Agentic AI and the Future of B2B Buying​​​​‌‍​‍​‍‌‍‌​‍‌‍‍‌‌‍‌‌‍‍‌‌‍‍​‍​‍​‍‍​‍​‍‌‍‌​‌‍​‌‌‌​‌‍‌‍​‌‍‌‌​​‍‍‌‍​‌‍‌‍‌​‍​‍​‍​​‍​‍‌‍‍​‌​‍‌‍‌‌‌‍‌‍​‍​‍​‍‍​‍​‍‌‍‍​‌‌​‌‌​‌​​​‍‍​‍​‍​‌‌​‍​​‌​‌​​‍‌‌‍​‌‍‌‍‌‌‍‌‌‍‌‌‌​‍‌‍​‌‍‌‌​‍‍‌‍​‌‌‍‌​‌‍‌‌‍‍‌‌‍‍​‍‍‌‍‌​‌‍​‌‌‌​‌‍‌‍​‌‍‌‌​​‍‍‌‍​‌‍‌‍‌​‍‌‍‌‌‌‍‌​‌‍‍‌‌‌​‌‍‌​‍​‍‌‍‍‌‌‌​‌‍‌‌‌‍‌‌‌‌‌​‌‍‌‌​​‌‍‌‌‌​​‍‌‌‍‌‌​‌‌‍​‌​‌‍‌‍‌‌​​‌‌‌​‍‌‌​‌‌​‍​​‌‌​‍‌‌​​‌​‌​‌‍‌‍‌‌‌‌‌‍‍‌‍‌‌​​‌‌‍‍​​​‌​​‌​‍‌‍‍‌‌‌​‌‍‌‌‌‍‌‌​​‍‌‍‌‍‌​​‍‌‍‌​‌‌‌‍‍‍‌‍​‌‍‌​‍‌‌‌​‌‌‍‍‌​‌‍‌‍​‌‌​​‌‌​​‌‍‌​‌‌‌​‍‌‍​‌​‌‌‌​‍‍‌‍‌‌​​‌​‍‌‍‌‌‌‍‌​‌‍‍‌‌‌​​‍​‌‍‌‍‌‍‍‌‌‍‌‌‌‍​‌‍‌​‌‌​​‌‍​‌‌‌​‌‍‍​​‌‌‌​‌‍‍‌‌‌​‌‍​‌‍‌‌​‍​‍‌‌

The forecasts are not exactly modest. Gartner predicts that by 2028, 90% of B2B buying will be intermediated by AI agents, moving more than $15 trillion of spend through agent exchanges (Digital Commerce 360). And buyers aren't waiting for 2028: 67% now prefer a rep-free experience, and 45% used AI during a recent purchase (Gartner).

Before anyone starts clearing out the sales team's desks, though: 69% of buyers still go back to a rep to check what the AI told them (Gartner). The robots aren't running procurement just yet. For manufacturers and distributors, the smart move right now isn't chasing the latest chatbot channel. It's making sure your product data, pricing and stock are clean enough for a machine to read and trust, because every version of this future runs on them.

Strip away the buzzwords and agentic commerce is simple: software acting on a buyer's behalf. It gathers requirements, compares options, builds a basket and, sometimes, presses "order".

Behind the scenes, the tech giants have been busy agreeing how agents and sellers should talk to each other. There are already four main standards (it's been a busy couple of years), and happily they work together rather than squabble.

The platforms aren't hanging about either. Shopify has made agentic storefronts generally available, and BigCommerce has an MCP server in beta.

The catch for B2B: none of these standards yet understands how trade selling actually works. There's no agreed way to handle contract pricing, account hierarchies, who's allowed to sign off what, purchase orders or credit terms. Shopify's own documentation is refreshingly honest about it: its agentic storefronts are for direct-to-consumer sales only, and B2B pricing doesn't appear on them. In short, the agents can go shopping, but they can't run your trade counter. Yet.

Picture a hotel group refurbishing its bathrooms. Its procurement agent checks three suppliers in seconds. Two drop out not because their valves are worse, but because the agent can't confirm the fit or the price. The third wins because its product data, pricing and stock are all machine-readable, and the order lands in its ERP without anyone touching a keyboard. Somewhere, a sales rep at Northfield Valves is still waiting for the phone to ring.

An agent can't be won over by a slick homepage, a firm handshake or a very good sales lunch. It picks suppliers whose information it can read, trust and act on in seconds. Four things decide whether you make the shortlist.

• Product data a machine can make sense of. Consistent attributes, units of measure, compatibility rules, stable product codes, and specs attached to the right SKU. If a key dimension only lives in a PDF, then as far as the agent is concerned, it doesn't exist.

• Prices that are actually current. Contract pricing that refreshes once a night at 2am is no use to an agent that wants an answer now.

• Honest stock and lead times. Agents take what you publish at face value, so "in stock" and "should be in next week, probably" need to be clearly different things.

• Systems an agent can reach. APIs (the connections that let software talk to software), punchout links into buyers' procurement systems such as Coupa or SAP Ariba, and clear account permissions.

That's why agentic commerce is mostly a data and integration challenge, not a design one. The good news is that the foundations that make a B2B portal great for humans are exactly the ones that make it readable for machines. Two birds, one well-structured stone.

You don't need a crystal ball to work out which standard will win. Every step below pays off whether orders arrive through an AI assistant, a procurement platform or Dave in sales.

1. Sort out your product data. Bring it into one place, ideally a PIM (product information management system) such as Akeneo, with complete, consistent details for every product. If you only do one thing on this list, make it this one. Tellingly, Gartner published its first-ever Magic Quadrant for PIM in October 2026 and named Akeneo a Leader (Akeneo). As an Akeneo Gold Partner we may be slightly biased, but it's nice when the analysts agree.

2. Make pricing and stock live. Use an integration platform to share ERP pricing, availability and lead times in real time, rather than in an overnight batch.

3. Make your account rules machine-readable. Who can buy what, at what price, and who needs to approve it.

4. Check your procurement connections. Review punchout catalogues and price feeds with your biggest customers. This is where buyer agents are most likely to knock first.

5. See what the machines see. Check your site's agent discovery files (such as llms.txt), and make sure no trade-only product is showing up to AI channels at a retail price. That's a generous discount nobody signed off.

6. Start small with one internal agent. Automating order entry from emailed purchase orders makes a great first project, and it's easy to measure in hours saved and typos avoided.

7. Agree the ground rules. Decide who owns data quality, set approval limits, and be clear about where a human still has the final say.

Agentic buying is real, but the hype is currently a few laps ahead of reality. A few things are worth keeping an eye on.

• Trust and authority. Nobody has yet agreed on a standard way to prove an agent is allowed to commit a company to a purchase. Until they do, an AI ordering £50,000 of components on credit terms remains, thankfully, theoretical.

• Prices turning up in the wrong place. Hybrid sellers on Shopify can find a product shown to AI channels at its consumer price. If trade prices are hidden by a third-party app, Shopify may not recognise them as B2B at all.

• Margin pressure. Agents compare suppliers tirelessly and never get bored. If your product data is thin, you risk being judged on price alone.

• Projects that don't land. Gartner predicts more than 40% of agentic AI projects will be cancelled by the end of 2027, thanks to rising costs, unclear business value or weak risk controls. It also reckons only around 130 of the thousands of so-called agentic AI vendors are the real thing, so a healthy dose of scepticism is advised (Gartner).

• People still matter. Most buyers still get a human to check the AI's homework, so your sales team isn't going anywhere.