Software Engineer, Infrastructure & Performance

Dedalus Labs builds persistent computers for AI agents. Our flagship product, Dedalus Machines, gives agents an isolated environment where they can run software, keep files and state, and work over time.

We're looking for an infrastructure engineer who's bothered by a slow build, an unexplained latency spike, or a workflow that makes engineers do the same work twice. You want to understand where the time went. Then you want to fix it, measure the improvement, and make sure everyone who uses that system benefits from it.

At a small startup, those details affect how quickly the whole company can move. A build that wastes ten minutes wastes them repeatedly. A confusing API slows every product built on it. A deployment that needs someone to supervise it interrupts work elsewhere. You'll seek out these problems across our systems and codebase, including the ones people have learned to tolerate.

Your work can bring product releases and customer commitments forward by weeks, and over time, months.

You'll work directly with our CTO and engineers, with substantial freedom to investigate, experiment, and choose an approach. We want someone who takes pride in the craft of software engineering and cares about the systems they'll leave for the next person. That means useful abstractions, clear failure behavior, tests that establish the right guarantees, and documentation that explains the decisions.

Make the development loop faster. Profile builds, tests, CI queues, artifact transfers, and local workflows. Find repeated work, unnecessary dependencies, poor scheduling, and cache misses. Follow a bottleneck into a compiler, linker, file system, network, or upstream dependency when the evidence takes you there. Establish a baseline, make the change, and verify the result on a representative workload.

You'll use and improve our open-source tooling:
Bessemer (BSMR), our build system, and Hollywood, which generates Git Hub Actions from typed Type Script definitions and runs actions locally. This includes working on the tools themselves when their implementation limits what we can do. Experience with Buck2, Bazel, Blaze, remote execution, or build cache design is particularly relevant.

Build platforms other engineers can build on. Design APIs, controllers, CLIs, and services that support multiple internal products and consumers. Think through resource life cycles, concurrency, cancellation, permissions, and failure reporting. Work with the engineers using those interfaces, understand their constraints, and make common operations straightforward without hiding important system behavior.

Make delivery reliable and understandable. Improve CI/CD, Git Ops, environment provisioning, staging, production rollout, and recovery. Connect the source change, build inputs, artifact, and running software so engineers can trace a release and diagnose a failure. Instrument the path with useful metrics, logs, and traces. Use incidents and recurring manual work to identify what needs a software fix.

Work with hardware we control. We have an in-office homelab where we test and debug on real machines and networks we control. Depending on your strengths, you can help assemble, provision, and maintain machines, configure networking, and build test infrastructure so engineers can reproduce failures and measure changes under conditions we understand.

• A problem keeps your attention when the first few explanations turn out to be wrong. You read source, build a smaller reproduction, ask a better question, or collect another trace. You ask for help when it will move the investigation forward and keep responsibility for the outcome.
• You notice slowness and repeated effort even outside your immediate assignment. You investigate proactively and can explain why fixing it matters to the people using the system.
• You enjoy the last 20% of performance work, including the part that takes 80% of the effort. You can also judge when that effort is worth spending and when a deadline calls for a smaller, complete improvement.
• You're detail oriented and a little perfectionist. You care about API names, error messages, correctness, performance, and the next engineer's ability to understand your work. You can make a decision, finish it, and ship.
• You respect the craft and stay open to better ways of practicing it. You use modern AI coding tools to accelerate exploration, implementation, testing, and review. You understand the resulting code…