Manufacturing’s Next AI Wave: Notes from the Frontier

See how the emerging AI manufacturing landscape could transform electronics manufacturing operations through intelligent automation, data integration, and decision-making tools.
by Shirin Khanna

I’ve spent the last few months talking with founders building AI for manufacturing, from pre-seed teams to growth-stage companies. As an operator at a contract electronics manufacturer, I wanted to understand where AI moves the needle on the shop floor and on the P&L. I came away with a clear view: This is one of the most underrated spaces in AI right now, and the next wave will look very different from the current one.

Most conversations about AI in manufacturing today focus on robots. Robotics startups pulled in over $10 billion in venture capital in 2025. Mind Robotics, a Rivian spinout, crossed $1 billion in cumulative funding inside a year.

Robots are easy to film and easy to fund. That’s why they get the headlines.

For most operators, though, robots represent only part of the manufacturing AI landscape. The other part lives in supplier inboxes, RFQ spreadsheets, bills of materials (BoMs) with 300 line items, ERP screens, CAD files, and shop-floor dashboards. That’s the part I want to talk about.

A Quick Map of the AI Manufacturing Space

The non-robotics AI manufacturing landscape falls into a handful of problem areas.

Manufacturing intelligence and shop-floor visibility. This is the most mature subcategory. Augury ($369M, Series F) leads in predictive maintenance. First Resonance ($19M, Series A) offers a manufacturing operating system for high-mix, low-volume electronics production. Guidewheel ($30M), Tulip ($152M), and Sight Machine ($87M) cover overall equipment effectiveness (OEE), shop-floor applications, and industrial analytics. Zelos Cloud represents the seed-stage entrants.

Visual inspection and quality. Instrumental ($80M, Series C) is the most established player in PCB and electronics quality inspection. Andrew Ng’s Landing AI ($57M, Series B) brings a data-centric approach to small datasets. At the seed stage, Bucket Robotics uses CAD-trained models to deploy visual inspection without manual labeling.

RFQ and quoting. Paperless Parts ($51M, Series B) has tackled this problem for years. The seed-stage wave (Korso, Soff, Arzana) is automating the full quote-to-order workflow.

Procurement and supply chain. This area has the largest cluster of seed-stage activity. Mandel, Lumari, Comena, Vendra, and Cavela are each building AI agents that work with supplier emails. At Series A, Didero ($37M, NYC) is bringing the same approach to enterprise scale.

Design, compliance, security, and ESG. These areas have less activity but still matter. Diode Computers and Prototyping.io focus on design automation. Noetic and Saphira address hardware compliance. Binarly specializes in firmware security, while Coolset focuses on ESG reporting.

One area stood out as underserved: high-mix, low-volume production scheduling. Most manufacturing execution system (MES) tools target high-volume, repetitive production runs. For an electronics manufacturing services (EMS) provider juggling more than 50 different jobs a week across SMT, wave soldering, and automated optical inspection (AOI) lines, spreadsheets still handle much of the scheduling. No seed-stage AI startup currently focuses specifically on this challenge.

Plot these companies by stage and funding, and the shape of the category jumps out: a dense seed-stage cluster on the left and a handful of growth-stage players up and to the right. The companies entering the market today are defining the category.

What I Heard Across the “AI for Manufacturing” Conversations

Three patterns repeated in nearly every meeting. These observations don’t reflect any specific team. They reflect the category’s current state.

Companies are building AI around unstructured manufacturing data. Suppliers won’t adopt portals, engineers won’t redo their CAD workflows, and shop-floor operators won’t use a new screen. So the dominant architectural approach involves reading data where it already lives. Inboxes and PDFs for procurement. CAD files and camera feeds for inspection. Machine-level signals for the shop floor. The same pattern appears across the category: Read what’s already there.

The workflows have converged. What separates teams now is configurability. Within each subcategory, companies now follow a standard core workflow.

• Procurement: ingest → distribute → extract → compare → push to ERP.
• Inspection: CAD → train → deploy → flag.
• Shop floor: sense → capture → surface → recommend.

What separates teams is how easily a manufacturer can encode its own playbook: follow-up cadences, escalation rules, approval hierarchies, and exception thresholds. Every shop runs its own playbook, and the better products let manufacturers keep theirs.

The customer counts are small, which is part of the opportunity. The teams I met range from zero customers to about a dozen. Even the well-funded companies continue to run pilots. I don’t see that as a knock. It’s the most exciting part of the category. The founders shipping products today are writing the playbook for everyone who comes next.

Today’s products make existing manual workloads faster and cheaper. For a manufacturer where manual cognitive work accounts for a meaningful share of operating expenses (OpEx), that translates into real money on the P&L. But executing yesterday’s workflow faster doesn’t necessarily create a fundamentally better business.

The bigger prize lies in expanding what an operator can decide. AI can surface patterns and possibilities operators couldn’t see before, letting them manage categories and programs they couldn’t handle before. That’s not “AI that does my job 10x faster.” It’s “AI that lets me do a job nobody could do at all.”

The teams that define this category long-term won’t necessarily have the cleverest agent architectures. They’ll be the ones that combine customers’ internal data (historical pricing, supplier scorecards, machine logs, quality history, and BoMs) with meaningful external signals: public benchmarks, government data, geopolitical inputs, commodity trends, and historical supply chain disruption data.

That combination helps operators make decisions they could not have reached alone.

Anyone can wrap a large language model (LLM) around a workflow. Very few can build the data graph that makes the model’s output useful enough to act on.

The pitches that impressed me most framed the product around a P&L line, not an architecture diagram. “We save $X per shift at a $100M EMS provider” is a different kind of sentence than “we use a 12-step pipeline with custom embeddings.” This matters beyond pitching. It shapes the product.

Tech-first products optimize for elegance. Business-first products optimize for the customer’s outcome. Over time, the second kind tends to win.

The use case I keep coming back to is an agent that works silently and only interrupts when it has something worth saying. In quality, that’s an agent watching every production line and flagging only the process deviations that deserve attention.

In scheduling, it quietly sequences jobs around a delayed shipment. In sourcing, it continuously scans new suppliers and only alerts you when one offers an advantage over your current vendor.

It’s the opposite of adding a chatbot on top of a workflow. The better model treats AI as a colleague who walks into your office only when something matters.

Companies are building this category right now. The current wave is laying the foundation: moving manual workloads off the page, building the underlying data infrastructure, and codifying the playbooks manufacturers currently manage by hand.

The next wave excites me most. It will bring products that combine internal and external data to help operators make decisions they couldn’t have reached alone. The underlying AI agents will stay quiet until they have something worth saying. And the entire system will focus on business outcomes, not the architecture.

If you’re a founder building in this space or an operator at a manufacturing business thinking about where AI moves your P&L, I’d love to compare notes.End of article content

Shirin Khanna is a board member at Indic Electronics, a contract electronics manufacturer, where she leads supply chain and procurement. She oversees the end-to-end RFQ process, drives digital transformation on the factory floor through SAP HANA, CalcuQuote and digital manufacturing platforms, and leads go-to-market initiatives that have secured multiple enterprise customers and introduced new revenue streams, including design for manufacturing (DfM) services. She focuses on making industrial operations faster, more transparent and more accountable, from sourcing decisions to shop-floor execution.