Fred de Villamil

Beyond the Build: Engineering Lessons from the #ApacheKafka CI Overhaul 🛠️ The 2025 modernization of Apache Kafka’s CI builds wasn't just a technical migration to GitHub Actions...it was a masterclass in removing organizational friction. The #DPE playbook from Kafka committers: 1️⃣ Build Acceleration: Using build caching and test distribution to reclaim thousands of engineering hours. 2️⃣ Reliable Signal: Using observability to eliminate flaky tests that erode developer trust. 3️⃣ Scalable Infrastructure: Moving away from brittle, hard-to-maintain legacy scripts toward a modular, observable CI pipeline. One year later, the results speak for themselves: a more productive community and a more resilient software supply chain. 👉 Join us Mar-12 for lessons learned and tools used: https://lnkd.in/dPdN3W5C #Kafka #ApacheKafka #Develocity #GitHubActions #GHA #CImigration #CI #Jenkins

1

GraalVM Native Image turns Java apps into native executables – faster startup, smaller footprint, and better fit for cloud workloads. In our new video, Catherine Edelveis shows how to take a Spring Boot app and compile it to native in one command. Along the way we compare distributions (Liberica NIK, Oracle GraalVM, Mandrel), explain how Native Image works, and troubleshoot common pitfalls. Watch the full tutorial here: https://hubs.li/Q03Kh_DR0

3

Your CI/CD pipeline takes 45 minutes. Developers merge to main and go to lunch. You call this continuous delivery. We call it continuous waiting. The symptoms: - 30-60 minute pipeline runs for basic changes - Flaky tests that fail 20% of the time for no reason - No clear rollback strategy - "just revert the commit and wait another hour" - Developers merge to feature branches to "test in CI" because local dev is broken - Teams bypass CI/CD entirely for hotfixes because it's faster to YOLO to production Here's the truth: if your deployment pipeline takes longer than a coffee break, developers will route around it. And they should. A slow pipeline isn't just annoying. It's a productivity tax that compounds: - Context switching kills flow state - Batch changes get bigger and riskier - Feedback loops stretch from minutes to hours - Fear of breaking things slows innovation What actually fixes it: - Parallel test execution - don't run 500 tests sequentially - Fail fast on static analysis before expensive integration tests - Incremental builds - rebuild what changed, not everything - Reliable tests or no tests - flaky tests are worse than no tests - Deploy with feature flags, not by waiting for perfect code At WANTED.solutions, we've cut customers pipeline times from 60 minutes to under 10 minutes while increasing reliability. The secret isn't exotic tools. It's brutal prioritization of what and when actually matters. How long does your deployment take? Be honest. #wantedsolutions #creativepioneers #ci #cd #deployments #pipelines

5

Big day for Linkup. We raised $10M to scale our web search engine for AIs, and we're launching Linkup /fast: the world's most accurate sub-second web search API. In 2013, I helped build Niland with Damien Tardieu, Christophe Charbuillet, Johan Pagès, and the rest of the team. We created a search engine that could find music by listening to it — using vector search and embeddings years before they became standard. Then Spotify acquired us. I thought I was done with search engines. Turns out I wasn't. There's something about the search problem that pulls you back. It's deceptively simple: someone asks a question, you find the answer. But the complexity underneath - understanding intent, ranking relevance, doing it at scale, doing it in milliseconds - that never stops being interesting. In 2024, Boris Toledano, Philippe Mizrahi, and I realized there was unfinished business. Google solved search for humans. But AIs don't search like humans do. They need structured information via API, in real-time, not blue links to SEO-optimized pages. We're building that infrastructure at Linkup. And we're looking for engineers and researchers who see search engines as puzzles worth solving. If that's you, let's talk.

145

15 commentaires

Ever had a CI/CD pipeline that started fast but gradually became painfully slow as your codebase grew? We’ve seen teams wait 10–15 minutes just to get basic feedback — not because their code was complex, but because everything ran sequentially in their pipeline. When linting, unit tests, security scans, and builds all execute one after another, the pipeline becomes a bottleneck instead of a productivity booster. The fix was simple: 🔹 Split the pipeline into clear stages 🔹 Run jobs in parallel wherever possible 🔹 Fail fast and surface feedback early The result? Pipeline time dropped by more than 60%, and developers stopped dreading every push. CI/CD should accelerate delivery — not slow it down. https://lnkd.in/gTGJKzwZ #DevOps #CICD #Automation #SoftwareEngineering #DeveloperExperience #DevOpsCulture #Productivity #CloudEngineering

7

If your team still says “works on my machine”, it’s not a people problem — it’s a platform problem. Environment drift across dev, QA, staging, and prod: - Slows releases - Increases failures - Burns cloud money Consistency beats heroics. Modern teams win by: ✔️ Standardized environments ✔️ Automated provisioning ✔️ Zero manual configuration drift Stability is a feature.

1

Thinking about migrating off Datadog? It’s not just a tooling decision — it’s an architectural one. We’ve talked with a lot of teams who underestimated what a migration actually involves: • Cost models that don’t scale with usage • Alert noise that follows you no matter the platform • Dashboards that don’t reflect how teams actually debug We pulled together 8 key considerations to help teams migrate with fewer surprises and less rework — from planning and data parity to alerting and team adoption. If Datadog costs or complexity are starting to feel heavier than the value you’re getting, this guide might help you pressure-test your next move. Get the Guide: https://bit.ly/4qChtPe

34

Today we’re launching GPT-5 — our most capable model yet for businesses and developers. It’s the smartest coding model we’ve ever built — state-of-the-art on real-world software engineering benchmarks and cutting tool-calling errors in half compared to other frontier models. Early partners are already putting it to work in production in the OpenAPI API: - GitHub is bringing GPT-5 to Copilot, empowering millions of developers with stronger reasoning across large codebases. - Amgen is exploring how GPT-5’s reasoning can accelerate discovery and improve decision-making in complex research environments. - Cursor has made GPT-5 the default for all new users — it catches deeply-hidden bugs, runs multi-turn background agents to completion, and is now their daily driver for scoping, planning, and end-to-end builds. - Lovable is using GPT-5 to help millions of users generate advanced applications from a single prompt and debug complex issues in large projects, excelling at reasoning over entire codebases. - BBVA reports it handled strategic tasks in hours that would have taken weeks, with standout speed and accuracy — even in Spanish. These stories capture what excites me most: GPT-5 is not just more intelligent, but more reliable, steerable, and adaptable to the unique workflows of every customer. Can’t wait to see what the wider developer community builds next. https://openai.com/gpt-5/

217

9 commentaires

"Centralized quality — slogans, KPIs, executive decrees — can produce positive results, but it’s brittle. Decentralized quality — continuous feedback, distributed ownership, emergent standards — builds resilience." https://lnkd.in/etMjmWYH

3