Embedder plugs its firmware agent into SEGGER's hardware debugging stack

Embedder's YC-backed agent can flash, inspect and debug firmware through SEGGER probes, moving AI coding onto the physical bench.

By · Published

Primary source: PR Newswire

Why it matters

Direct control of widely used debug probes gives firmware agents the physical feedback needed to test their own code, though independent reliability data remains the key test.

A close-up view of an electronics workbench, showing a hardware debugging probe connected to a circuit board, with digital patterns flowing from a screen into the chip.

Embedder founders Ethan Gibbs and Bob Wei are giving their firmware agent direct control of SEGGER J-Link and J-Trace probes, a September 10 partnership that moves the agent from producing code to testing its work on physical devices.

Under the integration described in the companies' September 10 announcement, Embedder can flash a build onto a target device, set breakpoints, inspect registers and memory, monitor live output through SEGGER's Real Time Transfer system and continue debugging after a failure. The intended loop ends when the firmware runs correctly on the device.

That last step defines the bet Gibbs and Wei have made since founding Embedder in 2025. A general coding agent can produce plausible C or C++ that compiles. Firmware has to operate against a particular chip, board layout, sensor, power budget and timing constraint. The physical device decides whether the code works.

Wei arrived at that problem after working as a robotics researcher and vehicle firmware engineer at Tesla. He left the University of Michigan's computer engineering program before graduating and later became a Z Fellow and Y Combinator alumnus. Gibbs, a Michigan computer science graduate, had worked as a software engineer at several YC-backed startups. The two met at Michigan and built Embedder around the view that an AI firmware engineer needs access to the bench equipment used by a human engineer.

"The biggest bottleneck in embedded development is shifting from writing firmware to verifying it meets functional and performance requirements on real hardware," Wei said in the announcement.

Giving the agent something to touch

SEGGER's probes supply a common hardware interface for that verification loop. J-Link is used to program and debug microcontrollers, while J-Trace provides trace capabilities for supported targets. As of September 10, 2026, SEGGER's J-Link supported-device directory lists 15,246 devices, while the partnership announcement cites the broader figure of more than 15,000.

That reach matters because embedded development remains fragmented across chip vendors, processor families, software development kits and proprietary board designs.

Embedder says its own catalog covers more than 500 platforms from 13 manufacturers and more than 5,500 peripherals. Its current product material also lists support for C, C++ and Rust projects spanning bare-metal systems, real-time operating systems and embedded Linux. Those are Embedder's figures. Y Combinator's listing still describes support for more than 500 microcontrollers and 3,000 peripherals, using an earlier count and a narrower label than Embedder's current "platforms" figure.

Embedder's workflow reaches beyond the debugger. Embedder says its agent reads datasheets, schematics and hardware errata, writes part-specific drivers, observes serial output and can operate logic analyzers, oscilloscopes and power profilers.

SEGGER product manager Erik Loehr said direct access to target hardware allows agents to investigate unexpected behavior, identify bugs and verify fixes on the device.

The benchmark still has to come from the hardware

The announcement establishes the connection between Embedder and SEGGER's probes. It does not establish how often Embedder can complete a bring-up or debugging task without an engineer taking control. The partners did not publish success rates, time comparisons, named deployments or a general availability date for the integration. Commercial terms were also omitted.

Those measurements will decide whether autonomous firmware development becomes a production workflow or remains a supervised engineering tool. A failed web application build can be rolled back. A firmware defect can drain a battery, disable a sensor, corrupt stored data or require access to a device already deployed in the field. Engineering teams will need reproducible test evidence for flashing, file changes and instrument activity.

Embedder says its platform provides cited, hardware-backed evidence for its conclusions. It also offers cloud, on-premises and bring-your-own-cloud deployments, an important requirement for customers working with unreleased devices, proprietary schematics or defense systems.

Gibbs and Wei are pursuing a category that has already attracted nearby entrants. BootLoop, another San Francisco company from Y Combinator's Summer 2025 batch, sells an AI firmware agent alongside hardware-in-the-loop testing. H2LooP is developing hardware-aware AI for safety-critical embedded systems, including automotive and aerospace software.

Embedder lists Y Combinator, CRV, Emerson Collective and BoxGroup among its backers, along with Liquid 2, Palm Drive and other early-stage investors. Embedder has not attached a confirmed funding total or valuation to that list. YC currently lists six people on the Embedder team.

The partnership gives that small team a broad route into hardware already used by embedded engineers. The harder task is proving that an agent can move safely through the full loop: understand the board, write the firmware, flash it, diagnose what happened and produce evidence that an engineer can trust.

Embedder and SEGGER plan to show the integration in a joint webinar on September 29 at 8:00 a.m. Pacific. For Gibbs and Wei, the demo will put their founding premise on the bench, where the code either works or it does not.

Reader comments

Conversation for this story loads after sign-in.