twentyfour26 is training its autonomous factory by making real parts

twentyfour26 wants to build factories that can manufacture a new part without requiring a person to work out every machining step. Its starting point is deliberately narrower: run its own machines, accept jobs that fit a standardized production cell, and use the work to improve the software.

The San Francisco company is taking orders for hardware components while developing models that interpret CAD files and decide how a part should be made. That puts actual production inside the research process. A machining plan eventually has to survive contact with a cutting tool and a piece of metal.

In its launch announcement, the team describes a cell that combines machinery, tooling, workholding and inspection under one system. The eventual goal is to handle varied designs with much less setup work, rather than only repeating a single part at high volume. The company says it is already supplying Bay Area hardware teams, with some orders completed in as little as 48 hours.

That turnaround is a company claim, not a promise that every uploaded design ships in two days. The ordering workflow starts with a STEP file, followed by a fit check, questions about the design and a quote. Production follows once the customer approves the plan. The site makes capacity and suitability part of the qualification.

There are useful limits in that description. The service standardizes materials, stock, tools and how parts are held during cutting. Its listed default material is 6061-T6 aluminum, with alternatives available on request. The site gives a standard machining tolerance of 0.005 inches and says tighter requirements need a drawing review. These details make the offer easier to evaluate than a general claim of automated manufacturing.

Standardization also explains the technical approach. Reducing the number of possible machine configurations gives the software a more bounded problem. The trade-off is that customers must first fit the available process, rather than expect a universal factory to accept any geometry or material.

The founders bring experience on both sides of that boundary. Hemant Mahendru, the CEO, studied aerospace engineering at the University of Illinois Urbana-Champaign and has worked across physical product development. Sam Mathew, the CTO, has built hardware and software for hardware businesses. Their company profile lists twentyfour26 in Y Combinator’s Fall 2026 batch.

The team’s launch video introduces the factory project, but the public materials do not establish that the entire production process is already autonomous. The founders still describe teaching the system to take on more of the work.

For a hardware customer, the near-term decision is therefore fairly practical: does the shop accept the design, meet the tolerance and deliver on the agreed schedule? For twentyfour26, every accepted job also provides another test of whether a standardized cell can learn to manufacture a wider range of parts.