The Reshoring Stack: YC Is Betting the American Factory Gets Rebuilt by Agents, Welders-in-a-Box, and Sourcing Brokers

The wave right now

For a decade, "manufacturing software" meant visibility — dashboards bolted onto ERPs that told a plant manager what was wrong without doing anything about it. The 2023–2026 wave has flipped the product category: this cohort sells work, not visibility. Arzana pitches AI agents that *do* quoting and order entry for US manufacturers. Advanced Metal Research sells a robot that *does* the welding, branded unapologetically as "machine intelligence for American welding." Saudara AI doesn't sell a supplier database — it *is* the sourcing broker, returning vetted quotes in 48 hours.

The macro engine is obvious and, I'd argue, durable: reshoring policy plus a generational skilled-labor exit. Every company in this cohort cites the same two forces — labor scarcity and supply-chain de-risking — and they're right to. The open question isn't whether the tailwind is real. It's which layer of the factory you can actually own before five other YC companies own it first.

The landscape today

1. The manufacturing back office is the new gold rush — and it's already crowded. The single densest cluster in this cohort is AI for the order-to-cash and procurement paperwork that mid-market manufacturers still run on email, PDFs, and a 20-year-old ERP. Arzana is building an agent-plus-ERP for US manufacturers and distributors; Korso runs autonomous quoting and supplier-communication workflows inside existing systems; Walter reads documents and writes validated data straight into the ERP; Smartbase does the same wedge but narrowed to metal finishing shops. Four near-identical theses in one batch is not a coincidence — it's a land grab, and the differentiation is going to come down to vertical depth and ERP-integration pain tolerance, not model quality.

P26

Arzana

AI automation for the manufacturing office

ActiveManufacturing Order-to-Cash5 ppl

P26

Korso

The Intelligence Layer for Manufacturing.

ActiveManufacturing Operations AI3 ppl

P26

Walter

AI Employee for Manufacturing Operations

ActiveManufacturing Operations AI

P26

Smartbase

AI-native ERP for manufacturers

ActiveAutomated Order Entry2 ppl

2. Sourcing and compliance are being rebuilt as AI-native brokers, not marketplaces. The last cycle tried to build supplier *marketplaces*. This cohort skips the two-sided cold-start problem by becoming the broker itself: Andustry manages end-to-end industrial parts sourcing for manufacturers; Saudara AI does the same for brands diversifying out of China into Indonesian factories — the purest onshoring/friendshoring play in the batch. The regulatory shadow of moving physical goods across borders is its own wedge: Complir is building "Vanta for physical products," automating the compliance documentation that every reshored or re-routed product line now generates. On the defense end of the same thesis, GUILD is attacking the defense supply chain directly as a "neoprime."

P26

Andustry

We help manufacturers find suppliers for industrial goods.

ActiveAI Industrial Sourcing Broker2 ppl

P26

Saudara AI

AI Native Sourcing Broker For Overseas Manufacturing

ActiveSupply Chain Sourcing Broker2 ppl

P26

Complir

Vanta for physical products

ActiveProduct Compliance Automation13 ppl

3. The factory floor itself: robotic labor priced like headcount. The hardware companies in this wave share one commercial insight — sell the robot as a monthly employee, not a capex project. InLoop Robotics rents bimanual warehouse robots on a flat monthly fee with remote human recovery; Advanced Metal Research targets the welding shortage, arguably the most acute named skills gap in American manufacturing; Prototyping.io goes further upstream and automates fabrication itself, turning CAD files into CNC-machined parts with AI design-for-manufacturability. Origami Robotics is the bet that general manipulation models eventually subsume all the task-specific players.

P26

Advanced Metal Research

Machine intelligence for American Welding

ActiveRobotic Welding Cells3 ppl

P26

Prototyping.io

Autonomous manufacturing for mechanical parts

ActiveAutonomous Manufacturing2 ppl

P26

InLoop Robotics

Staff your warehouse with robotic employees today

ActiveAI Warehouse Robotics4 ppl

W26

Origami Robotics

Manipulate Anything Robot

ActiveDexterous Manipulation Robots5 ppl

4. The operational nervous system: maintenance, monitoring, and logistics intelligence. Once production is reshored, the scarcest resource is the person who knows why machine four is vibrating. Pairio is building AI-native maintenance software that captures retiring technicians' knowledge before it walks out the door; Control Seat layers AI on top of existing SCADA systems for root-cause analysis and predictive maintenance; Transload turns the security cameras already hanging in every loading dock into a freight-dimensioning system — a genuinely clever zero-hardware wedge. Haladir plays the same game at network scale, unifying logistics data into one decisional layer.

P26

Pairio

AI-native industrial maintenance software

ActiveIndustrial Equipment Maintenance AI3 ppl

S26

Control Seat

Control and Monitoring of Industrial Operations

ActiveIndustrial Operations Monitoring2 ppl

P26

transload

Measure freight items with security cameras

ActiveLogistics Freight Measurement3 ppl

The cohort signal

This is a deliberate, accelerating program bet. Spring 2026 alone contains the back-office cluster (Arzana, Korso, Walter, Smartbase, Hexa), both sourcing brokers, the welding and CNC plays, and the maintenance tools — easily fifteen-plus companies on one thesis in one batch. Winter 2026 added Origami Robotics and Haladir; Summer 2026 is already seeding Control Seat and GUILD. Batch-over-batch, the bets are also moving down the stack — from paperwork automation toward physical execution and real-time control. Notably, a striking share of the cluster carries rename histories — Walter, Pairio, and Saudara AI among them — founders are actively steering *into* this thesis from adjacent ideas, which is itself a market signal.

Lessons from the last cycle

The veterans tell a sobering, useful story. Flexport proved a software-defined logistics company can reach massive scale — and also that freight is viciously cyclical and operational, not a pure-software margin business. Nowports showed the regionalization thesis works: a LATAM-focused forwarder rode nearshoring before it was a headline. GrubMarket is the quiet counterexample to venture impatience — owning an unglamorous vertical supply chain end-to-end compounds for a decade. And Faire demonstrated that in B2B trade, the winner isn't the best catalog; it's whoever takes on working-capital risk for the demand side. The shared lesson: in atoms-adjacent businesses, the moat is operational liability you're willing to absorb, not the interface.

If you're building here

Openings worth taking this quarter:

• Pick a trade, not "manufacturing." Smartbase (metal finishing) and Advanced Metal Research (welding) show the pattern: the generic horizontal is contested, but every named trade — heat treatment, casting, electroplating, CNC programming — has its own workflows, ERPs, and nobody serving it. There are dozens of these niches and roughly one startup each.
• The reshoring compliance layer. Complir is nearly alone on physical-product compliance, and the surface is enormous: country-of-origin verification, NDAA/defense supplier qualification (the gap between GUILD's neoprime ambition and a 40-person machine shop), tariff engineering. Boring, mandatory, sticky.
• Sell captured expertise as uptime. Pairio and Control Seat point at the real crisis — the retiring-technician brain drain. The version that wins prices against downtime cost, not per seat.

Tarpits to avoid by name: generic "AI order entry / quoting on top of your ERP" — Arzana, Korso, Walter, Smartbase, and Hexa are all already in market, and the fight is going down to integration grind and distribution; and yet another supply-chain visibility/optimization layer without an execution wedge.

What you'd have to believe: that reshoring policy survives electoral cycles; that mid-market manufacturers — historically the last buyers of software on earth — are finally forced to buy because they literally cannot hire; and that agents can be made reliable enough to write into a production ERP without a human-in-the-loop tax that eats the margin. The cohort is betting yes on all three. On the first two, I'm with them. The third is where the next two years of winners and casualties get decided.