Agent-to-Agent commerce in B2B: how AI procurement agents are changing vendor selection in 2026. Agent-to-agent commerce in B2B refers to automated commercial transactions where an AI-powered buying agent - deployed by a purchasing organisation - discovers, evaluates, and initiates contact with v Gritnord Team September 11, 2026 Agent-to-agent commerce in B2B refers to automated commercial transactions where an AI-powered buying agent - deployed by a purchasing organisation - discovers, evaluates, and initiates contact with vendors without human involvement, while a corresponding AI selling agent responds on behalf of the vendor. As of 2026, this model is operational at enterprise scale, driven by platforms including Coupa, SAP Ariba's Joule integration, and OpenAI's Operator product, and it is beginning to affect how mid-market B2B vendors are discovered, shortlisted, and engaged. Why agent-to-agent commerce is no longer theoretical. Enterprise procurement teams at Fortune 500 companies are deploying AI agents to handle intake-to-purchase-order workflows for repeat vendors and software renewals. OpenAI's Operator, commercially expanded in early 2026, is executing live procurement sequences under defined spend thresholds - typically sub-$25K - without human sign-off. On the vendor management side, Coupa launched its Agent Marketplace in Q2 2026, allowing enterprises to deploy pre-vetted purchasing agents with role-scoped permissions. SAP Ariba's Joule integration now handles vendor shortlisting and RFQ generation autonomously for catalogue items. Zip, a procurement intake platform, has shipped an AI agent layer covering the full intake-to-PO cycle for repeat vendors. The implication for B2B sellers: an automated agent is now running the first stage of vendor evaluation at a growing number of enterprise accounts. If your product is not visible to that agent, you are not in the shortlist. How AI purchasing agents actually discover vendors in 2026. The discovery layer for agent-to-agent commerce is currently fragmented across three mechanisms: Vendor registries and agent directories. AGNTCY - an open agent directory backed by Cisco and LangChain - is emerging as a public index where vendors publish machine-readable capability cards. These cards contain pricing tiers, API endpoints, compliance certifications, and SLA parameters. Buying agents query these directories the way a human might search G2. Platform-native marketplaces. Coupa's Agent Marketplace and SAP Ariba function as curated procurement environments. Vendors listed here are accessible to enterprise buying agents by default. Not being listed is equivalent to not existing within that procurement stack. MCP (Model Context Protocol) connectors. Anthropic's MCP has become the de facto integration protocol for enterprise agent tooling. HubSpot, Salesforce, and SAP have all shipped native MCP connectors as of Q3 2026, with over 4,000 MCP servers now registered in public directories. A buying agent can call directly into a vendor's product catalogue or pricing API as a tool call - no human demo required. The agent-to-agent transaction: discovery, negotiation, and contract. Once a buying agent identifies a candidate vendor, the transaction follows a structured sequence: Negotiation is not conversational. Buying agents submit JSON-schema RFQs - structured requests for quotation - and selling agents respond with qualified offers inside defined parameters. Google's A2A (Agent-to-Agent) protocol, currently at v0.9, defines the message format, task lifecycle, and capability declaration for these exchanges. Contract execution remains the weakest layer. Most live deployments rely on pre-approved master agreements, with agents filling variable fields such as volume, term, and price. DocuSign has an agent API in beta that allows an authorised agent to execute against pre-signed framework agreements, but full autonomous contract signing is not yet standard. Settlement for software purchases uses Stripe's agent billing primitives - metered, credentialed API keys with spend limits - as the dominant mechanism. Physical goods remain human-gated. The critical gap is authentication. There is no cross-platform standard for verifying an agent's authority or its principal's identity. Enterprises are managing this with human ratification checkpoints, which limits full autonomy but keeps transactions auditable. What this means for B2B sales teams at mid-market saas companies. For VP Sales and founders at companies with €5M-€30M ARR, the operational priority in the next six to twelve months is not building a fully autonomous selling agent - it is ensuring your product is legible to buying agents running automated shortlisting. Immediate actions: * Publish a machine-readable product data layer: pricing, packaging tiers, compliance documentation (SOC 2, ISO 27001, GDPR alignment), integration specs, and SLA parameters - accessible via API or structured webpage. * Evaluate whether your target enterprise accounts use Coupa, SAP Ariba, Zip, or Workday procurement agents, and pursue listings in those environments. * Audit your HubSpot or CRM data layer. If your product details are not structured consistently, a buying agent calling your MCP connector will return incomplete or conflicting data. Tools like Clay are already enabling agent-to-agent handoff workflows - a sales agent initiates contact with a buying agent endpoint, and the two systems assess fit before escalating to a human. This is not science fiction; it was shipped in June 2026. The open problems that still require human oversight. Three unresolved issues limit full agent-to-agent autonomy in 2026: Pricing negotiation between agents is absent in real production deployments. Agents accept listed prices or follow rigid discount schedules. Dynamic negotiation requires intent verification that does not yet exist at scale. Liability for agent-executed deals - wrong pricing tier, incorrect contract term - is legally unresolved in every major jurisdiction. Enterprise risk teams are managing this with human ratification gates. Discovery fragmentation means that AGNTCY, Coupa, MCP directories, and proprietary vendor APIs are not interoperable. A buying agent must be configured per vendor, which remains the immediate bottleneck for widespread adoption. Frequently asked questions. What is agent-to-agent commerce in B2B? Agent-to-agent commerce in B2B is an automated process where an AI buying agent deployed by a purchasing company discovers, evaluates, and initiates commercial contact with a vendor, while the vendor's AI selling agent responds - completing stages of a transaction without human involvement on either side. Which enterprise platforms are deploying AI purchasing agents in 2026? Coupa, SAP Ariba (via Joule), Zip, Workday, and ServiceNow have all shipped AI procurement agent functionality in production as of 2026. OpenAI's Operator is also running live procurement workflows at several Fortune 500 accounts. How do AI buying agents discover vendors? Buying agents query vendor registries such as AGNTCY, platform-native marketplaces like Coupa's Agent Marketplace, and MCP-connected product APIs. Vendors must publish machine-readable capability cards - structured data covering pricing, compliance, and integration specs - to be visible. What is the A2A protocol and why does it matter? Google's Agent-to-Agent (A2A) protocol, currently at v0.9, defines the message format, task lifecycle, and capability declarations for structured exchanges between buying and selling agents. It is the closest thing to a universal standard for agent-to-agent B2B negotiation. Do B2B sales teams need to build a fully autonomous selling agent now? No. The immediate priority is ensuring your product data is machine-readable and structured for agent discovery. A selling agent can begin as a simple endpoint that escalates to humans - the interface must exist, but full autonomy is not yet required or standard. The shift from human-mediated sales to agent-mediated discovery is not arriving all at once. It is layering into existing procurement stacks quietly - starting with renewals, tail spend, and catalogue purchases. B2B sellers who structure their product data for machine readability now will be shortlisted by buying agents that their competitors never even knew were running.

Network Routing Optimization: go from raw data to actionable insight. By: Sulaiman Panjwani Senior Product Marketing Manager Table of contents. Every logistics executive has felt this pain: Your network design team delivers a "fully optimized" blueprint for your distribution facilities, only for your transportation team to take one look at it and say, "This is completely unexecutable on the road." For decades, supply chain organizations have been forced to treat network footprint design and transportation routing as two completely disjointed, sequential decisions. * Network Optimization (NO) software typically assumes simplified, point-to-point shipping costs. It lacks the granular math required to accurately model multi-stop routing (like milk runs), fleet limits, or driver regulations. * Transportation Optimization (TO) software understands these route-level details perfectly, but it operates in a vacuum. It cannot factor in overarching enterprise constraints like plant production capacity, inventory holding costs, or multi-echelon network flows. The result? A costly, highly inefficient loop where modeling teams are forced to manually iterate back and forth between NO and TO five or more times. This slow, manual process delays critical decisions and creates a dangerous gap of "inconsistent costs," where the strategic financial projections approved by leadership do not match actual operational execution on the road. Unifying strategy and execution: Introducing Coupa NRO. To permanently bridge this gap, Coupa has introduced Network Routing Optimization (NRO). Network Routing Optimization (NRO) is a powerful, cloud-native problem type that unifies strategic Network Optimization and tactical Transportation Optimization into a single decision model. NRO is a powerful, cloud-native problem type that unifies strategic Network Optimization and tactical Transportation Optimization into a single decision model. Instead of treating logistics and facility design as sequential steps, NRO dynamically evaluates precise multi-stop routing costs, fleet limitations, and driver constraints simultaneously alongside overarching network variables. By natively embedding TO constraints within the strategic NO solver, NRO supports: * True cost structures: Stop costs, distance-based costs, and duty-time costs * Fleet limitations: Vehicle capacity limits, average speeds, maximum travel time, maximum distances, and maximum stop counts * Regulatory & safety compliance: Hours of Service (HOS) regulations, including maximum drive time per shift, maximum duty time per shift, and mandatory minimum break times This unified approach is powered by four unique, cloud-native capabilities that redefine the standard for supply chain modeling: * Multi-stop route sampling: Rather than relying on simple point-to-point assumptions, NRO evaluates realistic, multi-stop route candidates directly within the overarching network model. * Parallel route generation: Instead of running sequential, time-consuming simulations, NRO utilizes Coupa's powerful Joule cloud infrastructure to generate route options in parallel, drastically accelerating solver speeds. * Automated output merging: Once the optimal routes are identified, the engine automatically adjusts and merges the chosen route data into your network optimization files. This provides a frictionless, ready-to-execute dataset for your downstream teams. * Frictionless commercial packaging: Coupa has eliminated any adoption barriers. NRO is available within SC Modeler at no additional cost. It is natively embedded under the standard Network Optimization umbrella, and each NRO run counts as exactly one solve against your existing limits. Real-world ROI. By solving the strategy-to-execution disconnect, NRO delivers massive operational efficiency and immediate margin protection: * Unmatched solve-time reductions: By completely eliminating the manual back-and-forth between planners and routing teams, NRO slashes time-to-value. In enterprise validations simulating five iterations, NRO drove an 84.5% reduction in solve times for a leading specialty chemical manufacturer, and a 34.06% reduction for a global food and beverage enterprise. * Significant cost reductions: Because traditional network design locks in sourcing decisions too early, it misses global optimization opportunities. NRO optimizes holistically, delivering a 40.36% reduction in NO costs for DHL for a global logistics provider, and a 41.73% reduction in TO costs for a global food and beverage enterprise. If your distribution network relies heavily on last-mile deliveries, retail store replenishment, or multi-stop inbound milk runs, continuing to design your supply chain on disjointed legacy systems actively erodes margins.With Coupa NRO, the network you design is the exact network you actually build and run. Enhancing tactical freight agility: Relay hubs and tractor-trailer optimization. Long-haul freight rarely moves in a straight line. Drivers hit DOT hours-of-service limits, need to get back to their domicile, or sit idle while trailers wait for loading. To complement strategic NRO modeling, Coupa has also introduced two new tactical capabilities within Transportation Optimization. 1. Keep freight moving with relay hubs. Picture a shipment that has to travel farther than any single driver can legally go in one shift. Traditionally, that meant either breaking DOT rules, delaying the load, or manually coordinating a driver or trailer swap partway through the trip. Coupa's new relay-hub functionality automates that coordination. You define a set of candidate relay hubs, and the solver figures out which hub should serve each customer. The system automatically routes the swap so the shipment keeps moving without breaking the load or violating driving-time and domicile rules. The result: Longer routes become possible, compliance stays intact, and freight doesn't sit and wait for a workaround. 2. Optimize tractors and trailers as what they are: Separate assets. Tractors and trailers are two different resources with two different utilization stories, but most optimization software has forced them into a single bundled unit. This makes it difficult for logistics teams and 3PLs to optimize trailer utilization independently. Coupa now optimizes tractors and trailers as separate assets, each with its own route, while still respecting every shared constraint between them. This means you can see, and optimize, tractor utilization independently. Early access results have been excellent. The solver can drop a trailer at a yard for staged loading or unloading and immediately send the tractor on its next shuttle move, without forced idle time. Or it can drop the trailer, send the tractor off to complete a customer delivery, and bring it back later to pick the original trailer back up. The tractor stays productive, and the trailer gets staged exactly when needed. The result: Tractor idle time is eliminated, trailer staging becomes seamless, and asset utilization increases across your entire fleet. Ready to unify your network design and transportation routing?

Coupa's Chief Information officer featured by Women We Admire. The organization highlights women executives and leaders across the U.S. and Canada. Published on Sep. 03, 2026 Rose Velazquez Coupa, a company specializing in AI-powered spend management software, recently had one of its executive leaders featured by Women We Admire, a member-led network of women executives and leaders based in the U.S. and Canada. Amy Sweeney, Coupa's senior vice president and chief information officer, was listed among The Top Women Chief Information Officers of 2026. As enterprise organizations continue innovating digital solutions to stay competitive in the modern age, chief information officers are investing in artificial intelligence, data infrastructure, cybersecurity, cloud platforms and optimizing critical systems. Women We Admire's 2026 CIO honorees are spearheading technology functions that support their organizations' global operations, strengthen their business performance and equip them for advanced enterprise computing. At Coupa, Sweeney leads technology initiatives and the company's global IT organization, overseeing infrastructure, business applications, data and analytics, enterprise architecture and program management. She served as vice president of business systems at Dynatrace before joining Coupa, bringing three decades of business operations and technology leadership to the company.