Claude AI + KiCad PCB Design Video: From Idea to PCB

This video explores AI-assisted PCB design using Claude AI and KiCad, combining a conversational AI tool with an electronic design automation (EDA) platform to build a printed circuit board. The workflow illustrates how AI can help translate design requirements into structured circuit information and support tasks involved in schematic development, component selection, PCB layout, and design review.

KiCad provides the engineering environment for managing schematics, footprints, electrical connections, copper routing, and manufacturing outputs. Claude AI can assist with design reasoning, documentation, and selected workflow tasks, depending on the available tools and project setup. However, AI-generated design data still requires verification against component datasheets, electrical requirements, and KiCad's electrical and design rule checks.

For engineers developing IoT devices, industrial control systems, and other electronic products, the key question is not simply whether AI can produce a board layout, but whether that layout is electrically sound and manufacturable. Material selection, layer stack-up, trace spacing, drilling constraints, and assembly requirements all affect the final result. Before production, designers can review PCB prototype services and obtain an online PCB quote to evaluate fabrication requirements and costs. For a board that also needs component sourcing and soldering, a PCB assembly quote helps connect design decisions with assembly requirements.

Key Highlights

• AI-assisted PCB development: Claude AI can help organize circuit requirements and support parts of the design workflow, while KiCad provides the schematic, footprint, layout, and fabrication-file environment.
• PCB layout still requires engineering checks: Component footprints, electrical connectivity, trace clearances, return paths, and design rule violations must be reviewed before a board is released for manufacturing.
• Manufacturing readiness goes beyond a completed layout: Fabrication outputs, material selection, copper requirements, drill specifications, and assembly constraints must align with the board's intended application.

How Claude AI and KiCad Work Together in PCB Design

AI-assisted PCB design combines natural-language problem solving with the structured engineering functions of an EDA tool. Claude AI can help interpret requirements, suggest circuit approaches, explain design errors, and assist with scripts or project data when the necessary interfaces and context are available. KiCad remains the environment in which the actual schematic and board design are represented and checked.

The degree of automation depends on the workflow. Asking an AI model for design advice is different from allowing an agent to modify KiCad project files or interact with the PCB editor. File-based workflows may involve generating or editing project data, while more integrated workflows can use scripts, command-line utilities, or APIs where supported. Neither approach guarantees that every generated change is valid.

A practical workflow starts with a clearly defined electrical specification. Engineers should identify the power supply, controller, interfaces, connectors, operating conditions, and critical component requirements before asking AI to help generate or modify a design. Without these constraints, an AI-generated circuit may contain incompatible components, incorrect pin assignments, unsuitable footprints, or missing supporting circuitry.

From AI Suggestions to a Verifiable Circuit

Circuit requirements should be translated into explicit connections and component values rather than accepted as unstructured recommendations. Verify pin functions, supply-voltage limits, decoupling capacitors, pull-up and pull-down resistors, protection components, and connector pinouts against the relevant datasheets.

For example, an AI-generated microcontroller circuit may appear logically correct while overlooking a boot-configuration pin, a required external pull-up, or the current limits of a power rail. These omissions can prevent startup or cause intermittent operation. Reviewing the schematic with the component datasheets open is more reliable than relying on a conversational explanation alone.

KiCad's Electrical Rules Checker (ERC) can identify many schematic-level issues, including certain unconnected pins and electrical conflicts. However, a clean ERC report does not prove that the circuit performs as intended. It cannot independently establish that the selected regulator supports the load, that the clock configuration is correct, or that every component meets the system requirements.

From Circuit Concepts to a Routed PCB Layout

Once the schematic has been reviewed, the next stage is converting logical connectivity into a physical board. This involves assigning footprints, importing component relationships, arranging parts, defining the board outline, and routing the connections. AI may help with selected operations, but layout quality depends on electrical constraints and physical geometry as well as connectivity.

Footprint assignment is a particularly important checkpoint. A schematic symbol can represent the correct electrical component while its assigned PCB footprint has the wrong package dimensions, pin numbering, pitch, or pad arrangement. Before routing, compare the footprint against the manufacturer's recommended land pattern and verify that connector orientation and component polarity match the intended assembly.

Component placement should reflect the function of the circuit. Decoupling capacitors should be placed close to the power pins they support, switching regulators should follow their recommended layout guidelines, and high-speed signals should have suitable reference planes and return paths. An AI-generated layout that passes basic clearance checks can still have excessive noise, poor thermal performance, or difficult assembly access.

Figure 2 | KiCad PCB layout showing component placement, copper traces, ground plane, and design rule checks

Routing is not simply the process of connecting every net. Trace width, spacing, via placement, copper geometry, and layer transitions influence electrical performance and manufacturability. High-speed designs may require controlled impedance, carefully managed return paths, and constraints on differential-pair geometry. Power circuits may need wider copper, additional thermal paths, or dedicated copper areas based on current and temperature-rise requirements.

Routing and Design Rule Checks

KiCad's Design Rules Checker (DRC) evaluates board geometry against configured constraints, including many clearance, connectivity, and manufacturing-rule violations. The configured rules matter: a board can pass checks against inappropriate or incomplete constraints and still be unsuitable for the selected manufacturer.

A successful automated check is a necessary checkpoint, not a complete sign-off. Engineers should review the remaining warnings, inspect critical routing visually, and confirm that the manufacturing constraints reflect the actual board supplier's capabilities.

Can an AI-Designed PCB Be Manufactured Reliably?

A board that looks complete in KiCad is not automatically ready for fabrication. The manufacturing package must contain consistent, correctly configured outputs, including copper Gerber files or an appropriate alternative, drill files, board outline data, and fabrication notes. If assembly is required, the package may also need a bill of materials (BOM), component placement files, and assembly drawings.

For a standard rigid PCB, material selection and the layer stack-up should reflect the operating environment and electrical requirements. FR4 is common for general-purpose boards, while high-speed or RF designs may require materials with more tightly controlled dielectric properties. Flexible and rigid-flex designs introduce additional constraints around bend radius, copper construction, and mechanical stress. These decisions should be made from application requirements rather than inferred by an AI model without adequate specifications.

Original Source: https://www.youtube.com/watch?v=-PVyx749kyU

Watch the original video on YouTube.

Recently, all the frontier AI models being released are showing separate benchmarks for building PCB boards. Actually, I have a huge desire to build PCB boards using these AI models. So, you'll always see that on our channel, we mostly cover videos on building software or agents using Claude Code, or building things from scratch. So, we are going to start a new series. That is it. So, in this episode, what we are going to see is that we will build a PCB board completely using AI. The tool we are going to use for this, if you ask, is Claude Code again. Because if you ask who is on top among the current hardware engineering tools, it is Claude Code. So, let's see in this video, end-to-end, how to build a PCB board using it. And in each episode like this, we are going to do all our IoT projects completely using AI. And if you don't want to miss such videos, don't forget to subscribe to my channel. And don't forget to put your comments in the comment section. Because the comments you leave are what motivate me to post more and more videos. Don't forget to like this video. Now, let's go into the video.

Well, after a long time, I was working on a project using the ESP32 module to try something new in IoT. Actually, I had done a project back in college. It was about predicting weather in agriculture and activating a machine learning model to open the tap automatically. So, you might have heard of this project already. I did this when I was studying in college. Now that AI models have advanced to the point of building PCB boards, I got an idea: why don't we build a PCB board? But I honestly tell you, I don't know what's inside a PCB board or what the components are. But okay, I decided to try building it using AI. The module we have chosen for this, if you ask, is this one, which we call the ESP32. If you ask what this ESP32 module is for, you can connect many sensors to it and collect your data. You can send it to the cloud and use a machine learning or deep learning model there to build an IoT solution. Now, this module itself comes with Bluetooth, Wi-Fi, and everything like that. Okay, so with this module, I thought, "Okay, this is a PCB board itself." I had a thought: why don't we build this board? Right, let's build it, so I have now finished all the necessary setup for that. So, I will show you live how to build it.

So, the tools we will need for this — if you search for "KiCad," look here, a software called KiCad will appear. What this software actually is, is a tool for designing PCB boards. Not only that, it has a lot of features available. As far as I've seen, we can do this for free using KiCad. Okay? So, if I just click on download here, a file will start downloading for me. And I have already installed KiCad as well. So, I will open it and show you. And this is how the KiCad view looks. Actually, I didn't know about this software before. I went on YouTube and watched a tutorial. So, I just opened it to see what's in it and how the ESP32 module works. Okay. So, in this, we can build a PCB board and view it in 3D. That is, we can completely see how it will look once rendered.

So, what we are going to do now is, to build this PCB board, we need the desktop version of Claude. If you have Claude Code, you can do it too. But I chose the desktop version. I will tell you why. Because if you go to its settings, you will see an option called "System" down here. If you click it and scroll down, you will see an option called "Computer Use." This is a really, really great option. If I need to perform any kind of task, I use this. So, if you ask what it basically does, you just need to do this: Turn on "Enable Computer Use." Once you turn it on, when you give Claude a task, it takes over your system completely and handles the work itself. Right now, I don't know how to work with KiCad at all. I haven't used this software either. I just watched an overview on YouTube. So now, this Claude will enable my computer use, open KiCad, and build the PCB board we need completely. All the benchmarks needed to build a PCB board are available in the OPUS 5.5 model. If you ask who is on top, it is GPT-6 Astra. But we don't have the money to buy that. So, the one we have is just Claude, that's why we are using Claude. So, OPUS 5.5 has a good benchmark in building PCB boards. Okay, now I will tell you how I build it.

So, just enable the option called "Enable Computer Use" here. Once done, I am just going to give a prompt. "I have installed KiCad, build a PCB board for the ESP32 DevKit." The one I had in my hand is called the DevKit v1. Okay? So, I am telling it to build that. I am saying, "Use my computer and use my installed KiCad." So, I am telling it to use my computer access and build the PCB in KiCad. Otherwise, you know what it will do? It will write a Python script and try to prepare it that way. Doing it that way won't be good. So, what I am doing is, I am giving it my computer access. And I am telling it to use that to operate KiCad. So, what it's doing now is looking at the task here, loading each skill one by one, and performing it through code. Meaning, it's using Python scripts to generate it. Instead of saying that, I'm telling it now, "Use my laptop." Okay, now you can see that our Claude has taken screen access. I have granted screen access. Now, if you look on the side, it will keep running. While this is running, it will open KiCad on our laptop screen and start building the PCB board. So, we can visually see how it is building it. We can see it doing it one by one. Okay, I think EEE is one of the toughest domains in engineering. Because I've seen the coding involved in it. They call it "embedded" coding; there hasn't been a day where my head didn't spin looking at that. So, the fact that AI is doing it today is a very incredible development in technology. Alright, let's see how far it builds this. So, look now, it has opened our KiCad, created the project, and saved it. After saving, it has now opened the project here. It created a project called "ESP32 Carrier KiCad Pro," and you can see all the files inside. In that, there is the "ESP32 Carrier KiCad PCB," which is what I asked it to build. PCB board document. We can also view that in 3D. And what you see below is what we call its circuit diagram. So, it will complete all of this and show us the output of how it looks. Not only that, beyond just building, it also performs tests. Only after doing a few tests will we know whether it works or not. So, look now, it has generated everything needed for our PCB board. If you look on the side, it's opening some option to check for bugs in the ESP32 module. Look, it's running tests, checking for any violations. I don't even know what all these are. It's doing something, opening a "Symbol Fields Table" and inserting many symbols inside. So, now you can see that our PCB board is ready. This is the overview of that board. Okay, that is what is running on our screen right now. This dev kit has also delivered the version we asked for. Alright, once this is finished, it will compile completely. We can ask it ourselves whether the compiled version is correct or not. My goodness, look at what it's doing; it's fixing bugs and violations one by one, and it's running the tests itself. After running them, it's testing whether the circuits are working properly. Now, look at the schematic part there; it showed nearly 70 errors. It did something, and look, all 70 errors are totally cleared. In the schematic area, you can see that it says zero now. I feel like I've got someone who has done a PhD in EEE sitting here with us. Watching all this work. In that violation section, you can see it showed almost 15 violations. But now, there is only one violation left. It's searching for what that one violation is and fixing them one by one itself. That one violation just won't go away. It has been looking at it for a long time, wondering what it is. Let's just see if it fixes it or not. The screen that's open now is what we call the 3D view. In this 3D view, all the components are rendered and placed. Actually, it hasn't finished yet. It is checking. It kept saying one violation was coming up repeatedly, but then it finally finished it. Now, you can see that after finishing, our PCB board is ready. Now it will just give us a report about what all has come up. Okay, let's see what it has said in this. "I built the ESP32 DevKit V1 board in your KiCad, it's open in 3D viewer on your laptop screen, it found zero errors, zero unconnected nets, and the board matches the schematic." It has matched the schematic correctly, zero errors, zero unconnected networks, it says. One cosmetic warning, which is that there's something called J3 with the terminal outline. There's something there; I don't know what it is. "I tried to move but change didn't take." So, what it says is, "To keep it accurate, I wrote the files first, then opened and finished on your screen," it says. So, it has made the board completely ready. Zero errors, okay? All the nets are connected, nothing is missed, everything is perfect, it has given that report too. Apart from that, it has listed what is on this board. Now, I am asking what I should do next. Before that, it says, "Before ordering check one thing, some KiCad clones have a pin row 2.2 mm apart from 2.4 mm, ensure you are first, when you are ready to order, use file fabrication output PCB editor to make the files PCB makers need." Okay, in this, it says first to check the output, meaning the length of that PCB board. Obviously, I don't know. If any EEE students are watching, don't forget to tell me in the comment section. So that's what it says, basically to take the output file out and give it to those who build PCBs, and they will make it for you. I am asking a question: "How are you telling this board will work?" So, how do you say this board will work? On what basis are you saying this? Let's ask and see. Okay. "Honestly, I can't promise it will work." It says it won't promise it will work, but it says one thing: "I checked the design is correct on paper." The design is correct on paper. There is a check called DRC (Design Rule Check). In that, you can see there are zero errors, so no track is close. Therefore, it will work completely. The unconnected items are also zero, which indicates that everything is indeed connected. You can see there is zero difference in the schematic as well. The schematic is very, very important. I have read about this. If the schematic difference is zero, our board will absolutely start working. So, in this, you have filled in all these pin orders, right? It says the confusion is whether it should go on the left or the right. Suppose we need to feed a little more data into this. That is, the research paper for the ESP32. Using that, we can do it even better. I haven't enabled all of that yet. There is an option here called Claude Chrome. If you enable that, suppose ... It will go into Chrome, fetch all the data you need, and do it end-to-end for you on the computer. So, what's the news, you ask? Until now, we only saw it doing coding tasks. From now on, it has started building PCB board tasks as well. We used Opus 3.5 and GPT, but compared to the others — like GPT-4 o — this Opus 3.5 model has its own separate benchmark for building PCBs. So, if we bring that model here, it will give much better reasoning confidence for these kinds of things. So don't forget to try this out yourself. This KiCad software is free and available as open source. Download it and try building a PCB board using Claude. And in this episode, we are going to look at an IoT project. Not just IoT, but AIoT — meaning we are going to develop it using AI. We are going to integrate AI into it as well. If you don't want to miss any of these episode videos, don't forget to subscribe. Many of you are watching without subscribing. And don't forget to put your comments in the comment section. I will reply to them personally. And don't forget to like this video, and if you swipe right, there will be a 'Hype' button. Click that as well. So, in this video, you have overall clarity on how AI builds PCB boards now. You should have also gotten complete clarity on what tool is used for that. And in the upcoming series, we will practically start building a complete application using AI and IoT devices.