The Future of Commerce: Agents are Rebuilding B2B Procurement from the Ground Up
For decades, B2B procurement has been stuck in a paradox. Companies spend billions on enterprise software, yet procurement teams still operate like it's 1995. They chase quotes via email, reconcile invoices in spreadsheets, and manually verify supplier credentials across fragmented databases. The average procurement cycle for industrial equipment takes 30+ days with inefficient systems costing businesses billions annually in wasted time and lost productivity. Despite all the enterprise software, 60% of procurement work remains pure administrative overhead.
Legacy systems failed on two levels. First, they couldn't automate the basics. Data sits fragmented across ERP, procurement software, and inventory systems with no single source of truth. Implementations take 18 months and fail 70% of the time. Even when they work, these systems require constant manual intervention. Procurement teams spend hours validating supplier data, resolving PO-invoice discrepancies, and navigating approval workflows that take weeks to complete. Second, they couldn't handle judgment. A procurement manager sourcing electronic components needs to know that "ISO 9001 certified" isn't the same as "ISO 9001 compliant," that a substitute capacitor requires voltage and package validation, that delivery lead time matters as much as price. Defense contractors need ITAR compliance. Medical device manufacturers need FDA approval. These aren't workflow problems or database problems, they're expertise problems. And until recently, that expertise only existed in people's heads.
Why Now
Foundation models changed the equation. LLMs can now parse technical specifications, match products across catalogs, verify regulatory compliance, and negotiate pricing. More importantly, they understand context: which certifications actually matter, which suppliers deliver on time, how to structure an RFQ that satisfies both legal and operational requirements. The market is responding. GenAI has already sparked a 4,700% year-over-year traffic surge to B2B eCommerce sites, and in a few years, 90% of B2B purchases, or $15T worth of transactions, will flow through AI agents rather than procurement teams. But models alone don't rebuild markets. What's happening now is a complete rearchitecture of how B2B procurement operates, from the agents executing purchases to the ERPs integrating them to the payment rails settling transactions. Each layer needs to work together.
Distribution opened from an unexpected direction. Next-generation ERPs stopped trying to own procurement. Modern finance platforms focus ruthlessly on accounting and inventory (their actual core competencies) rather than building procurement features in-house. They partner with specialized procurement agents instead. When NetSuite or SAP controlled the full stack, procurement startups had to displace entrenched relationships. Now they integrate alongside the new infrastructure as the embedded "procurement module."
The moats are different too. Supplier networks create winner-take-most dynamics that didn't exist in traditional software. The first procurement agent in a vertical builds direct relationships with suppliers that compound as transaction volume grows. More suppliers generate better pricing data, which improves matching accuracy, which in turn attracts more volume. Competitors entering later aren't just behind on product, they're locked out of the relationship graph that makes the product work. Domain expertise accumulated through thousands of transactions can't be replicated by better models. Knowing which suppliers actually deliver on time, which certifications buyers scrutinize versus rubber-stamp, and how to structure an RFQ to satisfy both legal and operational requirements are intelligence that gets embedded through customer feedback and supplier interactions in ways foundation models can't shortcut.
Taking a deeper look, there are three important sub-sectors redefining the way B2B procurement is designed and executed.
Vertical Procurement Agents
The procurement software market is splitting along vertical lines. Government contracting requires navigating NAICS codes, small business set-asides, and agency-specific proposal formats. Electronic components demand real-time allocation negotiation and counterfeit risk assessment. Heavy machinery procurement needs ISO certification validation and lead time optimization across global suppliers. Generic LLMs can't handle these workflows without deep domain training or heavy customization.
Diagon automates heavy machinery sourcing for manufacturers. Built by ex-Tesla procurement leaders who spent years managing capital equipment purchases at scale, their platform connects buyers to tens of thousands of verified equipment vendors. Makat tackles electronic component procurement for contract manufacturers and OEMs. When production lines need semiconductors and strategic suppliers are out of stock, their AI screens global databases in real-time. Magentic focuses on value leakage recovery. Large manufacturers leak roughly 2% of total spend through contract non-compliance and missed rebates. Their agents audit invoices against contracts and recover savings. The commonality isn't technology because everyone uses similar foundation models. It's vertical depth. Diagon knows heavy machinery suppliers. Makat knows electronics component distributors and allocation strategies. Magentic knows contract enforcement patterns. That specialization creates defensibility horizontal platforms can't match.
AI-Native ERPs
While vertical procurement agents are capturing industry-specific workflows, a parallel shift is happening in finance software. A new generation of ERP platforms recognized that mid-market companies don't need everything a legacy ERP does. They need a few things done exceptionally well. Doss and DualEntry focus on fast financial close and automated reconciliation. GoodDay Software serves Shopify merchants who need multi-channel inventory and landed cost tracking, not generic manufacturing workflows. These platforms win by solving acute pain points in days, not months, and letting specialists handle everything else. Rather than building procurement modules in-house, they integrate with vertical agents. This creates a distribution advantage. As AI-native ERPs capture mid-market finance operations, procurement specialists plug in as the natural next layer. The modular approach means startups don't need to displace entire systems. They integrate where the new infrastructure already exists.
Agent-to-Agent Payment Infrastructure
Autonomous agents executing purchases creates a payments problem traditional infrastructure wasn't designed to solve. Credit card rails can't handle micro-transactions at fractions of a penny. ACH takes days when agents need millisecond settlement. Fraud detection flags "card not present, human not present" as suspicious by default. There's no way to verify delegation authority or enforce contextual limits like "office supplies under $500" without manual approval workflows.
Startups are building the missing layer. Skyfire uses stablecoin settlement for instant clearing, which is critical when agents execute thousands of API calls and data purchases that traditional payment economics can't support. Payman targets regulated environments where banks want AI agents but need white-label infrastructure that preserves compliance control.
Major technology companies are launching protocols with strategic partners, validating that agent payments become foundational infrastructure. Early movers defining standards now have the opportunity to become the rails powering autonomous commerce at scale.
B2B procurement is being rebuilt from the ground up. The transformation isn't incremental, it's architectural. Vertical specialists are capturing industry-specific expertise that generic software can't replicate. AI-native ERPs are partnering rather than competing, creating natural distribution channels for procurement agents. Payment infrastructure is being rebuilt for a world where software initiates transactions, not humans.
What makes this moment different from decades of failed procurement automation isn't just better AI. It's that the entire stack is being rebuilt simultaneously. Legacy systems tried to automate workflows while keeping humans in the loop for judgment calls. Autonomous agents remove the loop entirely. They don't need approval to order office supplies or validate supplier credentials. They execute based on encoded policies and accumulated transaction data. The infrastructure finally exists to support this with payment rails that verify agent identity and enforce spending limits, ERPs that integrate specialists rather than building everything in-house, and supplier networks dense enough to handle edge cases without human escalation.
For procurement teams, this shifts the role fundamentally. Instead of chasing quotes and reconciling invoices, they set policies, monitor exceptions, and negotiate strategic relationships. The administrative burden that consumed 60% of their time gets compressed into background automation. The companies that win will be those that remove traditional blockers entirely, build data moats through proprietary supplier networks, and embed at the infrastructure layer where switching costs become prohibitive. Superior technology gets you in the game, but network effects determine who wins it.
If you're building in autonomous B2B commerce, we'd like to hear from you. Please reach out to our team at np@interplay.vc.