Vibe Manufacturing: The Next AI Opportunity Has a Shop Floor

The bolts on a boat lift were scratching someone's pontoon. The solution was obvious: cover them with custom bumpers. Designing those bumpers had sat on the owner's to do list for months.

Then the owner found an AI design tool.

In a Hacker News discussion about Adam's open source CAD software, the user reported that five minutes of prompting had put four custom bumpers on the printer. A small irritation, postponed for months, had become a physical project almost immediately.

The owner already understood the problem. What changed was the ability to turn that understanding into a shape a machine could make.

For years, the internet has been full of people who know exactly what they wish existed. A better camera mount. A replacement for a discontinued plastic fitting. A bracket that would let two otherwise incompatible devices work together. Between the idea and the object sits a collection of skills, suppliers and unfamiliar decisions. For a small enough project, abandoning the idea has usually been easier than assembling all of them.

That calculation is changing from both directions. AI is making the design easier to attempt. Digital manufacturers let customers upload a file, choose a material, see a price and buy a custom part almost as casually as they order something already on a warehouse shelf.

The two developments are beginning to meet. A person can start with knowledge of a problem, use AI to help express the solution and rent the equipment needed to make it.

Call it Vibe Manufacturing.

The ambition is a direct relationship between the creator and an AI agent. You describe the object, inspect what the agent designs, ask for changes and approve a parts order. The agent operates the software and coordinates the suppliers. The factories make what you created.

The phrase has already appeared in an Automation Alley playbook and a Genpire essay this year. The more interesting evidence is the behavior behind it: makers producing printable objects through conversation, engineering software acquiring agents, and manufacturers accepting digital orders. Separate advances are starting to look like stages of the same journey.

The production infrastructure is already a serious business. In May, SendCutSend announced a $110 million investment at a valuation of $1.01 billion. Xometry reported $215 million in marketplace revenue in the second quarter, up 45 percent from a year earlier. AI arrives at an industry that has spent years learning to fulfill a digital order.

My bet is that the next expansion comes from people who previously would have abandoned the project before asking for a quote. They bring the demand. The emerging stack supplies more of the ability to act on it.

For entrepreneurs, this creates an unusually tangible way to participate in the AI wave: open a small manufacturing shop that turns those new designs into prototypes. As software makes it easier to create, someone still gets paid to make the object.

SendCutSend has made custom manufacturing feel like something with a checkout button.

A customer uploads a file, selects material and processing options, receives a quote and orders. The company cuts sheet material, bends it, inserts hardware and applies finishes. Its catalog includes CNC machining and a defined welding service. Laser cutting has no minimum quantity. Standard orders generally ship within two to four business days once the files are ready for cutting, with extra time for additional services.

The customer can discover whether an idea is affordable before finding an estimator willing to spend time on it. That changes the economics of curiosity. A small project can reach a price instead of becoming a series of unanswered inquiries.

A defined menu of materials, dimensions and operations makes the interface possible. Production knowledge has been translated into rules: the scale of the file, the position of a hole, the geometry of a bend. SendCutSend's applications team reviews orders, while the customer does much of the preparation through the website.

Those same rules are useful to an AI agent. It can consult the manufacturer's requirements while drawing, then move a hole or alter a flange before requesting a quote. The open source Text to CAD project already includes a file checking workflow for SendCutSend. The design conversation is beginning to reach the factory.

SendCutSend also offers a parts builder for configurable shapes. AI could extend that invitation to more unusual designs. The customer arriving at checkout increasingly needs to know what the part should do; software and services can supply more of the knowledge about how to make it.

Founded in 2018, the company has manufacturing capacity in Reno, Paris in Kentucky and Arlington in Texas. Founder Jim Belosic called himself “the most impatient person I know” in the funding announcement. SendCutSend is raising capital to add equipment and facilities that can accommodate that impatience.

Others have built different versions of the same bridge. Xometry routes work across a supplier network. Protolabs combines factories with its manufacturing network. Fictiv, which joined MISUMI in 2025, manages manufacturing and supply chain coordination. OSHCut translates sheet metal designs into work its equipment can form, including checks for bending constraints.

The factories are becoming easier to buy from just as the ability to design for them is spreading.

In software, a similar change happened with remarkable speed. A person could describe an application, ask an AI to write it, try the result and request changes. Vibe coding made the experience familiar: an intention could become something usable before its creator understood the underlying code.

The CAD story has been quieter, but its progression is revealing. When Zoo launched Text to CAD in late 2023, it generated individual files. By 2025, its software was generating parametric models: widen a bookshelf and its other dimensions adjust with it. Zoo reported cutting errors in the generated code from 50 percent to 16 percent over several months. Then came Zookeeper, an agent that can edit an existing model, inspect its changes from different angles and look up references.

Each advance lets the conversation carry more of the project. Make it wider. Move the mounting holes. Leave room for the connector. The creator explains the change while the software handles more of the translation.

Astra, OpenAI's GPT 6 model, brings coding, image understanding and computer use to existing engineering software. CAD can be operated through code: a model writes instructions that construct a part, inspects the result and revises those instructions. A photograph of a prototype can become feedback for the next version.

One maker's account in OpenAI's developer community shows what that feels like. After building software for a small ESP32 display, the maker needed a stand for the screen. They had owned a 3D printer for about a week. Through Codex, Astra generated CAD with Python and CadQuery, showed the design in Blender and helped prepare files in Bambu Studio. The maker supplied measurements, printed parts and sent photographs back. Bulky proportions and a troublesome cable clip prompted further revisions. Someone new to CAD and printing ended up with a finished display stand on their desk.

The loose clip became a design change in the same conversation. The project could move across tools and return from the physical world with information the model could use. That is the beginning of a much richer relationship between a creator and engineering software.

Adam's open source CADAM tool also generates parametric models. Specialist products give this behavior a place to happen; general models such as Astra offer another route through tools that already exist.

The improvement is visible with the software held constant. In gNucleus's September evaluation, Codex using GPT 5.6 Sol scored about 30 percent on 100 CAD tasks. Astra scored about 57 percent with the same Codex version. The task set includes creating parts, editing models and working from engineering drawings.

I expect the frontier to move toward coordinating more of the project. Component dimensions, supplier rules and quote feedback can become inputs to the same conversation that creates the geometry. The creator could ask for an entire device, with the agent developing its parts, selecting components and preparing an order within a budget.

Imagine a traveler who wants a folding phone stand that works on a cramped airplane tray, accommodates a thick case and keeps the charging cable out of the way. They sketch it on a napkin. This is our fictional consumer product: a few plastic pieces, a hinge and a problem its creator understands.

Before conversational AI, the traveler could learn CAD or pay someone to translate the sketch into a design. Cad Crowd advertised freelance design rates of $30 to $65 an hour in 2018; its current hourly service lists $45 to $65. The expensive part of a small project could arrive before a printer had made anything.

Consider a simple budget. Twenty hours of paid design at $50 an hour, plus a $100 allowance for fabrication and delivery, puts the first prototype at $1,100. Those are assumptions for our stand, but they expose the decision: is an untested accessory worth a four figure commitment?

Today, the traveler can ask an agent to create the CAD, compare variations and change the angle or cable channel. The person still checks the dimensions and tries the printed result. If that approach reduces the paid design and review work to four hours, adding a $20 software allowance and the same $100 fabrication budget brings the total to $320.

The physical order is already within reach. Formlabs' Form Now service publishes example prices of $16.93 for a knob and $51.02 for a snap fit box, and says a high percentage of orders ship within one to two days. A creator can rent professional printing without buying the printer.

The next step is an agent that carries the whole stand through the process. It asks for the phone's measurements, produces an editable design, chooses a material and checks the hinge and stability. The creator asks for a wider base. The agent revises it, compares quotes and prepares an order for approval.

For an object this simple, I expect the bill eventually to approach the cost of making and delivering the prototype. In our budget, half an hour of paid review, $20 of software and that unchanged $100 fabrication allowance would total $145. The design work has fallen far enough that a personal annoyance can justify an experiment.

The physical order keeps its $100 budget in every version. Most of the saving comes from getting the idea ready to manufacture. A lower total bill can bring the shop more customers without forcing it to charge less for the physical work.

This pattern is already visible in the machinery. In a 2024 Formlabs case study, Unilever and packaging manufacturer Serioplast used printed molds to make pilot runs of 200 bottles. Tooling that cost $2,500 to $10,000 in machined metal cost $500 to $1,000 with printed molds. Pilot testing took two weeks instead of six to eight. The team could try more designs before committing to production tooling.

AI extends that freedom to the design itself. The traveler receives the stand, notices that it tips when the phone is tapped and sends the agent a video. A wider base becomes the next file sent to the printer. The useful discovery happened on an airplane tray; the next revision happens in a conversation.

I expect the first visible boom in vibe manufacturing to be a boom in prototypes. More people will attempt a product, and each will be able to try more versions. The first order earns the manufacturer a customer. The revisions give that customer a reason to return.

One of the biggest beneficiaries of vibe manufacturing could be a small shop down the road. Its next customer has never opened CAD. Their agent arrives with the files, checks the price and prepares the order. A hobbyist's idea becomes work on the shop's schedule.

That changes how a manufacturer wins business. The creator wants to try the phone stand; the agent needs to know who can make it well and get it there quickly. SendCutSend has moved much of that conversation into its website. Xometry's matching models use machine characteristics, supplier quality and shipping performance. A local shop could compete by making its capabilities and available capacity just as easy to discover.

Prototype buyers have a different calculation from buyers ordering a production run. A dollar saved on each of 100,000 units is $100,000. On a single prototype, it is one dollar. Getting the object in time to test it, photograph it or put it in front of a customer can be worth far more.

Protolabs' 2024 report, based on a survey of 767 engineers and designers, captures that distinction. Speed ranked first when choosing an early prototyping partner. Quality led in later prototyping. Eighty two percent of respondents were looking for ways to accelerate development. For a shop selling prototypes, a dependable delivery date and a part that fits are part of the product.

This is an opening for small local manufacturers. A nearby shop can offer pickup, inspect a troublesome fit with the customer and turn a revision around without another shipping journey. It can charge for finishing, assembly, measurement and rush capacity. The traveler saves on developing the design and can spend some of that saving on a better physical prototype.

A shop does not need every manufacturing process. A few printers, finishing tools and measurement equipment can serve one class of product; a CNC machine and reliable outside partners can serve another. The valuable specialty is making a particular kind of prototype quickly and consistently. Repeat customers give the shop a growing record of which materials, fits and processes work.

The economics deserve attention. Protolabs reported a 44.5 percent gross margin on $533.1 million of revenue in 2025 across its manufacturing business. That is before operating expenses, but it demonstrates that custom manufacturing can support substantial value above the direct cost of production. A small shop's opportunity is to earn a premium for responsiveness while keeping quoting, setup and rework under control.

The demand is expanding already. Xometry reported 89,557 active marketplace buyers in June 2026, up 20 percent in a year. That is the existing custom manufacturing market; conversational design could widen it to people who have never commissioned a part.

The industry is already built from small businesses. Census figures cited by Connor Love and Collen Larson show that 83 percent of machine shop firms operating throughout 2022 had fewer than 20 employees. AI could bring these businesses orders from a much wider population of creators.

I would build this business around a category and a cluster of customers: consumer accessories near a design community, fixtures near an industrial district, research parts near a university. Stock the materials those customers use, reserve capacity for revisions and make ordering easy for their agents. Start with the process you can execute well and buy the remaining operations from other shops.

My bet is that well run local specialists could see demand expand dramatically. They can earn from the first attempt, the improved version and the small batch that follows. A customer who once needed to justify a design project can now justify a prototype. For the shop owner, that is the opportunity: many more people reaching the counter.

The boat owner needed four bumpers. A company designing for a national market would have little reason to develop that particular solution. For the owner, making four was enough.

As the cost of developing a design falls, fewer customers are needed to justify creating it. A product can be worthwhile for a small community, one workplace or the person making it.

The demand is already there. OSHCut's founders wanted 53 different sheet metal parts for a motion simulator, one of each. Conventional shops were reluctant to take the job. OSHCut grew out of the difficulty of getting those parts made. AI could let many more people reach the point of submitting such an order.

3D printing gives them a place to start. A creator can print the stand's hinge, try it and ask the agent to change it. A design that will eventually be made in metal can first be tested for fit on a desk. The prototype becomes a cheap way to learn what to order.

The same appetite for particular parts exists inside large companies. BMW reported producing more than 400,000 parts annually through 3D printing worldwide in 2024, including production tools and customized robot grippers. Even mass production needs objects designed for specific jobs.

Cheaper attempts also change how many objects get made along the way. In a 2019 Formlabs account, the founders of homebrewing device Plaato reported making roughly 1,000 clear part prototypes, sometimes six a day. Printing let them test, measure and revise while the next part was being made. AI could spread that habit to people who can describe a design but cannot draw it themselves.

A successful personal project can then become a small business. SendCutSend lists an example aluminum part, with cutting, bending, deburring and welding, at $55.62 for one and $8.37 each at 100. The creator can order a first part, then reach better unit economics if other people want it. Production grows with demand.

E commerce gave obscure products a way to find buyers. AI and digital fabrication could make more of those products worth designing in the first place. The useful accessory for an uncommon camera setup, or the fixture that fits one workshop, gets a chance to exist.

Prototype demand can grow in two directions at once. Ten times as many creators, each trying twice as many versions, would mean twenty times as many builds. That is an illustration of the multiplier, not a market forecast. Even projects that never become a retail business can produce paid orders for the shops helping people test them.

The seductive phrase for this future is AWS for atoms.

The idea predates today's AI tools. Fictiv was described as an AWS for hardware manufacturing in 2019. Shapeways launched a public 3D printing API in 2013, inviting developers to create applications that produced physical objects. Programmatic access to factories has been an ambition for more than a decade.

AI adds the ability to translate an ordinary description into more of the instructions those factories need. The person can begin further away from a manufacturing file and still reach an order. Physical creation gains an interface that understands what the creator is trying to do.

A software founder can build with cloud computing without owning a data center. A physical product creator could use a network of manufacturers without owning a machine shop. The agent supplies more of the ability to design for that network, making its equipment useful to people who have never prepared a production drawing.

Engineers will help build the tools, simulation methods, component libraries and design rules that make this possible. Their expertise can reach creators through the software, while expert review and specialist testing can be purchased for projects that need them. The person developing a simple object can keep working directly with their agent.

The commercial effects follow the creative ones. A photographer makes an accessory for a particular setup and other photographers want it. The creator returns to the agent for a small run, then a variation, then a family of products. A business forms around understanding the customer, with manufacturing purchased as demand appears.

That is the part of the vibe coding analogy I find most persuasive. The tool changes who gets to create from scratch. Taste, imagination, an audience or knowledge of an obscure problem becomes enough to begin. The creator can reach a prototype with less money and less expertise committed before anyone has used it.

A decade from now, I suspect designing and ordering certain physical objects will feel as ordinary as building a small website does today. People will create things for their homes, hobbies and workplaces because describing the need is a practical way to start making the solution.

For the entrepreneur opening a workshop today, the wager is on those people. The agent makes them capable of commissioning the work. A responsive local manufacturer gives them a reason to keep creating.

The boat owner needed four bumpers and had postponed making them for months. Five minutes of prompting put them on the printer. The prospect of that experience spreading to more useful, more complicated objects is what makes vibe manufacturing worth watching.

The next thing you wish existed could become something you make.