Indeed’s blog post introduces the idea of a Web 4.0 “agentic” era and argues we must redesign identity and access management for large-scale autonomous/AI agents. It highlights the need for richer authorization, trustworthy delegation, verifiable auditing, and preservation of speed, resilience, and privacy. The post is the first in a series exploring Agentic IAM.
Indeed migrated its CI platform from Jenkins/Hudson (CloudBees) to GitLab CI to address scaling, resiliency, and maintainability problems. They implemented a "golden path" of GitLab CI templates, used InnerSource to accept external contributions, ran batch MRs (Sourcegraph), nudged/stewarded teams, and retired Jenkins before their license expiration — achieving higher pipeline throughput, lower costs, and broad template adoption.
Indeed describes their workload identity platform built on SPIRE and SPIFFE to provide x509 and JWT identities for Kubernetes workloads and Istio. The guide covers architecture (root/intermediate CAs, nested topology), integration and compatibility issues with Istio (SPIFFE ID format, agent socket path/SDS), key management (disk vs AWS KMS KeyManager), OIDC federation with Confluent and AWS (JWKS limits, certificate thumbprints, audience/trust differences), a custom CredentialComposer plugin for session tags, and operational lessons/tuning (k8s scheduling, VPA, DaemonSet rolling updates). It highlights pitfalls encountered and practical recommendations for operating SPIRE at scale as an OIDC provider.
Indeed argues that single-point metrics (JSV delay, domContentLoadEnd, LCP, TTI) fail to capture holistic perceived client-side performance across different page architectures (SSR/CSR/SPAs). The author recommends using a composite metric (Lighthouse score composed of LCP, TBT, CLS, FCP, SpeedIndex) and corroborating metrics with page load behavior to better represent user experience.
Indeed Interview converted dashboard clicks into features, trained a CatBoost tree-based model and used SHAP decision/summary/dependency plots to identify positive and negative UI predictors. They translated those insights into A/B test hypotheses and UI changes that produced ~5–10% uplifts in interview-invite metrics.
Indeed's Dev Blog describes the company's SRE organization: what SRE is, how Indeed's embedded SRE model works (SLOs/SLIs, IaC, automation), technical initiatives like AWS migration and Kubernetes adoption, incident analysis and operational excellence, interviews with SRE staff, and a recruiting call-to-action.
Indeed measured page load performance using firstContentfulPaint and domContentLoadedEventEnd, improved load times via techniques like flushing the HTML head early, reducing render‑blocking files, and precompiling Emotion CSS‑in‑JS components with a webpack plugin. Using RUM and the Network Information API, they found device and network differences dominate performance; reliability (e.g., via CDNs) often matters more than small speed gains for engagement.
An Indeed engineering-leadership blog arguing that accountability is better built through a culture of ownership and an opportunity-driven coaching model than by naming obligations. The author describes practices (blameless retrospectives, intake processes, RACI/OKRs) and gives infrastructure examples (postfix, message queues, key-value store → ceph/s3, cassandra, oracledb) to illustrate when teams take or decline responsibility.
Alex Elman (Indeed) argues for shifting from a "prevent & fix" safety mode to a "Learn & Adapt" mode so organizations can sustain resilience as scale and complexity increase. The article focuses on changing culture and practices around incident analysis, normalizing assumptions and questions, cross-role cooperation, broad communication of postmortems, and fostering advocates to drive organizational change. Examples include incident practices and mentions of technologies/practices like RabbitMQ, MySQL (safe mode), Slack, chaos engineering, and postmortems.
Indeed announces 2021 updates to its FOSS Contributor Fund: it will award three projects per quarter (each receiving $10,000), changes voting eligibility and cadence, opens nominations and voting windows, and published a GitHub Pages resource to share how the fund is organized and help other companies adopt a similar program.
Indeed’s Open Source Program Office coordinated Hacktoberfest activities by selecting six Indeed-used projects with at least one Indeed maintainer, asking maintainers to tag issues for contributors and to target a three-day turnaround for responses. They supported contributors via 11 external office hours, 29 internal virtual study halls across four time zones, and a timed mentorship program pairing participants by timezone and experience. For automation and discovery, they open sourced Mariner Issue Collector as a GitHub Action to find beginner-friendly issues and used Starfish to generate accurate contribution lists and FOSS fund eligibility. Reported metrics include 437 commits, 328 submitted pull requests, 100 total participants, and an increase of over 200% in active recurring participants (contributors contributing on two or more days in the quarter).
The post describes k8dash, an open-source Kubernetes dashboard that interacts with the Kubernetes API to provide cluster visualization, resource management, and a YAML editor. k8dash uses a lightweight single-service architecture: a ~150-line Node.js server that proxies frontend requests to the Kubernetes API (using express, http-proxy-middleware, @kubernetes/client-node, and openid-client) and a React client built with create-react-app. It supports quick installation via a few YAML manifests, OIDC-based AuthN/AuthZ via environment variables without additional proxies or databases, and integrates with Metrics Server to surface CPU and memory metrics. Runtime features include real-time dashboards and charts, streaming cluster events, viewing pod logs, and an in-browser terminal to exec into pods. The project exposes context-aware API documentation and points to its GitHub repository for contribution and issue tracking.
Tatu Saloranta (Jackson creator) outlines challenges from Jackson’s growth: documentation structure and freshness, project website/branding, managing/prioritizing large changes, collaborative issue triage, and testing compatibility with downstream frameworks. He proposes concrete steps (how-to guides, improved Javadoc/wiki tooling, JSTEP mini-specs, issue templates/labels, volunteer triagers, and cross-project SNAPSHOT-based integration tests) and calls for community contributions via Hacktoberfest, Gitter, and the jackson-dev mailing list.
Overview of Jackson (Java JSON library): core JSON read/write capabilities, usage in frameworks (example with Spring Boot), and its extensible module system (data format modules, data type modules, and JVM-language modules) that add support for XML, YAML, CSV, binary formats (CBOR, Smile, Avro, Protobuf, Ion, MessagePack, BSON), and integrations with libraries like Guava, Hibernate, and Joda-time.
An experience-driven guide for engineering managers considering doing IC coding: explains the challenges (context switching, priorities), argues benefits (empathy, modeling vulnerability, mental health from shipping), and gives practical advice — ensure team stability, pick small non-urgent "unicorn" projects that play to your strengths, block deep-work time, and design delegatable MVPs. Examples use data-focused work (A/B tests, data wrangling) and include brief code/library references.
Indeed describes a cross-team effort to remove exclusive and harmful terminology from their codebase—explaining the rationale, recommended replacements, governance (wiki, quick reference), tracking (Jira labeling), a Git branch renaming case study (master -> primary), challenges encountered, and the ongoing grassroots process to make the codebase more inclusive.
Announcement of the FOSS Responders working group and a virtual funding event to support open source conference organizers and individuals impacted by event cancellations, with calls to donate, request help, and join the initiative.
An Indeed SRE reflects on shortcomings in the company’s incident retrospective process and recommends changes to increase learning: decouple remediation-item creation from retrospectives, use blank documents instead of rigid templates, make retrospectives optional, develop a debrief-facilitator program, and focus more on human factors and organizational communication to drive lasting improvements.
The article introduces the d-curve: a method to define non-contractual churn by choosing the churn-period d that minimizes a weighted average of type I and II errors. It provides theory, an implementation experiment (select validation period D, compute errors e1(d), e2(d) over candidate d values), and an applied example using Indeed job-sponsorship monthly data (optimal d=3 months for w=0.5, D=12). The approach outperforms naive percentile heuristics and can be used for threshold selection in classification more generally.
The article investigates a CPU throttling regression caused by a change to CFS-Cgroup expiration logic in the Linux kernel (commit 512ac999). The author explains how the scheduler's per-CPU slice expiration amplified throttling on high-core-count machines for bursty multi-threaded workloads, describes the debugging and patching process, and summarizes the upstream fix (removing expiration logic) that landed in 5.4+ and was backported to various stable kernels and distributions. The post closes with mitigation advice (monitor throttled percentage, upgrade kernels, use whole CPU quotas in Kubernetes) and pointers to related kernel/Kubernetes discussions.
