Skip to content

BazelCon 2026 Mega-Recap: The EngFlow Takeover, Scale in the Age of AI, and Building the Future of Developer Experience

By the EngFlow Team

Amsterdam hosted more than 500 build engineers at BazelCon 2026, and one reality framed nearly every discussion: automated coding tools are hammering developer infrastructure.

When autonomous agents generate thousands of pull requests an hour, local laptops lock up and traditional CI pipelines buckle. Across three days of keynotes, workshops, and late-night hallway debates, our team at EngFlow focused on solving that exact crunch. Google Bazel lead Tobi Werth joked from the stage that the event felt like an "EngFlow takeover," reflecting the nine technical talks, four sold-out training tracks, and new open-source projects our engineers brought to the community this year.

The Foundation and the Roadmap

Helen Altshuler joined Glydways' Chuck Grindel to open the event with a Bazel Community Update (14 min) that announced the Linux Foundation's new BUILD Foundation. The foundation establishes a vendor-neutral home to finance open-source developer tooling and coordinate long-term roadmaps across enterprise adopters and the core Bazel team at Google.

During the State of the Union (44 min), Google's John Field and Tobias Werth laid out the technical milestones slated for Bazel 10:

  • Java 27 & Chunking: Upgrades to Java 27, Skyframe cache pruning, and content-defined chunking (CDC) drop memory footprints and speed up execution by roughly 15%. Lower overhead is especially relevant for developers running concurrent local Bazel instances alongside active background agents.
  • Skymetal: Google revealed Skymetal, an exploratory project testing a progressive rewrite of Bazel's core engine in Rust to improve overall throughput and memory usage.
  • Native Starlark Typing: Bazel 10 introduces formal type-checking for Starlark rules, building directly on the syntax parsers landed in recent releases.

Technical Training Courses - Now Recorded!

EngFlow hosted several packed training courses just before the conference officially kicked off, and they are available in full, long-form for those who weren't able to get into the sessions.

Field Notes from the EngFlow Team

Our sessions focused on production bottlenecks, real-world customer migrations, and the engineering required to support high-velocity build environments.

Nicolas Lopez teamed up with Julio Merino from Snowflake (29 min) to examine how agentic workflows break developer machines. Snowflake relies on heavy Git worktree usage and frequent parallel builds; routing that bursty load into EngFlow's managed remote execution cluster keeps developer environments responsive.

In a related systems talk, Helen Altshuler and Plaid's Nikita Chepanov (12 min) walked through the failure modes of Plaid's CI cluster after AI tooling doubled their build volume, detailing the autoscale thresholds and AWS SLAs they established to stabilize performance.

On the developer tooling front, Armando Montanez and Ulf Adams debuted any2bazel (31 min), an open-source project that uses diff-checking to translate CMake and MSBuild configurations into deterministic Bazel rules. Han-Wen Nienhuys demonstrated Horapha (11 min), using Linux CRIU to snapshot warmed-up Bazel JVM states and cut cold-start delays.

Next, John Cater and Antonio Di Stefano shared an API (31 min) for persistent test runners that untangles test startup from run invocations, trimming Bazel's internal test suite runtime by 46%. Di Stefano also teamed with Yannic Bonenberger (32 min) to present rules_supply_chain, extracting CycloneDX and SPDX software bills of materials straight from target dependency graphs.

Jay Conrod (creator of Gazelle) presented (10 min) how to support new rule sets by writing an extension with Gazelle's new v2 API, which supports better error handling and dependency resolution.

With all this going on, we also formalized our partnership with Cursor to design build infrastructure on Cursor that is specifically tuned for autonomous development lifecycles.

CI at Massive Concurrency

Case studies from across the industry reinforced the scale challenges modern teams face:

  • Massive Concurrency: OpenAI outlined (29 min) the architecture behind its internal build systems, which run over 100,000 parallel agents against shared Buildbarn and EngFlow remote caches on Kubernetes.
  • Storage and Deduplication: Tyler French (BuildBuddy) and Benjamin Ingberg (Meroton) demonstrated (34 min) that rolling-hash content-defined chunking cut disk cache utilization by 40% and written bytes by 35%. Epic Games walked through Lore, their BLAKE3-backed asset store integrated with Bazel's remote cache, while Canva explained how Build Event Service streaming keeps automated screenshot tests from polluting Git history.
  • macOS Sandboxing: OpenAI's David Zbarsky shared actiond (11 min), a daemon that executes strict, fast sandboxing on macOS through a local Linux VM over vsock. Aspect's Borja Lorente detailed limitations in the macOS FSKit layer and proposed a pluggable --sandbox_backend interface.
  • Fast Developer Tooling: Tamir Duberstein showcased sty (x min) for fast, cross-file Starlark linting, while David Zbarsky's rules_rs cut Cargo lockfile regeneration cycles from ten minutes down to 100 milliseconds.

Between sessions, our coffee cart served steady lines, and conversations at the booth with engineering leads from Datadog, Rivian, Volvo, ASML, Spotify, and Qualcomm centered on the same priority: keeping build loops under a few seconds when code generation runs non-stop.

The shift toward machine-authored code makes reliable remote execution and fast caching mandatory infrastructure for any serious engineering organization. We are already turning these discussions into production deployments, and we will see everyone back in Amsterdam for BazelCon 2027.