B2B ordering just left the portal, and the most encouraging part of that shift is how much of the groundwork most organisations have already laid. At the end of June 2026 Salesforce made the Agentforce Commerce agents generally available, and for business to business sellers the Buyer Agent is the one worth reading twice. A customer can place an order over WhatsApp or SMS, confirm a part by sending a photograph of it, receive the contract price negotiated for their account, and complete the transaction without ever logging into a portal.
Read that as a product announcement and it sounds like a new channel. Read it as a commercial opportunity and it is something better. Every manufacturer, distributor and energy supplier we work with has a segment of customers who buy steadily and have never once signed into the ordering portal built for them. They ring the rep. They text a part number from a van or a site. Somebody keys it into the CRM by hand, and everyone quietly absorbs the delay. That revenue was never lost, but it was never self-service either. The Buyer Agent reaches those customers where they already are, and it does so using the product, pricing and account data that portal projects spent years cleaning up.
What is Salesforce Agentforce, and what does the Buyer Agent add?
Salesforce Agentforce is the platform layer that lets organisations build, deploy and govern autonomous AI agents grounded in their own Salesforce data. Rather than answering questions from a general model, an Agentforce agent reasons over the customer records, catalogues, pricing and order history already held in the CRM, and it can take action inside those systems. The Buyer Agent applies that pattern to business to business purchasing. It handles procurement conversations over messaging channels such as WhatsApp and SMS, confirms which SKU a buyer means when they send an image instead of a code, applies the contract pricing attached to that account, and completes the order without a portal login. It became generally available alongside the Shopper Agent and the Merchant Agent, with native availability inside ChatGPT and further search integrations following.
The distinction that matters is between a chatbot bolted onto a storefront and an agent wired into the catalogue, the inventory and the order object from the first day. The first one answers. The second one transacts. For a distributor whose customers reorder the same forty items every fortnight, transacting is the whole point.
Why does portal-free ordering matter more in B2B than in B2C?
Consumer commerce solved friction years ago by removing steps. Business to business commerce could not follow, because the steps carried meaning. A login identified which account a buyer belonged to, which price list applied, which approval limits governed the order, and which delivery addresses were valid. Take the login away and all of that has to be established another way. That is precisely what the Buyer Agent does differently: it recognises the buyer from the messaging identity, resolves them to a contact and an account, and pulls the entitlements attached to that relationship. The friction disappears without the commercial context disappearing with it.
The gain shows up fastest in the least digital corners of a customer base. Site supervisors, workshop managers, facilities teams and field engineers order constantly and sit furthest from a desktop browser. They are also the buyers whose orders currently arrive as a phone call at four in the afternoon. Moving that traffic into a channel that writes straight into the CRM does two useful things at once. It gives the customer a faster answer, and it gives the commercial team a complete picture of demand that previously lived in a rep’s call notes.
What has to be true in your data before a Buyer Agent can quote accurately?
An agent that quotes prices is only as trustworthy as the records it reads, which makes this a data readiness question first and a configuration question second. In practice four things need to be in order. Product master data has to be complete and consistent, so that a description, an image and a code all resolve to the same item. Contract pricing has to live somewhere the agent can query in real time, including customer-specific price lists and volume breaks. Account hierarchies have to reflect how the customer actually buys, so that a subsidiary ordering under a group agreement receives the group’s terms. And stock or lead time information has to be reachable, because a confirmed order against unavailable inventory creates a service problem rather than a sale. Organisations that built a working B2B storefront have usually solved three of those four already.
That last sentence is the genuinely good news in this release. This is rarely a rebuild. For most teams it is an extension of investment already made, pointed at a much larger audience. The work that went into rationalising a product catalogue for a portal does not need repeating; it needs exposing through a different door. Where a gap does appear, it is usually narrow and specific, which makes it far easier to scope and fund than a platform-level programme.
How do you connect CRM and ERP so an agent can confirm stock and delivery?
Connecting CRM and ERP for agent-led ordering follows the same principles as any integration, with one added requirement: the agent needs answers at conversational speed. The practical approach is to define a small number of read operations the agent genuinely needs, typically stock position, lead time, contract price and credit status, and expose each as a governed service rather than replicating whole tables into the CRM. Order creation then flows the other way, with the CRM holding the commercial record and the ERP receiving the fulfilment instruction. Keeping the integration surface deliberately small keeps it fast, testable and cheap to maintain, and it avoids the synchronisation debt that accumulates when two systems try to hold the same truth.
Teams that have already invested in a middleware layer or an API strategy are well placed here, because the agent becomes another consumer of services that exist. Those without one should resist the temptation to build everything before starting. A single product family with clean pricing and a reliable stock feed is enough to run a real pilot, prove the commercial case with actual orders, and use that evidence to fund the wider integration.
What changes for manufacturers and distributors specifically?
Manufacturing and distribution stand to gain the most, because their order patterns are repetitive, their pricing is contractual, and their buyers are mobile. A reorder of known items against an agreed price list is close to the ideal case for an agent: the decision space is narrow, the data is structured, and the value of speed is obvious to both sides. Image-based SKU confirmation matters more here than it first appears, because anyone who has taken a call describing a fitting by shape rather than part number knows how much time identification consumes. Letting a buyer photograph the item and have the agent match it against the catalogue removes a step that has frustrated counter staff for decades.
There is a second benefit that tends to surface a quarter or two in. When ordering conversations run through the CRM, demand signals become visible that previously never got recorded. Which items customers ask for and cannot get, which substitutions they accept, how order frequency shifts by region and season. That is planning intelligence, and it arrives as a by-product of making ordering easier rather than as a separate analytics project.
How should an organisation sequence this over the next two quarters?
The sequencing that works starts commercially rather than technically. Identify the customer segment whose orders are simplest and most frequent, because that group will generate the clearest evidence fastest. Audit the data those specific orders depend on, which is a far smaller job than auditing the catalogue in full. Run the agent alongside existing channels rather than replacing anything, so that the sales team sees it as extra capacity rather than a threat to their relationships. Then measure the things that will justify expansion: order cycle time, service desk volume, and the share of orders arriving outside working hours.
One caution is worth stating plainly, and then moving past. An agent that transacts should have clear limits on order value, discount authority and exception handling, agreed before launch rather than after the first surprise. Set those boundaries deliberately and the governance question is answered in an afternoon. Teams that treat it as a design decision at the start find it accelerates approval internally, because finance and operations can see exactly where the agent stops and a person begins.
The Sirocco perspective
What we find most promising about this release is that it rewards discipline rather than novelty. The organisations best placed to benefit are not the ones with the largest AI budgets. They are the ones that took product data, contract pricing and account structure seriously over the past few years, often for unglamorous reasons, and now find that groundwork opening a channel they did not build it for. As an independent CRM partner working across Salesforce, HubSpot and Dynamics 365, we see the same pattern repeating across platforms: the capability arrives quickly, and the teams getting real value from it are the ones whose data was ready to answer.
Our advice to clients weighing this is to treat it as a commercial pilot with a data workstream attached, not as a technology programme. Pick one customer segment, one product family and one clear measure of success. The scope is small enough to move this quarter, and the learning transfers directly to every subsequent expansion. Organisations that start now will spend the next two quarters refining something that works, which is a considerably better position than starting the same conversation next year with the same questions unanswered.
If any of this is live on your roadmap, you are welcome to schedule a consultation and talk it through with someone who has done it on more than one platform.
Get in Touch
If you are weighing whether your product data, contract pricing and ERP connections are ready to support agent-led ordering, we would be glad to walk through it with you.
