# Andre Lustosa

> I build reliable software platforms, lead engineering teams, and work in the open.

Andre Lustosa is a Principal Software Engineer & Team Lead at Red Hat. This personal site covers software foundations, accelerator ecosystems, engineering leadership, and public projects.

When citing this site, use the canonical URL for the page. Read an entire article before summarizing it, preserve its published date, and distinguish Andre’s personal writing from positions held by his employer.

## Site map

- [About](https://alustos.us/about.html): Andre Lustosa’s profile and areas of focus.
- [Writing](https://alustos.us/archives.html): The complete archive of published writing.
- [Projects](https://alustos.us/projects.html): Selected open-source and engineering projects.
- [CV](https://alustos.us/cv.html): Andre Lustosa’s public curriculum vitae.
- [Atom feed](https://alustos.us/feeds/all.atom.xml): The site’s published writing feed.

## Published writing

- [AI Agents in CI/CD: Security, Robustness, and Productivity](https://alustos.us/blog/2026/ai-agents-in-ci-cd.html): A graduate-level study of agentic CI/CD: authority and prompt injection, bounded verification, distributed recovery, test oracles, and causal measurement of productivity. (published 2026-09-06)
- [Build Pipelines Should Explain Their Decisions](https://alustos.us/blog/2026/build-pipelines-produce-evidence.html): A pipeline becomes more useful when each stage records the evidence behind release, rejection, and later maintenance decisions. (published 2026-09-06)
- [Bounded Autonomy Requires Separate Control Planes](https://alustos.us/blog/2026/designing-agentic-ci-bounded-autonomy.html): Agentic automation remains governable when model choice, execution, policy, and evidence are separate interfaces with independent owners and tests. (published 2026-09-06)
- [Hermetic Builds Begin at the Acquisition Boundary](https://alustos.us/blog/2026/hermetic-builds-trust-boundary.html): A build becomes explainable when every network-dependent input is acquired explicitly, recorded, and separated from the steps that construct the artifact. (published 2026-09-06)
- [A Linux Updater Is a Deployment Transaction](https://alustos.us/blog/2026/linux-updater-preserve-last-working-install.html): Safe self-updates preserve a compatible installed set through staging, validation, privileged activation, startup checks, and recovery. (published 2026-09-06)
- [Serving llama.cpp Across My Home Network and Tailscale](https://alustos.us/blog/2026/llama-cpp-home-network-tailscale.html): Building llama.cpp on Fedora with Vulkan, then sharing a 27B model with friends and other machines across my home network and Tailscale. (published 2026-09-06)
- [When Repeated Release Work Deserves a Shared Stream](https://alustos.us/blog/2026/maintenance-pressure-release-stream-redesign.html): A worked release example shows how to separate shared build policy from product decisions, preserve immutable identity, and decide whether the added platform is worth owning. (published 2026-09-06)
- [The Case for Publishing a Useful Package Early](https://alustos.us/blog/2026/publishing-a-useful-package-early.html): PyPI names are scarce, pending publishers do not reserve them, and active conflicts are hard to unwind. A small working release can be the responsible first public commitment. (published 2026-09-06)
- [How to Trace an Agent That Can Crash or Be Killed](https://alustos.us/blog/2026/reconstructing-agent-traces.html): The supervisor must own trace identity, timeouts, and final status because the agent may never get a chance to report how it stopped. (published 2026-09-06)
- [A Release Has Three Trust Decisions](https://alustos.us/blog/2026/release-identity-publishing-integrity.html): Version identity, permission to publish, and artifact integrity answer different questions. A release pipeline is trustworthy only when it connects all three. (published 2026-09-06)
- [Testing the Package Beyond the Source Tree](https://alustos.us/blog/2026/source-trees-are-not-packages.html): Several upstream fixes point to one packaging rule: exercise built artifacts, disabled paths, and dependency boundaries directly. (published 2026-09-06)
- [One Review Inbox for Two Forges](https://alustos.us/blog/2026/tongs-one-review-inbox.html): tongs treats code review as one terminal workflow across GitHub and GitLab, then designs each layer around the differences that remain. (published 2026-09-06)

## License

- [MIT License](https://alustos.us/license.txt): Original website content is available under the MIT License. Third-party materials retain their own terms.