current, verifiable stack — no roadmap items presented as products.

Capability	What it does	Status (July 2026)
Live Search semantic search	Resolves buyer intent instead of matching keywords	Generally available since June 2026
Intelligent ranking configuration	Merchant control over how and why results rank	Public beta
Product Recommendations	Behavioral recommendation units	Production-stable for years; widely under-configured
Agentic protocol support (UCP, ACP)	Makes catalog, pricing, and inventory readable and transactable by AI agents (Google and OpenAI standards)	Committed; rolling out through 2026
Commerce MCP server	Permissioned real-time agent access to catalog, cart, pricing, inventory, checkout	Introduced at Summit 2026
LLM Optimizer	Measures and shapes how your products get cited in AI-driven discovery	Available

A competent Magento agency should be able to configure all of this. But configuration is table stakes. The differentiating work sits underneath.

Where buyers feel AI first: on-site search

Before any agent buys anything, your human buyers are typing into your search box. In B2B, that’s where AI produces the fastest measurable wins.

Search improvement	How it works	Business outcome
Customer part-number indexing	Index the buyer’s own SKUs for logged-in portal users	Self-service replaces “call Dave, he knows our number”; fewer support tickets
Zero-results elimination	Map industry slang to catalog terms; semantic search covers the long tail	Recovered sessions that previously ended at a dead end
Keyword-to-cart models	Learn from what buyers do after searching; auto-link discontinued SKUs to replacements	Captured revenue with no manual spreadsheet maintenance
Search analytics as procurement intel	Zero-results reports show demand you don’t stock	A ranked list of products worth adding

None of this requires exotic technology. It requires understanding B2B buying behavior — and doing the data work underneath.

AI in the ERP: your system of record becomes something AI can talk to

The ERP has always been the source of transactional truth — pricing, inventory, order history. What’s new in 2026 is that AI can now read it directly. Through MCP-style interfaces like the one built into our Sirius platform, an AI assistant can query live ERP data — stock levels, contract pricing, a customer’s order history — with the same permissions and guardrails as any other integration. No nightly export, no stale copy of the data.

That unlocks practical use cases that were previously manual:

• Replenishment suggestions from order history. AI reads a customer’s past ERP orders and proposes a season-appropriate reorder list — the digital version of a counter rep who remembers what you buy every spring.
• Business-aware search ranking. Feeding ERP margin and inventory data into search means results can favor high-margin items or stock you need to move, not just textual relevance.
• Answering operational questions in plain language. “What’s on backorder for this account?” becomes a query an assistant can answer from live ERP data instead of a ticket for someone in customer service.

The caution: AI should read from the ERP, never around it. An assistant working from a cached or partial copy of your transactional data will confidently quote prices and stock levels that are no longer true. The connection has to be live and permissioned — which is an integration problem before it’s an AI problem.

The data work that makes AI trustworthy

Your search is only as smart as your data, and so is every capability above. In a typical B2B setup, the ERP holds transactional truth and the PIM holds marketing truth; search and AI both depend on the two feeding one clean, unified index.

The industry numbers back this up. Gartner predicts that through 2026, organizations will abandon 60% of AI projects that aren’t supported by AI-ready data — and 63% of organizations surveyed either lack the right data management practices for AI or aren’t sure they have them. The model is rarely the problem. The data underneath it is.

This is where AI earns its keep in real projects. Three examples from our recent B2B integration work — the pattern behind our Sirius integration platform, documented in our ERP integration case study:

Mapping legacy ERP data to a clean schema. A typical industrial catalog is a 20-year accumulation of fields named ATTR_47 and SPEC_OLD, half of them populated. Classification pipelines read the existing values, descriptions, and spec sheets, then propose mappings into a clean attribute model. Every mapping gets a confidence score: high-confidence auto-applies with spot checks, mid-confidence goes to a merchandiser review queue, low-confidence stays manual. In a recent project, roughly three-quarters of a five-figure SKU catalog cleared the auto-apply threshold — six weeks of review work instead of a multi-year manual backlog.

Enriching product content without inventing it. Thousands of B2B SKUs carry descriptions like 1/2″ SS HX NPLE 304 SCH40 — legible to a procurement engineer, invisible to search, useless to an AI agent. Generation from the cleaned attribute set produces titles, buyer-facing descriptions, and spec content. One hard rule makes it safe: the model only describes what exists in the structured data. No pressure rating in the source, no pressure rating in the copy. That constraint is the difference between enrichment and fabrication.

Cross-system product matching. When one physical part lives in the ERP, the B2B portal, and a Shopify store under three SKU formats, string matching breaks and fuzzy matching confuses stainless with brass. Embedding-based matching clusters products across systems and surfaces likely matches for human confirmation. The output: one canonical record per product, mapped everywhere it sells.

Integration first, intelligence second

The order matters. AI applied to broken data doesn’t expose the problems — it covers them with confident-sounding output. Polished descriptions for products whose stock numbers are wrong. Clean titles at prices that aren’t real. An AI agent transacting against that catalog makes your data problems customer-facing at machine speed.

The sequence that works: fix the data flow first — middleware between ERP and storefronts, real-time inventory and contract pricing, order validation with failure queues. For Adobe Commerce merchants specifically, this is the core of our Magento ERP integration services. Then apply AI on top of data you can trust.

Not sure where your data stands? Our free eCommerce health check reviews performance, integration gaps, and catalog readiness — with clear next steps.

Five questions to ask your Magento agency about AI

These separate agencies doing the work from agencies listing the keyword:

1. Show me an AI project where you rejected the output. A team that has never thrown away AI-generated work has never checked it.
2. How do you keep the model from inventing product attributes? Look for a concrete answer: confidence thresholds, source-data constraints, review queues.
3. What happens to my catalog before the AI touches it? If the plan starts with generation rather than a data audit, expect polished output on top of broken data.
4. How will AI agents read my store? The answer should reference UCP, ACP, structured product data, and what Adobe Commerce exposes today versus what’s on the roadmap.
5. Which AI tools have you stopped using, and why? The landscape turns over fast — Adobe retired Aero barely a year after agencies across the industry were recommending it. An agency that has never re-evaluated its stack isn’t paying attention.

Where this leaves you

AI in Adobe Commerce in 2026 is real, and it’s concentrated in two places: the platform’s discovery and agent-readiness layer, and the data work that makes that layer trustworthy. A Magento agency worth hiring is fluent in both — from custom Magento development to the integration layer underneath — and clear with you about which vendor claims have shipped and which haven’t.

If your catalog, integrations, or agent-readiness could use a second opinion, talk to us. We’ll tell you what’s real, including the parts that don’t require our help.