If you have an iPad, you can vibe code robots. And potentially save your company millions of dollars while you’re at it. Today, New York-based Standard Bots launched its AI platform with new skills for programming robots. The pitch: any frontline worker can set up a robot to do complex tasks in minutes, with no code, no fixtures and no outside integrator.

“It effectively allows you to vibe code your robot,” CEO Evan Beard told me on the Humanoid Daily podcast. “So for those that have tried Claude Code or Codex, you can just talk to the robot and say, ‘Hey, this is what I wanna automate.’”

That’s a massive change from how industrial automation – slow, painful and expensive – typically works.

Traditionally, if you want to automate a task, you buy a robot and then you hire someone to make it do something useful. That someone is usually a systems integrator, and the bill can be eye-watering, even compared to the cost of the robots themselves.

“You have to hire an expert to install your robot,” Beard says. “It often costs more than the robot itself to have the robot installed, and all that the end user can do is press start, stop, and reset. They can’t really edit the program usually.”

Innovation in robot training

This is part of a huge wave of innovation in robot training, which is getting faster, easier and cheaper seemingly by the day. Just this week, Google’s new Gemini 4 Argon taught Apptronik’s Apollo humanoid to play cornhole from a single video of a person playing. It built its own simulation, practiced there, took a few real throws, and then sank seven in a row. In September, Nvidia-backed General Robotics told me its GRID platform cuts robot onboarding time by as much as 99.7%. It uses AI agents for the grunt work of robotics engineering: calibration, simulation, and turning video of a motion into a deployable skill.

The common thread here is that AI is starting to do the work that used to take a team of robotics engineers: labeling data, building simulations, tuning parameters and analyzing failures. That’s vastly increasing the number of things team can train robots on.

Standard Bots is applying that idea at the far end of the chain, on the factory floor: post-training in industry jargon. If the industry can make post-training cheap and fast enough, the economics of automation change a lot, especially for companies that have never been able to afford it before.

Standard Bots’ head of applied AI, Leif Jentoft, puts a price on the benefit: “You used to have to hire an integrator; it costs a million dollars. You spend months dialing it in. And when something breaks in the middle of the night, you can’t fix it yourself.”

The new robot training skills

The first of the three new skills, Agent, is the vibe coding part.

It’s a chat interface trained on Standard Bots’ manuals and best practices. Ask it to build a machine tending routine and it asks about your setup, then builds the program with you. Beard says it can pull from “thousands of examples of canonical … programs to figure out … how should your program be structured, and what skills do you need to pull?”

Locate helps robots find objects in space that they’ll need to work on.

Most industrial robots, as Beard describes them, are glorified playback machines. “It’s like crane game at Denny’s,” he says. “You move them around the workspace, and you save points, and you save point every inch … and they replay motions, and that’s all they do for the most part is just replay motions.”

That’s fine when every part arrives in exactly the same way at the same spot every time, but it fails when anything changes. So manufacturers spend a lot of money on fixtures and part presentation to make sure nothing changes.

With Locate, any front-line work can tap an object on the robot’s camera feed, or just describes it in text, and the robot finds it, tracks it and moves around it. Standard Bots says that cuts programming time by 90%. If that holds up in the real world on shop floors, that’s huge. And it especially matters most for high-mix, high variety work: small shops running many different parts a day on the same machine, where reprogramming and re-fixturing for every changeover kills the economics of automation.

“We’ve taken AI models trained on billions of images, and we’ve customized these models so that a user can just click,” Beard says. “And something that used to be weeks, that used to cost tens of thousands of dollars, takes seconds.”

The third skill, Inspect, is quality control. Built with computer vision company Roboflow, it lets customers train custom vision models on their own parts, using photos the robot itself takes as it moves around a part, to flag defects like scratches. Customers can also bring their own models rather than being locked into Standard Bots’ stack.

“Basically what it comes down to is if a human can tell the difference, we can train an AI model to tell the difference,” Beard says.

The big target here is the long tail of American manufacturing.

Nearly 200,000 of America’s 240,000 manufacturers have fewer than 20 employees, according to the National Association of Manufacturers. Deloitte and the Manufacturing Institute estimate the industry could need 3.8 million new workers by 2033, and that 1.9 million of those jobs could go unfilled. And yet only 6% of U.S. manufacturers use robots today, according to Standard Bots. Meanwhile, China installed more industrial robots in 2024 than the rest of the world combined, as I noted when covering the company’s Series C fundraise.

Not shockingly, small shops don’t have robotics engineers on staff. And they probably can’t afford to pay an integrator every time a product line changes, let alone the first time they install an automation solution.

That’s the whole point, Beard says. Customers’ parts change and their processes change, and “if you have to call somebody to do it, it’s out of your control.”

“This is all about making sure customer is confident they’re going to see the ROI,” he adds. “That’s how you sell product.”

Standard Bots claims its robots deploy and pay back twice as fast as the industry standard. Customers include Sunoco, Lockheed Martin, NASA and the U.S. Army, but also, Beard says, small businesses in almost every state, including a baker. The company designs its own hardware, software and AI in Glen Cove, and has said it plans to manufacture everything in America.

Adrian Nistler, CEO of Path Machining Automation, a contract machining company in the Minneapolis-St. Paul area, is one of those customers.

“This is a big thing for American manufacturing because we can actually compete,” he said in a statement. “Quality’s improved, on-time delivery is improved, and price competitiveness is improved.”

Beard’s ambitions go well past a chatbot on an iPad. Standard Bots is building its own foundation models, and he frames all of this as a race to the physical world’s version of the ChatGPT moment, which he argues won’t arrive just because the AI is capable enough.

“It’s also when … you’re on the steep part of the adoption curve, and this is taking off,” he says. “The only way to make that happen is it has to be accessible to a normal person to be able to unbox it and train an AI model.”

Or, as he put it more simply:

“The robot shouldn’t be harder to use than your iPhone.”