Why We Invested in Hermetiq

Gi21 Capital has invested in Hermetiq's $2.2M seed round. Hermetiq builds the intelligence layer between AI coding agents and the build systems that have to compile, test and ship what those agents write. Here is why we backed the team.

The bottleneck nobody budgets for

At the largest engineering organisations in the world, the build process runs on Bazel, the open source build system originally developed at Google. Bazel is powerful and it is complex. When a build fails, the diagnosis is rarely obvious. It requires an engineer who understands the toolchain deeply, and those engineers are scarce everywhere. While one specialist works through the logs, the rest of the team waits.

That waiting is invisible in most budgets. It does not appear as a line item, it appears as slower releases, longer review cycles and senior engineers spending their week on infrastructure archaeology instead of product.

Why this problem is about to get much larger

Agentic coding tools have changed the volume equation. Teams now generate far more code than humans can review, and every commit still has to pass through continuous integration. More code means more builds, and more builds mean more failures. The cost of AI-generated code does not land on the model, it lands on the build system.

This is the part of the AI stack we find most interesting as investors. The models attract the attention and the capital. The infrastructure underneath them absorbs the consequences, and it is chronically underbuilt.

Evidence, not narrative

Early-stage investing usually means betting on a story. In this case there was already a track record to examine.

Hermetiq runs in production at one of the top four frontier AI labs, handling telemetry from more than two million builds per day. Etsy is running a pilot. Bazel lists Hermetiq as an official product partner and as a remote execution service, which is a meaningful signal in a community that is notoriously hard to win over. The company holds SOC 2 Type I and offers both cloud and self-hosted deployment, which matters for exactly the kind of customer that has this problem at scale.

For a company at seed stage, that combination of production scale and enterprise readiness is unusual.

The team

Jake Newfield and Tim Potter come out of the build infrastructure world rather than adjacent to it. They were working on remote build execution before AI coding agents made the problem urgent, which means they were early to the market by conviction rather than by timing. In a category this technical, that distinction tends to decide who wins.

We are pleased to be investing alongside a strong group of operators and angels, including Niranjan Tulpule, VP of Developer AI at Google, Gokul Rajaram, and Jason Calacanis.

Where this goes next

Diagnosis is the first step. The direction of the product is automated remediation, where the system does not just explain why a build broke but proposes and eventually applies the fix. Hermetiq also exposes structured build context to coding agents through MCP, so the agent writing the code can see what happened when that code failed.

Put those together and the shape of the company becomes clear. Every build to green, every agent closer to autonomy, with the build system as the source of evidence rather than another source of questions.

Congratulations to Jake, Tim and the entire Hermetiq team. We are glad to be part of this one.