Jonathan Perichon

🚀 Two days before WhatTheStack kicks off, I’m running a crash course in building apps that survive the real world. Real-World React: The Architectural Playbook for Performance, Resilience & Scalability is built around React and Next.js as our shared language, but it’s not exclusive. I’ve had Vue, Svelte, and Angular devs attend, and most of the concepts we’ll cover are framework-agnostic and apply to any modern frontend stack. We’ll dig into: - Architecture patterns that implicitly boost performance - The real story behind React reconciliation (and anti-patterns to avoid) - Resilience engineering for modern frontends - DORA metrics and delivery velocity at scale - Building post-deployment confidence so you can ship without fear This isn’t about pixel-perfect UIs, it’s about patterns, decisions, and trade-offs that keep apps fast, resilient, and scalable when they’re serving millions, not just running locally. 🎯 If you work in frontend and want to level up your architecture game, this is for you.

5

After Claude released Legal and Tax related skills, the software/SaaS companies (The Companies) stocks experiences a cliff jump in the past months with a strong "AI fear" that the business model could be disruptive by AI. It's not the first time we see that as the coding has been significantly changed since Claude Code / Codex and Cursor/Windsurf released. What exactly changed? What will be "unchanged" in this fast changing world? What can the company do? What exactly changed? AI lowers the barrier of knowledge/experience based moat, leading the PE ratio drops for The Companies. The Companies always hold a high PE (40~50X) compared with normal company (10~20x) due to fast growth and deep moat on complicated logic through accumulated competitive advantage and sunk cost for new competitor to enter the market. The key moat is business logic itself. But AI makes the knowledge easier to available and copy a similar function faster. The Companies may not be total abandoned but PEs won't be that high with more competition. What will be "unchanged" in this fast changing world? Ground Truth is unchanged part in the fast changing world. AI has the knowledge in the past to build "memory"/knowledge, and the "thinking logic" to consume latest data and generate result. It means it depends on the ground truth to perform well, otherwise it's "hallucination". The company with the ground truth (owned the data, not just store the customer data) will build new moat. What can the company do? The Companies needs to shift the logic complication to logic flexibility with AI and focus on AI output stability with minimized hallucination. The heavy logic and customer experience will not a blocker for customer to shift the "software", and they would be attracted by the flexibility to optimize processes meeting business needs faster. In the meantime, LLM based AI is still a statistical model which will have hallucination in a tiny number of randomized but disaster cases. A guardrail is critical to allow AI makes mistakes but don't caused disaster business impact. The same logic also fits to individual professional person - the training and knowledge to solve the problem is learned by AI. Our job could shift to identify and define the problem, fast learning to choose the right tool and audit the AI output. #AI #SaaSCompany #SoftwareCompany #AIChangeJobs

9

Startups are hard. No one is ever startup ready. Coming from a big tech the adjustment phase is much more demanding because you need to unlearn a bunch of stuff. Processes are not defined, it all works on trust. Trust that your team aligns on a singular goal and delivers time after time, no matter how tough it gets. There are sleepless nights and anxious moments at its peak. The problems are not well defined. That paves way for innovation. Working on problems that have never been solved before makes you give up at times. But when it works, all the frustration, the tiredness just cease to exist. Seeing your feature being used by thousands of users makes it all worth it. Since there's a singular goal, there's less to zero politics and bureaucracy and the focus lies in building stuff. There are conflicts, and oh so how many there are, but it just means you care to build the best. Startups filter out the people who don't fit, and what remains is a singular entity. People leave for many reasons — timing, life, priorities. But you buckle down and build from ground up, again and again. Slowly but surely, the processes are built and it takes effort for people to follow them. To move fast, you break things, you come under the scanner and that's when it teaches you accountability. When you build a no-blame culture, the mistakes are not just yours but the org's as a whole. Working at a startup, you're not restricted by biases. For you, OpenAI or Google are both the same and you pick the tools that help you move faster and efficiently. In a startup of 30 people, you'll see people using 5 different tools, all of them the best ones and whatever gets the job done. People are not SMEs, they slowly become jack of all trades. There's is so much work and so much exposure that you see people standing out and taking on responsibilities that they have never taken before. It molds them differently and changes them to take on any challenge in open world. I'm proud to be a part of that culture and glad that I could contribute in building it further. As 2025 comes to an end and I reflect on the last two and a half years, I am lucky that I got the opportunity to work with superstars who give their everything. Because, startups are not for everyone. But if you can thrive at a startup, you can bring down any wall in the world.

50

2 Comments

New in /experimental mode in Copilot CLI: "Fleets" 🛸 Run `/fleet` to dispatch parallel subagents to implement your plan. The secret sauce here is a sqlite database per session that the agent uses to model dependency aware tasks and TODOs.

111

8 Comments

Great news for AsyncAPI Initiative and OpenAPI Initiative tech writers and developers! The Arazzo Specification is getting support in JetBrains IDEs! 🚀 After a quick survey of the ecosystem, I realized there was no existing plugin available. Fortunately, I already had a well-crafted foundation. And I’m now building a dedicated plugin specifically for Arazzo While the community is actively working on integrating AsyncAPI support into Arazzo (see: https://lnkd.in/egm4XFGC), I’ve decided to move ahead and release a JetBrains plugin tailored for Arazzo 🔧 Out of the box, the plugin will support: - File recognition - Real-time validation and inspection - Context-aware autocompletion The first release is coming in the upcoming few weeks. Stay tuned! #JetBrains #Arazzo #OpenAPI #AsyncAPI #IDEplugins #DeveloperTools

38

1 Comment

🔥 NUMA: The Hidden Latency Penalty in Multi-Socket Servers When we talk low-latency C++, we often obsess over branch predictors, inlining, or cache lines. But in multi-socket production servers, a bigger invisible cost lurks: NUMA (Non-Uniform Memory Access). 💡 What’s going on under the hood? • Each CPU socket has its own local memory. • Accessing memory on your own socket is fast. • Accessing memory on the other socket can be 2–3× slower, since it travels across the interconnect (Intel UPI, AMD Infinity Fabric, etc.). Now imagine a thread pinned to Socket 0, but allocating memory the OS placed on Socket 1’s NUMA node. 👉 Every access now pays a remote memory penalty — adding tens of nanoseconds per load. 👉 In a trading engine, that’s the difference between hitting the market first… or missing it entirely. ⸻ ✅ NUMA-Aware Best Practices: • Thread Affinity: Pin threads to cores deliberately. • NUMA-Aware Allocation: Use numactl, libnuma, or OS APIs to control where memory lives. • Partition Work: Keep data + compute on the same socket. • Socket Discipline: Some workloads isolate latency-critical tasks to a single socket to keep memory fully local. ⸻ 🔑 Key Takeaways • NUMA effects don’t show up on dev laptops — but they do on real multi-socket servers. • Cross-socket memory access can silently add tens of nanoseconds. • Profiling and locality awareness are essential to keep latency deterministic. 👉 Have you profiled your system for NUMA penalties? What’s your strategy — pinning, binding, or redesigning the workload? #Cplusplus #LowLatency #HFT #NUMA #PerformanceEngineering #SystemDesign #TradingSystems

68

2 Comments