# Sam Alba

Sam Alba builds Mendral, an AI-native DevOps platform that uses agents to automate platform-engineering work, including CI maintenance and supply-chain security reviews.

- Profile: https://b2binfluencer.marketing/creators/sam-alba/
- LinkedIn: https://www.linkedin.com/in/samalba
- LinkedIn followers: 7,025
- Topics: artificial-intelligence, supply-chain, startup, ai-agents, software-supply-chain-security
- Based on 12 dated posts, latest 2026-09-21

## Summary

Sam Alba is most useful for technical posts on CI automation and supply-chain security for platform engineers.

- Best for: A DevSecOps platform hiring a founding engineer or a technical launch for platform and security engineers.
- Not for: Brands seeking sponsored content, fixed deliverables, reach guarantees, or broad non-technical consumer audiences.
- Check first: Confirm whether they offer creator packages or only Mendral product access and hiring.

### Takeaways

- Supply-chain security belongs in CI, before merge.
- Treat dependency updates as untrusted code contributions.
- Mendral weights packages published in the last 72 hours as the active attack window.

## Main ideas

- **Supply-chain security belongs in CI, before merge**: Sam Alba argues that CI is the right enforcement point for supply-chain security: it sits between a developer opening a pull request and production shipping. Every dependency change, new package, version bump, and action invocation passes through CI. The described agent triggers when a pull request touches a dependency manifest, lockfile, Dockerfile, or Actions workflow. It reads source code and transitive dependencies, flags typo-squatting, malware, and suspicious lifecycle scripts, and scrutinizes packages published in the last 72 hours more closely because that is the active attack window. Waiting for an advisory leaves a gap a normal scanner will miss.
- **Treat dependency updates as untrusted code contributions**: In a post pointing to Olivier Gambier’s argument on the Mendral blog, Sam Alba highlights the position that auto-merging Dependabot pull requests is now the leading route for supply-chain compromise. The cited incidents include Shai Hulud, Nx, chalk/debug/qix, and tj-actions/changed-files. The stated model is to review a dependency update like a pull request from a stranger: run the package in a sandbox, check for post-install scripts, new transitive dependencies, and behavioral drift, then approve, request changes, or block. Separately, they describe agents that open upgrade pull requests by reading the changelog, examining how a package is used in the codebase, and making required code changes, with CI validating the result and anything unverifiable marked for human review.
- **Automate maintenance or accept the risk**: Sam Alba argues that CI pipelines are critical infrastructure, but the work that keeps them secure—upgrading pinned actions, fixing broken scans, and keeping dependencies current—is still manual at most companies. They propose an always-on agent that treats delivery reliability as a first-class problem, catching issues before they surface rather than scrambling after a disclosure.
- **Let user workarounds—not procurement checklists—set product direction**: In an early-startup post, Sam Alba warns that large enterprise prospects can hand over long requirement checklists and promise multi-year contracts if every box is checked. Teams can spend months building role-based access controls and custom corporate networking for purchasing managers who will never use the software, then still lose the deal—while making the product harder for actual engineering teams to use and slower to onboard. They argue that product direction should come from watching where users expend their own energy. In this view, user workarounds are the roadmap.
- **Keep the agent harness outside the sandbox to limit token bleed**: Sam Alba describes a technical priority behind Mendral’s autonomous work: a tiered model strategy and an agent harness kept outside the sandbox to avoid token bleed. The goal is to make autonomous CI investigations practical at scale without treating every task as an expensive model call. The company describes connecting an agent to ClickHouse for a SQL interface across terabytes of raw logs, ingesting over a billion log lines a week, and handling 16,000 autonomous CI investigations a month.

## Work to read

- [Nine months of Mendral, month by month](https://www.linkedin.com/posts/samalba_heres-all-the-crazy-stuff-thats-happened-activity-7477771705638227970-wmYe)
- [The Trivy compromise and the upgrade that came 11 days earlier](https://www.linkedin.com/posts/samalba_last-week-aqua-securitys-trivy-action-github-activity-7441853851546316801-cMy7)
- [Dependency updates treated as untrusted code](https://www.linkedin.com/posts/samalba_you-should-not-update-your-dependencies-activity-7465224581201698818-QEta)
- [How the supply-chain agent decides what is suspicious](https://www.linkedin.com/posts/samalba_something-we-watched-play-out-several-times-activity-7463608366808940544-ZhvT)
- [Watch the workaround, not the enterprise checklist](https://www.linkedin.com/posts/samalba_during-the-early-days-of-a-startup-large-activity-7481033204037230593-_1JK)

## Brand relationships

- **Mendral** (owned_business)
- **Dagger** (employment)
- **Y Combinator** (disclosed_partnership)
- **Anyshift.io** (client)
- **PostHog** (client)
- **Aqua Security** (mention)

## Partnerships

Sam Alba’s posts promote Mendral, the AI-native DevOps platform they co-founded, or discuss engineering strategy; they don’t describe sponsored content, brand collaborations, affiliate arrangements, or creator packages. Two posts invite interested readers to get in touch: one asks readers curious about how Mendral works to “ping me,” and another offers access to limited accounts to try the product. Both invitations concern trying or learning about Mendral, not working with Sam as a creator. The posts publish no partnership format, deliverable, fee, sponsorship policy, or official inquiry route. Ask Sam for current sponsorship terms and a partnership contact; the posts don’t say whether they accept sponsorships.

## Alternatives

- Selim Erünkut: For the same supply-chain security campaign, choose Selim Erünkut when the buyer needs a forensic walkthrough of a real attack or an architecture-level argument, not a product build story. The March 2026 post reconstructs the Trivy compromise step by step: a stolen personal access token three weeks before disclosure, the aqua-bot takeover, a malicious v0.69.4 tag pushed through the official release pipeline to GitHub Releases, Docker Hub, GHCR and Amazon ECR, 75 of 76 version tags force-pushed in trivy-action, malware reading runner memory from /proc and harvesting SSH keys, cloud credentials and Kubernetes tokens to a typosquatted domain, a persistent systemd backdoor polling a blockchain canister, and then CanisterWorm hitting 47-plus npm packages with 28 hijacked in under 60 seconds. It closes with defences a security audience can act on: pin commits, verify signatures, monitor CI runners as production workloads, and rehearse full credential invalidation because partial cleanup creates the next incident. The May 2026 post argues that agent failures are identity, context and governance problems rather than model quality problems, and lays out four architectural bets around first-class agent identities, platform-level integration, durability beyond a session, and buying platforms instead of rebuilding plumbing. Briefs needing incident analysis or a platform-architecture perspective fit this voice more directly than Sam Alba's first-person build updates.
- Bartosz Jurga: For the same supply-chain security campaign, choose Bartosz Jurga when the audience is the board and the message is risk, resilience and regulatory exposure rather than implementation. The September 2026 post frames the keyv compromise and the ChainDrop campaign in executive terms — blast radius across CI/CD runners that hold deployment secrets and production access, compliance exposure under NIS2 and ISO 27001, and business continuity — and links to an executive brief. The August 2026 post takes the ransom-payment decision to leadership: low restoration rates when victims pay, repeat attacks against organisations that pay, legal and regulatory sanctions, and the argument that immutable, air-gapped and tested backups make ransom demands obsolete, plus the shared-responsibility gap where cloud providers guarantee platform uptime but not the buyer's own data recovery. For a CISO- or board-facing resilience programme, this framing does work Sam Alba's engineering posts do not attempt.
- Toby Van de Grift: For the same supply-chain security campaign, choose Toby Van de Grift when the campaign is about third-party and supplier risk governance — the accountability question rather than the tooling question. The October 2025 posts use JLR, M&S and the Co-op to argue that supplier breaches are a resilience problem, not a procurement checklist: GDPR keeps the controller responsible even when a breach starts at a processor, NIS2 requires third-party assurance to be demonstrable and continuous, and DORA requires ICT third-party risk to be monitored, tested and reported. One post asks boards whether supply chain risk still sits in procurement or has become a resilience function; the other insists partners prove their security through SAFE reports, SBOMs and CBOMs on the principle that you inherit every weakness a supplier fails to mitigate. The hashtag set (#TPRM, #GRC, #NIS2, #DORA) and the ReversingLabs mention place this squarely in the governance and third-party-risk conversation, where many enterprise buying decisions actually get made.

---
AI-assisted summary of public posts. Audience demographics are not available. Engagement appears only where we checked recent posts. Method: https://b2binfluencer.marketing/methodology/