Figma Closes the Gap Between Designers and Developers — While OpenAI Builds Its Own Chip
Today’s AI news has a theme running through it: the tools we use to build and run AI are changing fast, and the changes are finally reaching everyday workflows. Figma just made a major push to blur the line between design and code. Meanwhile, OpenAI is quietly becoming a hardware company. Both moves point to the same underlying pressure: AI is getting woven deeper into the work, not just bolted on top.
Figma Lets Designers Touch the Code — Without Leaving the Canvas
Figma released a wave of new features at its Config 2026 conference this week, and the headline addition is something called code layers. Think of it this way: until now, a designer would create a button in Figma, and a developer would have to translate that into actual code somewhere else. Code layers collapse that gap. Designers can now see and edit the underlying code directly on the same canvas where they’re doing their visual work. TechCrunch and The Verge both confirmed the details of the update.
The update also adds AI-powered tools for motion graphics and shaders — shaders being the small programs that control how light, color, and texture appear on screen. Figma is using AI to generate these effects automatically, which previously required serious technical skill. There’s also a new AI plugin builder, which lets users create custom Figma add-ons by describing what they want in plain language instead of writing code from scratch.
For anyone who works in a team that includes both designers and developers, this matters right now. Fewer handoff meetings. Fewer miscommunications over what a design is supposed to do versus what it actually does when built. The motion tools also mean that animated UI elements — things like loading spinners, transitions, and hover effects — can be created without exporting anything to a separate app.
Why this matters: Figma is quietly becoming the operating system for product teams. Every feature it adds pulls more of the workflow into one place.
“Figma adds code layers, animations, motion graphics, and AI-powered plugin builder”
OpenAI and Broadcom Built a Chip Together — Here’s What That Actually Means
Inference — the process of running an AI model to generate a response — is one of the most expensive operations in tech right now. Every time you ask ChatGPT a question, a massive amount of computing power fires up to produce your answer. OpenAI and Broadcom have designed a custom chip called Jalapeño specifically to make that process faster and cheaper at scale. Ars Technica broke down the technical details of the announcement.
General-purpose chips like Nvidia’s GPUs are incredibly capable, but they’re designed to handle many different kinds of computing tasks. A purpose-built chip like Jalapeño can ignore everything except LLM inference — large language model inference, meaning the specific math involved in generating text from AI systems. That specialization typically translates to better performance per dollar. Apple took a similar approach with its M-series chips, which outperform many general processors on tasks they’re tuned for.
Most people will never hold a Jalapeño chip, but they’ll feel its effects. Faster responses. Lower costs for AI services. Potentially cheaper API access for the businesses that build the apps you use every day. OpenAI has spent years depending on Nvidia for its hardware needs. Designing its own chip signals that the company wants more control over that stack — and more predictable costs.
Why this matters: When AI companies start building their own chips, it means they’re playing a longer game. This is infrastructure, not a product feature.
“OpenAI and Broadcom unveil Jalapeño, a custom AI chip built for LLM inference”
Also Happening in AI
A few other stories worth keeping an eye on today. Researcher departures at Google are accelerating — according to TechCrunch, prominent AI scientists Jonas Adler and Alexander Pritzel are among the latest to leave for rivals, continuing a pattern that has been quietly reshaping which labs have the strongest technical talent. On the hardware side, Cerebras — a company that makes chips designed to run AI faster — saw its stock drop sharply after its first earnings report as a public company, with the CEO saying investors misread the company’s margin forecast, per TechCrunch. Meanwhile, Wired is reporting that Google now holds on to images and media you upload during searches to train its AI systems, and there’s a way to opt out if you’d rather not contribute. Finally, Meta is testing a standalone AI companion app aimed at content creators, replacing parts of Creator Studio, according to TechCrunch.
What to Watch
The Jalapeño chip and the Cerebras earnings stumble are both symptoms of the same tension: running AI at scale is brutally expensive, and nobody has fully solved it yet. Watch whether other major AI labs — Anthropic, Google DeepMind — announce their own custom silicon in the next few months. The researcher exodus from Google is worth tracking closely too. Talent tends to move before breakthroughs do.