A 0–3 score across the 8 moats, anchored to comparable companies. It surfaces where Faraday's defensibility is real, where it's thin, and where the story outruns the structure.
Faraday scores 3 out of 24, with no moat at 2 or higher. That is normal for a two-founder company in YC's Fall 2026 batch with no named customers yet, and it is not a verdict on the product. The design is thoughtful: workflow mapping across ERP, MES and WMS, customer-specific RL environments where agents practice before they act, a shared runtime for people, agents and machines, and a public benchmark. But each of those pieces is either per-customer, built on a model and systems of record Faraday doesn't own, or open to a well-funded rival doing the same thing. The work here is foundational: finding the first wall this design can actually hold.
Faraday calls itself an industrial aOS and a “unified control layer,” and its Research page leads with a benchmark where its agent scores 96.2 against 75.2 for the best frontier model. Both claims imply the position an operating system holds: the layer everything else depends on. Structurally, Faraday says it sits on top of fragmented systems of record, so the ERP, MES and WMS stay the record, and the vendors who own them (SAP, Infor, Epicor) are shipping their own agents into that same layer this year. The benchmark lead is real engineering, but it is a harness on GPT-5.5, scored on 74 synthetic tasks Faraday wrote itself, and the page is honest about that. In AI-native operations software, the loudest claim is usually the agent's intelligence, while the harder-to-copy asset is whatever the agent learns inside a customer that a rival would have to rebuild from scratch.
Each moat is scored 0–3 for structural durability — not for how useful or popular it is. The tier is set by how many moats reach 2+, not by the total.
| Moat | Meter | Score | One-line take | |
|---|---|---|---|---|
| 01 | Data | 1 | Customer-specific RL environments are rich, but per-customer by design. | |
| 02 | Workflow | 1 | Built to embed in approvals and exceptions, but no live deployments are visible yet. | |
| 03 | Regulatory | 0 | No certifications or licensed position visible, and its segments have none to win. | |
| 04 | Distribution | 1 | YC badge, founder networks and an open benchmark: borrowed reach, not an owned channel. | |
| 05 | Ecosystem | 0 | Consumes ERP, MES and robot-controller interfaces; nobody builds on Faraday yet. | |
| 06 | Network Effects | 0 | Each customer gets its own environment; no value moves between customers. | |
| 07 | Physical / Infra. | 0 | Dispatches to robots and machines it doesn't own; pure software. | |
| 08 | Scale | 0 | Two people and customer-specific builds; no cost or pricing leverage. | |
| Total | 3 / 24 |
The tier is set by counting how many moats score 2 or higher — not by the total out of 24. One strong moat isn't a stack.
| Exposed← This company | 0 moats at 2+ | No defenses; open to attack from every direction. |
| Partial | 1–2 moats at 2+ | Some defense, but below the stack threshold. |
| Stacked | 3 moats at 2+ | Stack threshold met; defensibility begins here. |
| Fortified | 4+ moats at 2+ | Multiple moats compound; the strongest state. |
Faraday has 0 moats at 2+, which lands it in Exposed.
Scores are anchored against comparable companies in the same category — not isolated opinion.
This is the most promising part of the page. Faraday combines operating procedures, enterprise data and expert corrections into customer-specific RL environments, and plans to bring live machine data into the same feedback loop. But each environment is built around one company's systems, approvals and exceptions, so what it learns stays inside that customer: a switching-cost argument, not a compounding Data moat. Clay and Attio score 1 here for the same structural reason; Gong reaches 3 only because its conversation corpus compounds across every customer. The public benchmark runs on synthetic records, so it isn't proprietary data either. A 2 would need an owned dataset of real industrial workflow outcomes at scale; a 3 would need those outcomes to make the agent better for every new customer.
The design targets real operational work: supplier delays, quoting, production scheduling, SPA claims and accounts payable, with human approvals inside the loop. If an agent ends up running a distributor's order entry, removing it would hurt. Today there are no named customers, case studies or retention data, and the product deliberately sits on top of the ERP rather than replacing it. That puts Faraday next to 11x at 1: the pitch is embedding, but lock-in hasn't formed. Calendly's 2 came from a daily ritual customers already ran through it; Salesforce's 3 comes from being the record itself. A 2 needs live workflows a customer would face a real migration to pull out.
The site lists no SOC 2, ISO, ITAR or CMMC posture, and its target segments, general manufacturers and distributors, carry no license that keeps competitors out. Even once Faraday adds SOC 2, that is the cost of entry, which is why Attio, Clay and HubSpot all sit at 0 here. Stripe's 2 comes from money-transmitter licenses competitors structurally can't clear; nothing comparable gates order entry or scheduling software. For a horizontal manufacturing and distribution platform, this moat is structurally unavailable rather than missing.
Acquisition today runs through a “Book a walkthrough” button, the YC badge and the founders' backgrounds: factory systems for Worthington Steel, medical devices at Stryker, enterprise engineering at Microsoft, founding engineer at Handshake AI. The open benchmark on GitHub is a smart credibility asset that could pull in technical buyers. All of it is borrowed or early, the same 1 that Attio, Day.ai and Rox carry on fundraise and founder reputation. Apollo's 2 comes from a freemium loop that acquires on its own; HubSpot's from an inbound engine. A 2 would need a channel Faraday owns that brings in industrial buyers without a founder on every call.
Dependency runs one way: Faraday connects into ERP, MES, WMS and robot controllers, and no one yet structures a product around Faraday. The benchmark invites outside teams to plug in their own agents, which is a small open surface, not a platform. Day.ai and 11x score 0 here as integrators; Calendly's 1 reflects an API surface others use; Salesforce's 3 is AppExchange, a whole labor market built on top of it. A 1 or better would need third parties shipping against Faraday's runtime, not the other way around.
Each customer's agent learns inside that customer's environment, so a new manufacturer joining does not make the product better for an existing distributor. “Learns how enterprises operate” describes a within-customer feature, which is the most common way this moat gets over-claimed. Rox, Day.ai and Qualified score 0 for the same reason. Apollo's 2 comes from contributory data that flows across its base; G2's 3 is a true two-sided network. A 1 or 2 would need a mechanism where what one customer's agents learn improves everyone else's, without exposing anyone's operations.
The robotics layer is real in the design: the runtime authorizes a task, a robot controller executes the motion, and status flows back to the business systems. But the controllers, machines and plants belong to the customer, and Faraday runs on commodity cloud. That is the universal 0 for software, the same score every company in the AI-native GTM benchmark carries except Salesforce, whose Hyperforce residency layer earns a 1 and marks the ceiling. Faraday's model coordinates hardware rather than owning it.
With two employees and an RL environment built per customer, Faraday's cost base grows with each customer rather than spreading across them. That is the high-touch pattern that keeps Attio, Rox and 11x at 0, and it is where the ERP vendors hold the edge: they can bundle agents into licenses customers already pay for. ZoomInfo's 2 rests on revenue and operating margin a sub-scale rival can't match. The 2.6x lower run cost Faraday reports against its GPT-5.5 baseline is an efficiency claim, not scale leverage; a 2 would need unit costs that fall as the customer count rises.
Not every moat fits every business. Three buckets narrow the field — they don't name which available moat to pursue. That's the Playbook's job.
Faraday has a thoughtful architecture and founders who know the shop floor. What it does not yet have is a wall. The Defensibility Stress Test pressure-tests each of these scores, runs Faraday head-to-head against 3 to 5 players across factory agents, ERP-native AI and process intelligence, and identifies which foundational moat its model can hold first. Once a foundational moat exists, the Stacking Moats Playbook takes over: matching the stack to a durable archetype and naming the missing leg.
Plus a head-to-head competitive analysis against 3 to 5 players in Faraday's field.
The rigorous follow-on to this scorecard, run on your own company with the full methodology visible.
Thanks — your scorecard and a short email series are on the way, and I'll be in touch within one business day. If you want the rigorous version, the discovery call on the left is the fastest way to start.