Faraday launches company-specific AI that runs factory workflows end to end

Faraday launched on Y Combinator Fall 2026 on 29 October. The YC launch page had 5 votes in its first hours. The company offers what it calls "company-specific AI and infrastructure" for manufacturing workflows. The product runs from the back office to the shop floor. The named founder is Oscar Rangel. The YC post is written by the co-founder, who is not named by first name on the page.

The product covers six workflows at launch. Engineering reviews check drawings, bills of materials and revisions against company standards. Quoting costs the job from the company's history and drafts the quote. Production scheduling replans jobs around machines, materials and people. Raw material inventory keeps components ahead of the schedule. Accounts payable matches invoices to purchase orders and receipts and routes approvals for sign-off. Sales handles orders in a form a factory can act on.

How Faraday says it works

A factory's work spans disconnected systems, machines, spreadsheets, email, and what the launch calls "tribal knowledge". Faraday connects these so agents can run across them. The agents also learn. Each workflow adds to a shared foundation that grows with the business. Faraday builds reinforcement-learning environments around each company's processes and exceptions, the launch says. Agents improve through real-world outcomes and human corrections.

The company sells to industrials. The launch names distributors, automotive suppliers and precision manufacturers as its customer types. The product site, faradaycompute.com, lists more workflows. These include uptime monitoring, machine integration, predictive maintenance, pricing optimisation, inventory optimisation and order entry.

Why the pitch is different from a copilot

A copilot writes an email or drafts a document. The person using it still has to send the email and type the next one. Faraday is pitched as a different shape. The agent reads the systems that already run the factory. It writes a quote or a schedule. The person approves it. If that works, the saving is not minutes per email. It is minutes per job, across every job the factory runs.

The size of that gain depends on how often the agent is right the first time. A machine-shop quote has 20 to 100 lines on it. A schedule change touches orders, suppliers and the people on the floor. Say a model-written quote is right nine times out of ten. That still leaves one quote a day for a person to redo. Say a model-written schedule is right nine times out of ten. That still leaves one plant day a week where somebody redoes the plan by hand. The honest number for a buyer is how often the agent is right, measured on that buyer's own data.

What is missing from the launch

No customer logo. No revenue figure. No published duration a plant has run on Faraday. No measurement of the agent's accuracy on quoting, scheduling or any other workflow. The ask on the launch is for introductions to industrial leaders. That is the usual shape for a Fall 2026 launch, which is early.

A reader at a manufacturer, a distributor or an industrial software team is the audience here. The question to ask is not whether a model can read a bill of materials. It can. The question is whether it can read this company's bills of materials, with this company's conventions and exceptions. That is the test the product site is making.

Source

• Primary source: Faraday on Y Combinator.
• Product site: Faraday.