We’re used to seeing AI in all the obvious applications: software engineering, photo generation, and so on.

But it’s the non-obvious applications that are most interesting! Uptool is an AI tool that deviates from the obvious use cases. Building machines is a great example here. In the past, generating a job estimate to build something would be a manual process. Now, that cost can be easily computed!

Arek and Ethan 🦄

Who builds the machine that builds other machines? Think about the robotic arms that weld your car together on the assembly line. Those machines were themselves built, part by part, by a machine shop most people have never heard of. But we need someone who takes the raw material and creates the metal parts that will later be used to build machines like those. These small manufacturers make up most of the U.S. industry; however, most of them are still untouched by the AI-driven productivity gains reshaping other sectors and still run manually through emails and spreadsheets, a slow process that makes everybody lose money and opportunities.

Uptool builds AI software to automate quoting for machine and fabrication shops. When a customer emails a shop asking what a part would cost, Uptool reads the email and its attachments, the design files and drawings, pulls out what is being ordered and in what quantity, and estimates the material and finishing costs. The shop still decides which machine to run the job on, how long it will take, and what to charge. The platform also tracks incoming requests, keeps a record of who is asking, connects to QuickBooks so an approved quote can turn into an invoice without anyone re-entering it by hand, and prints the job instructions that go out to the floor.

Check it out: uptool.com

Standing out online usually means paying a designer tens of thousands of
dollars. So we did it for you. For free.

The Bolt Template Marketplace is here: real, working apps built by a handful
of the world's best designers and agencies. Dashboards, storefronts, finance
trackers, portfolios, and more. Open one, tell Bolt what to change, publish
it live.

Not mockups. Not themes. Finished apps, ready to make yours.

Browse the marketplace.

This is sponsored content.

Uptool sells a subscription directly to machine and fab shops, priced by shop size rather than usage.

If you’d like to explore opportunities with Uptool, let us know by replying directly to this email!

Benny Buller and Alex Huckstepp have spent their careers in manufacturing, approaching the same problem from opposite ends. Buller trained as a physicist before leading engineering teams at First Solar and Applied Materials. He then moved on to investing at Khosla Ventures and founding Velo3D, the metal 3D-printing company he took public as CEO in 2021. Huckstepp came up on the commercial side, running growth and business development across four back-to-back manufacturing startups, including Machina Labs, ARRIS, Digital Alloys, and Carbon.

No matter which company he worked for, Huckstepp noticed a pattern. A capex-heavy machine (meaning one that demanded millions in upfront investment before it produced anything) took years to be developed and a long time to reach the customer.

Both had spent years around 3D printing and were reintroduced by a VC who noticed they were circling the same space. The conversation turned to building a next-generation tech stack for manufacturers, aimed at helping smaller machine and fab shops keep pace with a new guard of hardware companies operating on a much faster clock. Huckstepp had worked with those shops for years and knew the gap firsthand. They partnered up, and Uptool followed.

Most people have never thought about how a custom metal part gets made. A company designs the part and sends the drawing to a machine shop, one of the roughly 600,000 small manufacturing firms that own the equipment to cut it out of raw material. Before any metal moves, the shop has to quote the job, which requires reading the geometry, deciding which operations it requires, estimating cycle time, pricing the material, adding labor and margin, and sending a number back. Quoting looks like overhead sitting in front of the real business, but it is the business. A slow quote loses the job to whoever answers first, while a wrong one either eats the shop's margin on an underpriced job or costs it the customer on an overpriced one.

The reason it binds so tightly comes down to the mix. A shop running high-mix, low-volume work might quote 50 parts in a month and make each once, so there is no volume to amortize the effort across. We also learned that it cannot be pushed down the org chart either, because pricing a part correctly requires knowing how this shop, with these machines and this operator, would actually make it. That knowledge usually lives with one person, the estimator, who ends up being the owner or the best machinist on the floor, and getting them to quote a job means pulling them off the value-adding work only they can do. 

On top of that, we have the win rate pegged at only a quarter to a third chance. Thus, most quote labor produces no revenue. Faced with that math, shops have no other choice than to quote less, triaging work they doubt they will win or letting Requests for Quotations sit until the buyer stops waiting.

The industry has already priced what fixing this is worth. 40 percent of OEMs would pay a premium of 10 to 20 percent to shorten delivery times by five weeks, while domestic shops now compete against imports on roughly a third of the quotes they send, losing most of those bids on price. U.S. manufacturers are losing on the one axis where they cannot win, in a market that has already said it will pay for the one where they can. Huckstepp explained engineers at next-generation hardware companies like SpaceX, Tesla, Rivian, and fusion startups all share one requirement, and if they cannot get speed locally, they will look elsewhere.

Manufacturing use cases are scoring lower than earlier AI waves would have predicted, because generative models are weak at the numerical optimization that dominates physical production. Scheduling a floor is exactly that kind of problem. Quoting is not. Most of the work is reading unstructured inputs, like a CAD file, a drawing with tolerances, an email with half the requirements buried in the body text, and turning them into structured data. 

In other words, every downstream system in a shop, such as scheduling, purchasing, capacity planning, job costing, and so on, consumes information first assembled during quoting. By connecting all that scattered information, Uptool lets a shop quote faster and get back to the client before a competitor does, so good deals stop slipping away. And it does this without pulling the most important people off the floor.

What makes it work commercially is where Uptool stops. The system automates extraction and the predictable cost math, but the shop still sets run times, machine selection, lead time, and final price. This is deliberate, since costing is one thing but final pricing is an art. A tool that automated pricing outright would ask a shop owner to bet their margin on a model they cannot inspect.

The open question is whether the learning compounds across customers or only within them. Today it is within, because onboarding loads a shop's equipment and rates, and if a user consistently overrides a variable, the software remembers. The cross-customer layer, also known as geometry matching to predict pricing on similar parts, has not been shipped yet. If the shared layer proves substantial, every new shop makes the product better for every other one. If it stays thin, the moat is switching cost on shop-specific data. That is the difference between a good quoting tool and the system every U.S. machine shop runs on.