AI Designs a Chip and Writes its Code, Shrinking Years of Work to Days

SAN FRANCISCO, CA – July 14, 2026 – In the world of semiconductor development, time is measured in seasons, if not years. The painstaking process of designing, verifying, and fabricating a new piece of silicon is a monumental effort. Once that hardware finally exists, a second, equally arduous journey begins: writing the low-level firmware that breathes life into the chip. Today, a partnership between two AI pioneers, Verkor and Embedder, promises to collapse that multi-year timeline into a matter of days.

Verkor, the company behind the autonomous chip design system Conductor, and Embedder, an AI agent for firmware development, have joined forces to create a fully integrated, AI-driven pipeline from concept to functioning system. Their first joint achievement is nothing short of revolutionary: Verkor’s Conductor platform designed a complete RISC-V CPU core, dubbed “VerCore,” from a 219-word text prompt in roughly 12 hours. Almost immediately, Embedder’s AI agent began writing and verifying the firmware to make that brand-new silicon work, a task it performs on real hardware.

This collaboration directly targets the industry's most notorious bottleneck. Historically, software development for new silicon could only begin after the chip design was finalized and physically available, creating a long and costly delay. “A system is only proven when software runs successfully on its hardware,” said Suresh Krishna, CEO of Verkor. “AI-based software and hardware co-development speeds up building and validating a fully working system, taking it from years to days.”

A New Paradigm in Silicon Development

The traditional chasm between hardware and software teams has defined the pace of innovation for decades. Chip design cycles are long, and the subsequent firmware bring-up—the process of writing the first code to run on a new chip—is notoriously difficult. Firmware engineers must grapple with dense reference manuals and new, unproven hardware, often spending weeks on what is colloquially known as the “hardest week of their career.”

Verkor aims to eliminate the first half of that delay. Its Conductor platform has demonstrated the ability to produce a verified, tape-out-ready layout from a high-level description, a feat that compresses months of human-led design work. The resulting VerCore CPU is not a toy; simulations show it running at 1.48 GHz with a CoreMark score of 3,261, and a functional FPGA implementation is already operational.

This radical acceleration in hardware design, however, only intensifies the pressure on the firmware side. “Chip design used to take months. Verkor just cut it down to a day,” noted Ethan Gibbs, CEO of Embedder. “That makes firmware the next bottleneck, and firmware is the problem we built Embedder to solve.”

Embedder’s AI agent tackles this challenge by building its understanding from the ground up. Instead of relying on pre-existing code libraries or vendor ecosystems, it ingests the chip’s reference documentation and schematics to generate context-aware firmware. Crucially, it then closes the loop by driving real-world test equipment—like logic analyzers and oscilloscopes—to prove the code works on the physical silicon. For the brand-new VerCore, this meant starting from a blank slate. As Gibbs puts it, “For Embedder it's a normal Tuesday.”

The Agentic Workforce: How AI is Redefining Engineering

The collaboration offers a compelling glimpse into the future of technical work, where autonomous AI agents act as highly specialized colleagues. This isn't about a chatbot suggesting code snippets; it's about end-to-end task execution. Verkor's AI navigates the labyrinthine process of physical chip design, while Embedder's agent performs a complete hardware-in-the-loop (HIL) validation cycle: writing code, compiling it, flashing it to the chip, observing its runtime behavior, and even correcting its own errors.

A key distinction from general-purpose AI is what Embedder calls “Documentation Intelligence.” The agent’s work is rigorously grounded in the technical specifications of the hardware. By indexing and cross-referencing datasheets, reference manuals, and even errata, the system avoids the “hallucinations” or plausible-sounding errors that plague less specialized models. This hardware-aware approach allows it to understand register maps, timing diagrams, and memory constraints with high fidelity, a capability essential for the unforgiving world of embedded systems.

This shift doesn’t necessarily signal the obsolescence of human engineers but rather a profound change in their role. Industry analysts suggest that engineers will move from manual implementation to high-level oversight. Their focus will migrate from writing boilerplate drivers to defining system architecture, validating AI-generated outputs, and solving the complex, non-deterministic problems that still require human intuition. The job becomes less about the “how” of coding and more about the “what” and “why” of the system’s goals, turning engineers into architects and validators of AI-driven work.

Supercharging the Open-Source RISC-V Ecosystem

The choice of a RISC-V core for this demonstration is particularly significant. RISC-V is an open-source instruction set architecture (ISA) that has been steadily gaining traction as a royalty-free, customizable alternative to proprietary architectures from firms like Arm and Intel. Its open nature has fostered a vibrant community dedicated to creating custom processors for everything from tiny IoT devices to massive data center accelerators.

However, a major hurdle for widespread RISC-V adoption has been the fragmentation and relative immaturity of its software and firmware ecosystem compared to established players. The Verkor-Embedder partnership directly addresses this gap. By enabling the creation of a custom RISC-V core and its essential firmware in a matter of days, the collaboration provides a powerful accelerator for the entire ecosystem.

As Verkor’s CEO Suresh Krishna highlighted, this approach “gives AI-designed silicon a software story on day one.” For developers in the RISC-V community, this could unlock an unprecedented pace of innovation. Instead of waiting months or years for a silicon vendor to provide a development kit and basic software, teams can now theoretically design a custom core for a specific application and have functional firmware ready almost immediately. This dramatically lowers the barrier to entry for hardware startups and researchers looking to experiment with novel computer architectures, potentially democratizing chip design on a scale previously unimaginable.

The ability to rapidly iterate on both hardware and software in tandem could lead to a Cambrian explosion of specialized RISC-V designs. This tight co-development loop is critical for optimizing performance in demanding fields like automotive, defense, and networking, where both Verkor and Embedder are already finding customers. The partnership doesn't just represent a new set of tools; it signals a fundamental re-architecting of the value chain that brings our digital world to life.