Rewiring Supply Chain with Multi-Agent Infrastructure

Procurement is a blend of negotiation, operations, risk management, and strategy. It’s the connective tissue of the supply chain, where decisions around cost, quality, timing, and risk are made under pressure and often with incomplete information.

Every decision carries a cost, whether it’s paying a premium for speed, accepting longer lead times to reduce spend, or choosing a supplier with less proven quality. So do the tradeoffs made in the name of preparedness (e.g. dual sourcing = higher overhead, buffer inventory ties up capital). In a world of shifting tariffs, supplier instability, and geopolitical volatility, there’s an expectation to anticipate disruptions while responding to them in real time. That’s hard to do without better visibility, clearer signals, and faster coordination across functions.

As long-term investors, we’re constantly evaluating how structural shifts in supply chains reshape the systems and people that underpin them. But the opportunity set isn’t always clear-cut. Procurement is messy. It’s not one problem. But an aggregation of a hundred small ones, much of which stem from siloed data, fragmented tools, and misaligned incentives.

Workflows have remained painfully manual. Teams stitch together messy inboxes, Excel trackers, rigid ERPs, and tribal knowledge. On the surface, this seems like a natural fit for AI. Point solutions have emerged to automate individual steps: parsing RFQs, benchmarking quotes, flagging contract terms, generating POs.

It’s no doubt these tools deliver value. But they operate in isolation; and over time, that creates new problems for everyone else.

AI-for-X solutions have flooded every industry in recent years, and procurement is no exception. There’s a tool for scoring suppliers, drafting sourcing events, flagging contract risks, and more. But each builds only narrow intelligence, optimized for a single step rather than the full context.

Real procurement decisions don't live in a silo. Choosing a supplier isn’t just a question of price or quality. It depends on forecast volatility, budget constraints, quality control flags, and geopolitical risk. These dependencies live in ERP, MES, QMS, and inboxes. A point solution might optimize a step, but it doesn't understand the cross-functional tradeoffs that define the job.

We believe a multi-pronged agentic approach changes that. Think of it not as a single AI that does procurement, but as a network of agents, each embedded across the supply chain stack. One lives in procurement workflows. Another in finance, tracking budget and supplier payment terms. Another in quality, flagging NCR trends. Together, they surface insights, adapt plans, and handle tasks that would normally require weeks of back-and-forth across departments.

Let’s look at the workflow:

• Demand signal hits from MRP or forecast adjustment

• Buyer issues RFQ via email or ERP sourcing module

• Supplier responses come back as PDFs, Excel files, or quotes in email

• Teams compare cost, lead time, quality history (usually manually)

• PO is issued, contract is signed, delivery is tracked

At every stage, decisions hinge on inputs from other functions. Finance approves spend. Engineering confirms specs. Legal flags contract risk. Quality approves first articles. If any link fails, procurement stalls.

This is why point solutions break down. AI can extract terms or analyze quotes, but it can't adapt to shifting constraints or pull judgment from a cross-functional context. The result? Teams revert to email threads and spreadsheet fire drills when real decisions need to be made.

The future isn’t "AI for procurement". It’s AI that owns the interface across systems and workflows, a lattice of agents that coordinate across domains. For example:

• When a new forecast drops, the demand agent alerts the sourcing agent to reevaluate suppliers.

• If a preferred supplier’s quality score drops (pulled from MES or QMS), the risk agent flags it and suggests alternatives.

• The finance agent monitors PO exposure and payment history to ensure spend stays within bounds.

This networked layer acts as an operational nervous system, surfacing risks, triggering actions, and accelerating routine decisions that would otherwise stall across teams. It doesn’t replace human judgment; it enhances the system’s ability to respond with clarity and speed.

Automotive supply chains are brittle. Lead times stretch over months. Supplier capacity is pre-booked in advance. And the cost of a late or failed part ripples across production lines. From sourcing battery trays to stamping brackets, every component introduces new risk.

After interviewing folks across companies from EV to legacy OEMs, fragility permeates these organizations and between functions. Supplier vetting is slow, often manual. Financial health is inferred from D&B reports. Performance metrics sit in silos. Quality issues like batch-to-batch variation go undetected until a warranty crisis hits. And when forecasts swing or a supplier underperforms, procurement must re-plan quickly, often without visibility into alternatives.

The case for AI isn’t just about speed. It’s about reframing how procurement operationalizes resilience. Multi-agent platforms can synthesize supplier data, infer delivery risk, and propose mitigation options in real time. When geopolitical events disrupt Tier 2 production in Taiwan, the system can surface alternates in Mexico with comparable specs and available tooling. When a supplier’s cash flow wavers, it can model the downstream impact on delivery timelines and payment exposure before it becomes a problem.

As we know, procurement doesn’t operate alone. It reacts to and triggers downstream functions: engineering, finance, manufacturing, logistics. Today, most of these teams operate with disconnected tools—CAD, PLM, MES, QMS, ERP, etc. AI embedded in just one of them can only go so far.

That’s why we believe the long-term opportunity is in agentic interoperability: agents embedded in each function that talk to each other, coordinate work, and collectively handle decisions. Engineering agents extract GD&T tolerances and feed them to DFM agents. Manufacturing agents flag production bottlenecks that trigger procurement agents to re-source. Quality agents surface part failure trends that inform supplier risk scores.

These agents don’t just automate tasks. They rewire how decisions are made. They collapse latency across functions and turn reactive processes into adaptive systems.

Let’s be clear: the goal isn’t to rip out your ERP. It’s to own the layer that makes them usable. ERPs weren’t built for real-time planning or exception handling. They were built for record-keeping. Multi-agent systems can sit above existing systems, extracting signals, writing actions, and driving decisions.

In time, this network of AI agents can become the new system of record for how decisions are made. Who got flagged? Why was a supplier switched? What was the delivery risk assessment tied to a forecast swing? That provenance lives in the agents.

So where do we go from here? The next generation of supply chain software isn’t a better sourcing module. It’s a distributed system that thinks, plans, and adapts across functions.

Not an AI assistant for the buyer, but a co-pilot for the enterprise.

We’ve spent the past few months speaking with leaders across the supply chain and industrial stack. If you are (or know of) someone who fits the following, I’d love to connect:

• Operators and leaders in procurement, supply chain, engineering, quality, finance, or operations navigating these challenges firsthand

• Founders building new infrastructure, agentic systems, or AI-powered tools for the industrial world

Let’s chat!