Luca Maraschi

TanStack Start, AI, and Performance: Rethinking Full-Stack Architecture (with Tanner Linsley) x.com/i/broadcasts/1…

Most AI agents don’t fail at the demo. They fail right after. State disappears. Scaling breaks. Orchestration gets messy. @platformatic built Regina to fix that. Stateful AI agents, built for production on Watt. 👇

Second Episode is out now! In the latest episode of 𝐓𝐡𝐞 𝐆𝐚𝐦𝐞 𝐨𝐟 𝐂𝐚𝐩𝐢𝐭𝐚𝐥, @lucamaraschi and Sam Pirzadeh sit down with @Rezaarbabian for one of the most unfiltered founder conversations we've had. Don't miss this one! youtu.be/DqWGSX8LyEM

@rauchg @dubeyansh_ @Railway @vercel @vercel's workflows have by far the best DX - if you are looking to run them in your K8s check @platformatic blog.platformatic.dev/durable-workfl…

@dubeyansh_ @Railway @vercel Workflows make the sloppy and unreliable, reliable. It's my default place to go for long-running execution. Steps can be up to 13 minutes right now, and have built-in observability, retrying, etc. It's also OSS and backend agnostic: useworkflow.dev

Been running long-lived backends on @Railway and loving it. But I want to go all-in on the @vercel ecosystem. What's the move for serverfull compute? Do Vercel Workflows solve this? With Vercel pushing into backend, the full-stack DX dream feels real. @rauchg 👀

Vercel CEO Guillermo Rauch signals IPO readiness as AI agents fuel revenue surge techcrunch.com/2026/04/13/ver…

We didn't want to lock you into Redis just to run background jobs reliably. @platformatic/job-queue ships with pluggable backends. Redis, filesystem, in-memory, and TypeScript-native support out of the box. Your job queue should fit your infrastructure. Not define it. Thoughts? 👇 blog.platformatic.dev/job-queue-reli…

@Zackary_Chapple @terminaldotshop Soon available also as Nespresso pods?;-)

Just finished another bag of @terminaldotshop dark roast

🎙️ New Banter Episode (a special one) coming soon! I sit down with @matteocollina and @tannerlinsley (founder of @tan_stack ) to talk about something that actually matters in production: SSR performance at scale. Not synthetic benchmarks. Not hello-world apps. Real workloads, real tradeoffs, real bottlenecks. We dig into: ☑️ Why SSR performance is often misunderstood ☑️ Where frameworks behave differently under load ☑️ The gap between average latency and tail latency ☑️ What actually breaks at 1,000 req/s ☑️ And how architectures like @platformatic Watt change the equation If you’re running SSR in production or planning to, this is the conversation you don’t want to miss. 🔗 Register: streamyard.com/watch/MBisytb7…

Skills, Agents & the Setup That Actually Works for AI-assisted JS Development x.com/i/broadcasts/1…

The payoff: ✅ Easier debugging ✅ Fewer moving parts ✅ No glue code Supports AVRO, Protobuf, JSON Schema out of the box. Design principle: keep async out of the hot path. Always. blog.platformatic.dev/platformatic-k…

The fix: Move async work out of the hot path. With @platformatic/kafka: ✅ Schemas resolved upfront ✅ Caches warmed before processing ✅Encode/decode stays 100% sync 👉 Predictable latency 👉 Stable throughput

Stop putting schema registry logic in the hot path. It’s one of the most common Kafka mistakes. Async schema resolution during serialization = unpredictable latency. @platformatic fixed this in @platformatic/kafka. blog.platformatic.dev/platformatic-k…

In production, failure is inevitable. Losing jobs shouldn't be. @platformatic built stalled job recovery and deduplication into @platformatic/job-queue from day one, because resilience can't be an afterthought. Does your current job queue handle crashes automatically? 👇 blog.platformatic.dev/job-queue-reli…

🎙️ New Banter Episode! 🗓️ Wednesday, April 8th - 8:00 AM (PT) 🔗 Register: streamyard.com/watch/Sy4qrdBF… In this episode of the @platformatic "The Node (and more) Banter", @matteocollina and I will dive into the emerging AI toolchain for Node.js development, from Agent Skills to coding agents like OpenClaw and Pi. 🚨 𝐒𝐩𝐨𝐢𝐥𝐞𝐫: @matteocollina will share his own personal skills repository (mcollina/skills), a collection of battle-tested Node.js, Fastify, and TypeScript best practices that any AI coding assistant can load and apply out of the box. Don't miss this episode!

@taishik_ I don’t see @platformatic there ;-)

Vancouver AI Startup Scene

"Fire and forget" shouldn't mean "fire and pray" 🔥 @platformatic built @platformatic/job-queue to give Node.js teams reliable background processing, retries, graceful shutdown, and pluggable storage, out of the box. What's your biggest frustration with job queues today? 👇 blog.platformatic.dev/job-queue-reli…

Bringing @workflow to your Kubernetes x.com/i/broadcasts/1…

⚠️ Durable workflows in Node.js look easy… until they silently corrupt data. On Kubernetes, version mismatches can replay wrong steps, mix results, and break your system. 🎧 @matteocollina and I, explain how to run version-safe workflows in production with @platformatic World. Register 👉 streamyard.com/watch/W2pmNaq2…

@kettanaito came to us with a question... @paolo_insogna and @matteocollina jumped in. What started as a quick answer turned into 50 minutes of going deep into Node.js internals. We covered: ✅ The full history of HTTP parsing in Node.js. From the old unmaintainable http_parser to llhttp ✅ How llhttp is actually built (𝘵𝘩𝘦 𝘴𝘰𝘶𝘳𝘤𝘦 𝘪𝘴 𝘛𝘺𝘱𝘦𝘚𝘤𝘳𝘪𝘱𝘵, 𝘤𝘰𝘮𝘱𝘪𝘭𝘦𝘥 𝘵𝘰 𝘊 𝘷𝘪𝘢 𝘓𝘓𝘝𝘔 — 𝘺𝘦𝘴, 𝘳𝘦𝘢𝘭𝘭𝘺) ✅ Parser pooling, the free list architecture, and why it's tied to the garbage collector ✅ The C++ binding layer, why crossing the JS↔C++ boundary is expensive, and how Undici switched to WebAssembly to make it faster ✅ How Artem is tapping into TLS/TCP wraps to intercept HTTP traffic at the socket level This is the kind of conversation we love at @platformatic. Answering hard questions about things most people take for granted, and two Node.js TSC members who spent quite some time of their lives in this codebase, trying to explain it. youtube.com/watch?v=DhRJvQ…