VirtusLab

For agents that make it out of the swamp 🕯️ 21:13. Your AI agent gets one job. First, the data swamp: six tables look the same. Which one is right? Is the agent even allowed to use it? Then the pipeline: code written in two minutes waits days for CI, review, security and release. The problem isn’t the agent. It’s everything around it. Visdom is the missing layer between AI coding and production: the right context, a risk check on every change before CI, and safe delivery. See it live in New York: Mon, Nov 2, 6-9 PM, 307 W 38th St. Live demos & Q&A. Register: (hubs.ly/Q04CYfC-0) Thank you, #Visdom!

Groundhog Night: a developer horror story 🎃 Change one line of code. Wait 12 minutes for tests. Repeat. All day. For a team of 200 engineers making 20 changes a day, that’s 800 hours of waiting, or 100 workdays lost every single day. You’re paying for that time, and getting nothing back. The fix is simple: reuse work that’s already done and test only what changed. The result in VirtusLab client projects: 4x faster builds and 3.3x lower CI costs. Your team ships faster, and you spend less doing it. Stuck in the loop? Let’s check → hubs.ly/Q04CTY4z0 Happy Halloween! 👻 #DX

Looking for a Linux internals geek who writes great Python We’re building an LLM sandbox on Nix + low-level Linux. Remote (EU), B2B 140–185 PLN net/h. Ever traced a weird syscall to see what’s going on? Let’s talk 👇 hubs.ly/Q04CS2qF0 #EB #Linux #NixOS #Python #Hiring

Full room at the talk. Full room at the workshop. 🔥 Thank you, Austin! 🤠 Data agents you can trust + AI code that actually ships to production, live on stage at #AustinTechWeek. Next up: Happy Hour at Bull & Bowl. Come join us! hubs.ly/Q04CQvm10 #Visdom #VisdomTour

🚨 Your AI is giving you outdated Bazel advice. Bazel changes fast. AI tools learned from old answers, so they often suggest things that no longer work. You follow the advice, and your build fails. So we built the Bazel Knowledge Hub: ✅ Learn Bazel at your level, from basics to advanced ✅ Plug it into your AI agent so it gives up-to-date answers ✅ Based on 15+ years of experience and 100 projects 👉 Check out the Bazel Knowledge Hub! hubs.ly/Q04CNnFR0 #Bazel #AI #DevTools #DX

🇺🇸 Austin, it’s today. 🤠 Two sessions, one problem every AI team hits: agents that can’t trust their data, and code you can’t trust in production. We’re fixing both at Austin Tech Week, with real projects, not only slides. 📍 STATION Austin, 701 Brazos Street 🎤 11:00–11:45 AM | Floor 1 | Talk + demo From a Data Swamp to a Data Agent You Can Trust, by Grzegorz Król 🛠️ 1:00–3:00 PM | Floor 16 | Hands-on workshop From AI Coding to Production, by Artur Skowroński, with a live game build & debug 👾 Our stand is open all day. Come say hi to Tomasz Orzechowski and Michał Ogrodnik from VirtusLab! 🤖 AI Track, sponsored by VirtusLab & Partly 🎳 Then join us for Happy Hour at Bull & Bowl, sponsored by VirtusLab & Michael Best ⚠️ Seats are first come, first served. Come early! 👉 hubs.ly/Q04CMfxs0 #AustinTechWeek #VisdomTour #Visdom

Looking for a Linux internals geek who writes great Python. We’re building an LLM sandbox on Nix + low-level Linux. Remote (EU), B2B 140–185 PLN/h. Ever traced a weird syscall to see what’s going on? Let’s talk 👇 hubs.ly/Q04CKj7M0 #EB #Linux #NixOS #Python #Hiring

🚀 12 months of AI-powered delivery. 12 hard-won lessons. One evening in Zurich. 🇨🇭 AI agents now read issues, pull dependencies, run tests and push code. The question is no longer whether to use them. It's how to restructure delivery, secure the pipeline and make your data trustworthy enough for agents to act on. On 10 December, 4 practitioners who've already done it share what worked and what didn't: 🔹 @Artur Skowroński (Head of Application Development, VirtusLab): AI-Native SDLC in 12 Lessons: 2026, Month by Month 🔹 @Fulco Houkes (former VP of Engineering, Bottomline Technologies): Securing the SDLC by Design and Architecture 🔹 @Grzegorz Król (Data Architect / Tech Lead, VirtusLab): From a Data Swamp to a Data Agent You Can Trust 🔹 @Jérémie Geus (Head of Software Delivery, Medgate): AI at Medgate in Software Delivery After the talks: networking and time to swap notes with peers facing the same challenges. 👥 Who it's for: CTOs, VPs of Engineering and tech leaders figuring out what AI-assisted development means for their teams, delivery pipelines and risk profile. 📍 Headsquarter – The Artistic, Zürich 📅 Thu, 10 Dec · 4:00-8:00 PM Seats are limited and registration requires approval, so save your spot early 👇 hubs.ly/Q04CG0RJ0 #AI #SDLC #Visdom

A maintainer approved the PR. Within a month, users reported two bugs it introduced. Radosław Kłos tested a “council” of AI reviewers on the same PR: several agents, each with one narrow job. They saw only what was available on merge day. In all three runs, they flagged both bugs. What made it work: → Each agent gets one job, not a personality. → Agents can say “nothing found.” Otherwise they invent issues. → Every serious finding gets challenged before anyone sees it. The cost: ~7x the tokens of a single agent. It’s not worth it for a config tweak, but it may be for auth, migrations, or shared contracts. Running this on one branch is manageable. Running it on every PR for a whole team is harder, and that’s what we’re building with #Visdom Code Review. Full setup and experiment in Radek’s article 👇 hubs.ly/Q04CCt4R0 #CodeReview #AI #Visdom #DX

For nights that end ghost-free. 👻 23:47. Your AI coding agent suggests a package to install. But that package doesn’t exist. The AI made it up. Attackers watch for these made-up names. They register them first, hide malicious code inside, and wait for someone to install. It’s called slopsquatting. This time, Visdom checks every new package before anything gets merged. The fake one gets blocked. Midnight. Your code is still safe. Thank you, #Visdom. 🎃 Happy Halloween!

Learn how investment banks can scale autonomous AI development without creating security vulnerabilities, runaway token costs, or code review bottlenecks - while ensuring full compliance with stringent financial regulations like NIS2 and DORA. Read the article: hubs.ly/Q04Crl_90 Don't hesitate to follow us for more news, videos, and more. #finance #banking #sdlc #ai

Every CI worker took ~15 minutes just to start. And one slow shard could hold up the whole job. For a global tech company’s main monorepo, VirtusLab is moving CI workloads to Bazel Remote Execution on EngFlow. The migration is still underway, but the results are already here: ✅ 20% lower average cost of main PR tests ✅ $3M+ in projected yearly savings ✅ 3× faster post-merge compilation ✅ Tests excluded from RBE: from ~11,000 to under 3,000 ✅ 100% of pull requests now run on RBE How we did it, and what made thousands of tests RBE-ready, is in the full case study 👇 hubs.ly/Q04CcPvl0 #Bazel #RemoteExecution #DX

“Virtual threads are good for I/O.” That’s true, but not always. On JDK 25, a simple file read can still block the carrier thread. 🤯 Michał Zyga tested how virtual threads behave in practice. Here’s what he found: 1️⃣ CPU-bound work holds the carrier thread. Virtual threads don’t make computation faster. 2️⃣ http://Thread.sleep() and other parking operations let the virtual thread unmount. This is where virtual threads shine. 3️⃣ File I/O: on JDK 21, ForkJoinPool added a compensating carrier thread. On JDK 25, it doesn’t, so the read just blocks. 4️⃣ Mixing CPU and I/O tasks can starve the I/O tasks. Takeaway: “I/O” isn’t one thing. Check whether your operation actually unmounts the virtual thread, and keep CPU-heavy work separate. 👉 Full article with code and results: hubs.ly/Q04C7byP0

26M lines of code. Navigation ready in 8.7 seconds. At Databricks, Metals v1 had to import the build before search or go-to-definition worked. In a monorepo with 285k Bazel JVM targets, that meant a long wait. Metals v2 rethinks that: a build-free index, compiler-backed pipelines for Scala and Java, and build integration kept off the startup path. It was developed together by engineers from Databricks and VirtusLab, and it’s open source under Apache 2.0. What Databricks reported from its production rollout: - 8.7s median to working navigation - 7ms median jump-to-definition - 2.4k engineers on one editor - 92% weekly-active IDE share for JVM work The lesson: at scale, don’t just make the slow step faster. Move it out of the way. 👉 The full story: hubs.ly/Q04C0Rtw0 #DX #Scala #DX

That's a wrap on @TechCrunch Disrupt 2026. 🚀 Huge shoutout to our team for absolutely crushing the Visdom demos. 👏 Nonstop conversations. A booth that never went quiet. And the same message from founders, investors and engineering leaders alike: they're ready to build differently. The interest in #Visdom our expectations, and it proved one thing: the future of software isn't written line by line. It's composed. Compose your AI-native software factory. ⚙️ 👉 Missed us at Disrupt? Book your Visdom demo: hubs.ly/Q04BpL180 #TechCrunchDisrupt #Visdom #AInative #AI #AInativesoftwarefactory

About 66,000 security vulnerabilities (CVEs) are expected this year. But the number attackers actually use? Almost the same as before. AI made finding bugs easy. That doesn’t mean every bug matters to you. A simple example: two apps use the same library with the same “high risk” vulnerability. App A reads data uploaded by users → fix it now. App B only reads its own built-in settings file → it can wait. The risk score is the same, but the real risk is completely different. Krzysztof Ciesielski explains how to tell which alerts really matter, how AI can help, and what to ask your security tool vendor 👇 hubs.ly/Q04BpxR_0 #Visdom