KiCad + Codex Isn’t a Team Workflow

What today’s hardware teams run on

Hardware teams look different than they did two years ago. Same EEs. Same MEs. Plus AI engineers. The humans hand over a requirement and spend their time reviewing, deciding and shipping.

In 2019, people thought we were lunatics for pitching AI for hardware. We kind of were. Back then, models couldn’t pick a voltage regulator that was in stock, let alone route a PCB. Now they can.

We’ve seen this movie before. When Cursor showed up, plenty of software engineers shrugged. They tried it once, judged it by what it could do that day, and went back to their editor. Now it’s hard to find a software team that doesn’t work with an AI coding agent, and the job itself changed: engineers stopped writing every line of code and moved to just reviewing an agent’s work.

We’re watching the same thing start in hardware. The early teams have already switched. Everyone else is next.

“Why not just use KiCad and Codex?”

I think there’s a more fundamental argument here.

How many engineers do you know who use VS Code through an MCP server to do AI coding? Zero!

AI coding didn’t take off because we connected LLMs to existing IDEs. It took off because we built entirely new, AI-native workflows around coding agents.

Cursor was an intermediate step: adding AI to the traditional IDE. But tools like Claude Code, Codex, and Devin represent a much bigger shift, where the agent becomes the primary interface and the tools operate behind the scenes.

I believe hardware engineering is going through the same transition.

KiCad + Codex is the equivalent of bolting an AI agent onto VS Code. Flux is about reimagining the entire engineering workflow around AI, with CAD, EDA, simulation, manufacturing, and collaboration deeply integrated into the agent’s environment.

“The real competition isn’t about which CAD tool an agent can control. It’s about who builds the best environment for AI agents to engineer hardware.” – Matthias Wagner, Co-founder

That means one shared project the whole team can reproduce, live collaboration instead of emailed files, and an agent that knows what’s in stock and what your fab can build.

So it’s not really a choice between Flux and your favorite AI coding agent. Bring Claude, Codex or another frontier model into Flux, and it gets our hardware harness, tools and shared context. Or connect your existing agent to Flux over MCP. Either way, you get the same integrated hardware engineering environment.

Bring the work you already have

A better environment only matters if your work can get into it. Flux now imports complete KiCad, Altium, OrCAD, Eagle, EasyEDA, PADS and CADSTAR projects. You read that right: parts, nets, schematics and layouts! Most people skip the import menu and just tell the agent “bring this in.”

This week, a customer with an audio product line imported an Altium project, had Flux replace the obsolete parts, and exported it for manufacturing.

Mechanical works the same way. Bring in a STEP, IGES or BREP model and say what you want: “make the flange 40 mm deeper and add five vent slots.” No more “why can’t I just make it bigger?”

Your work never gets locked in, either. Export Gerber, ODB++ or STEP anytime.

Flux also plugs into the rest of your stack. Connect any MCP server or web API: push firmware to GitHub, pull field crash reports from Memfault, link your PLM. Admins add API keys once, and the agent never sees them.

One designs things. The other gets working hardware built.

That’s the difference between an AI agent and an AI engineer like Flux. Hand it a requirement and it picks parts you can buy today, designs the PCB, models the enclosure, writes the firmware, checks its work in simulation, and reports back. Big jobs run in parallel threads. Anything that moves an interface waits for your approval.

Why this is the moment

Hardware was always a team effort. The tools just never acted like it. Teach Flux your guidelines once, and every project follows them. Use OpenAI, Anthropic or Gemini models, with no lock-in to one AI lab.

We’ve been working on AI for hardware for seven years. Seems like the world is finally catching up. And the gap between teams that put AI engineers to work and teams that don’t is about to get very wide.

We’re on day one of the AI age. The best hardware teams of the next decade won’t be the biggest ones. They’ll be the ones that put AI engineers to work first.

See Flux for Teams →