TL;DR: Build infrastructure is the bottleneck determining who wins in AI.
Something strange is happening in software development. The industry just handed every engineer a 10x productivity multiplier — AI coding tools like Claude, Cursor, and Copilot — and yet some software processes have gotten slower.
In the last two years, agentic AI has fundamentally changed how software gets written. Cursor became the fastest SaaS product ever to hit $100M ARR. Claude, Cursor, and Copilot are now embedded in the daily workflows of millions of developers. The code is flowing.
But code doesn't become a product until it builds, passes tests, and deploys. And that infrastructure — the build and compilation layer — was never designed for this volume.
Here's the math that should worry every engineering leader: if your AI tools are helping developers produce 3-10x more code, but your build system processes at the same speed, you've just created a bottleneck that scales linearly with your investment in AI tooling. More code means more build targets. More build targets mean longer build times. Longer build times mean developers sitting idle, context-switching, and losing flow state.
Across the industry, this adds up to a staggering problem. There are over 47 million developers worldwide, and surveys show that engineers spend an average of 57 minutes per day just waiting on builds and tests. Multiply that across the global developer workforce, factor in fully-loaded engineering costs, and you arrive at a bottleneck measured in the hundreds of billions of dollars.
The companies feeling this most acutely? The ones driving the AI boom itself.
The Great Bazel Migration
The industry's response has been a mass migration to Bazel, Google's open-source build system. Originally developed internally at Google as "Blaze" to handle one of the largest monorepos on Earth, Bazel was designed from the ground up for hermeticity, reproducibility, and massive parallelism. It's the build system you reach for when nothing else scales.
And right now, everyone is reaching for it. The official list of Bazel adopters reads like a who's who of elite engineering organizations: Stripe, Uber, Pinterest, SpaceX, and dozens of other major engineering organizations run on Bazel. The AI companies are following suit — many staffed by ex-Google engineers who know firsthand that Bazel is the only build system that can handle what's coming.
The migration makes sense. Bazel's architecture — content-addressable caching, remote execution, fine-grained dependency graphs — is uniquely suited to the demands of modern monorepos. When you're iterating on a million-line C++ codebase for an autonomous vehicle stack, or a massive Python monorepo for a foundation model, Bazel is often the only viable option.
But there's a catch that nobody talks about until they're deep in it.
Bazel Is Powerful. It's Also a Black Box.
Bazel solves the scale problem. It does not solve the understanding problem.
When a Bazel build fails — and at scale, failures are constant — debugging it is notoriously difficult. The hermeticity that makes Bazel reliable also makes it opaque. Build graphs become enormously complex. Cache invalidations cascade in non-obvious ways. Remote execution introduces network-dependent failure modes. A single misconfigured rule can silently break builds across an entire organization.
Engineers who've operated Bazel at scale know the feeling: staring at a failed build, scrolling through thousands of lines of logs, trying to figure out why. The build system that was supposed to save time becomes its own time sink.
This is the first gap we're attacking: the ability to see inside the build.
Hermetiq: Observability for Build Systems
Hermetiq provides observability purpose-built for Bazel. The simplest way to understand what we do: think Datadog, but for build infrastructure.
We instrument the entire build lifecycle — from target resolution to remote cache hits to test execution — and surface the insights that let engineering teams understand, optimize, and debug their builds in real time. When something breaks, Hermetiq shows you exactly what failed, why, and what changed to cause it. When builds are slow, we show you where the time is going and what to fix first.
We launched invite-only a few months ago. After processing millions of build events per day, we launched publicly on February 9th, 2026.
Jason Calacanis — host of the All-In Podcast and founder of Launch.co, backed Hermetiq in December. Hermetiq also received investment from a VP at Google.
Why "Build Observability" Is Becoming an Infrastructure Category
We're not the only ones who see this opportunity. BuildBuddy raised a $3.15M Series A. EngFlow raised $18 million from Tiger Global and Andreessen Horowitz. Aspect Build raised $3.85 million. The capital is flowing into this space because the pattern is unmistakable.
If you were paying attention in 2015 and 2016, this looks familiar. That was the era when "observability" went from a niche concern of SRE teams to a recognized infrastructure category. Datadog, which IPO'd in 2019, is now worth roughly $45 billion. The insight was simple: as systems became more distributed and complex, the ability to understand what was happening inside them became critical infrastructure, not a nice-to-have.
Build systems are undergoing the same transition. As codebases grow, as AI tools accelerate the volume of changes, and as companies migrate to sophisticated build systems like Bazel, the ability to observe and understand build behavior becomes essential. It's not optional infrastructure. It's the layer that determines whether your engineering organization can actually ship.
The AI & Robotics Wave Will Make This Urgent
Here's what makes the timing especially critical: we're on the edge of a robotics AI wave that will push build infrastructure to its absolute limits.
Companies like Physical Intelligence are already demonstrating what's possible — they're building general-purpose robot foundation models that require massive, heterogeneous monorepos. Physical Intelligence cut their iteration cycle — what they call "time-to-robot" — from 90 minutes down to 5 minutes by investing heavily in their Bazel build infrastructure.
In a field where iteration speed determines who reaches general-purpose robotics first, build infrastructure isn't a backend concern. It's a competitive weapon.
And Physical Intelligence is just one company. The entire autonomous vehicles industry, every humanoid robotics startup, every company training and deploying foundation models — they all face the same constraint. The speed at which they can iterate on massive, multi-language monorepos determines how fast they can ship.
Build infrastructure is the bottleneck determining who wins in AI.
The Endgame: Self-Healing Builds
Observability is just where we start.
Every build event Hermetiq processes — every failure, every cache miss, every flaky test, every configuration change — adds to what is becoming the largest proprietary dataset of build intelligence in the world. We're not just watching builds. We're learning from them.
That dataset is the foundation for what we're building toward: AI agents that automatically diagnose and fix build failures. Self-healing builds.
Imagine a world where a broken build doesn't page an engineer. Instead, an AI agent identifies the root cause, generates the fix, validates it, and merges it — all before the developer who triggered the failure finishes their coffee. This is the natural endpoint of combining deep build observability with the kind of AI reasoning that already exists today.
The companies building the most sophisticated software on Earth — autonomous vehicles, foundation models, robotics systems — run monorepos with millions of lines of code across multiple languages. When their build infrastructure can fix itself, they don't just save engineering hours - they compress their entire R&D cycle and ship faster.
See Hermetiq in action. Explore how our build intelligence platform gives you real-time visibility into cache misses, build costs, and performance — so you can ship faster.
Related Articles
- What Is Build Observability? — The discipline of applying production-grade monitoring to your build system.
- What Is Hermeticity and Why It Matters — The property that makes caching, parallelism, and reproducibility work in Bazel.
- Why Everyone Is Talking About MCP — How the Model Context Protocol connects AI assistants to build infrastructure.