Jakub Dobry

Faster responses, fewer rate limits, and 10% off premium requests. We are launching auto, now available in Visual Studio Code! Auto picks the right model for the job in GitHub Copilot, chosen automatically based on capacity and performance. Great products these days are rarely powered by a single model. They rely on a system of models working together, each playing to its strengths. With auto, we are bringing that ensemble approach directly to customers.    Over time, auto will get smarter. It will dynamically scale between small and large models based on task complexity, and fold model selection into the development loop itself. 🔗 https://lnkd.in/gTHA-ZsR

707

19 komentářů

13 painful lessons I learned building fast React apps (number 9 is my favorite.) 1. "It works" is not the same as "it scales." 2. Clear naming > clever naming. Always 3. Reusability speeds up development time. 4. Don’t chase perfection → ship, test, improve 5. Always build on core concepts → not just code 6. You’re not done after building → test real usage 7. Global state isn’t always the answer → start local 8. UI logic and business logic don’t belong together 9. Component structure determines debugging speed 10. State management should be boring → not confusing 11. Data fetching needs its own strategy → not just useEffect 12. Use tools because they're performant → not just popular 13. Performance issues usually come from poor architecture, not React itself I’ll break down 7 of these in the 📌 comments. Repost this free value for your team or other devs ♻️ P.S. Which one (1-13) resonates with you the most?

222

111 komentářů

I've interviewed 1,500+ developers in 6 years.  85% failed on the SAME backend question. The question? "Explain how you'd scale a Node.js application to handle 10,000 concurrent users." Most answers I got: ❌ "Add more servers" ❌ "Use load balancers"  ❌ "Cache everything" What I was looking for: ✅ Event-driven architecture principles ✅ Database connection pooling strategies ✅ Horizontal vs vertical scaling trade-offs ✅ Memory leak prevention techniques The gap is MASSIVE. Frontend developers are getting replaced by AI tools like v0, Cursor, and GitHub Copilot. But backend architecture? That requires deep understanding that AI can't replicate (yet). If you're a full-stack developer without solid backend foundations, you're competing with 10,000+ other developers for the same junior roles. The solution isn't more React tutorials. It's becoming the developer who can architect systems that scale. 🎯 Want to know the exact backend skills that made developers stand out in my interviews? Comment "BACKEND" and I'll share my insider list of what actually matters in 2025. (Only sharing with the first 50 people who comment)

62

59 komentářů

Everybody is talking about MCP. No one's telling you how to create a server. EDIT: AgentPass Link https://lnkd.in/gucgBnKB Well, if you want to build an MCP server from scratch, here's the manual way according to the official documentation: 1. Set up a Python or TypeScript development environment. 2. Write code for a protocol handler to manage requests and responses. 3. Build custom connectors for each data source (APIs, databases, etc.). 4. Implement your own query processing logic. 5. Code a robust security and permissions layer for authentication and authorisation. OR you can save yourself hours by hopping onto AgentPass Your no-code MCP server builder. It took me five minutes with AgentPass to do the above, Just give it the API spec YAML and it automatically: ↳ 𝗖𝗼𝗻𝘃𝗲𝗿𝘁𝘀 𝗮𝗻𝘆 𝗢𝗽𝗲𝗻𝗔𝗣𝗜 𝘀𝗽𝗲𝗰 into a secure, production-ready MCP tool. ↳ 𝗛𝗮𝗻𝗱𝗹𝗲𝘀 𝗲𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝘀𝗲𝗰𝘂𝗿𝗶𝘁𝘆 with built-in auth and access control. ↳ 𝗗𝗲𝗽𝗹𝗼𝘆𝘀 𝘁𝗼 𝗮 𝗳𝘂𝗹𝗹𝘆 𝗺𝗮𝗻𝗮𝗴𝗲𝗱 𝗰𝗹𝗼𝘂𝗱 𝘀𝗲𝗿𝘃𝗲𝗿—no infrastructure required. ↳ 𝗚𝗶𝘃𝗲𝘀 𝘆𝗼𝘂 𝗳𝘂𝗹𝗹 𝘃𝗶𝘀𝗶𝗯𝗶𝗹𝗶𝘁𝘆 with detailed analytics and audit trails. ↳ 𝗣𝗿𝗼𝘃𝗶𝗱𝗲𝘀 𝗮 𝗽𝗼𝘄𝗲𝗿𝗳𝘂𝗹 𝗖𝗟𝗜 for easy, programmatic control. ...all in a matter of minutes (really). I’ve been trying it out for my automation use cases, And I’m impressed! AgentPass.ai is open for beta users. I got in for free by finding an Easter Egg on their website Want to try it for free? Comment “MCP Server God” and I’ll tell you how ♻️ Repost this to save someone hours

270

70 komentářů

When developers use "tech debt" to refer to any code that they didn't write (and some they did), how do you decide where to invest? I find this 2x2 quite helpful for framing the discussion. In many situations, the trouble with debt is not the principal, it's the interest. When it comes to technical debt, that's the increase in the cost of change that's created by the debt. And you're not really required to pay until you're working in that area. The axis for "ease of change" includes everything it takes to get a change into production. Is the code easy to reason about? Do you trust the tests? Is the pipeline fast? Anything that makes it harder is bad. However, if the system or component isn't undergoing change, it's not hampering your business as badly. Hence the need to incorporate "desired rate of change." In other words, if we could, how much work would we be doing in this area? If something is slowing you down or blocking you in an area of your system that presents critical opportunities to your business, it's probably time to get it cleaned up. Like any 2x2, it's pretty simple. But I assure you someone, somewhere, is arguing right now about how to improve a thing that's already good and pretty stable. PS. I used this chart in a workshop at LeadDev's LDX3 and it led to some great discussions. If you'll be at LeadDev in NYC or Berlin this year, come have a chat. The workshop will be repeated at both.

61

2 komentáře

🤔 Is distributed tracing worth its overhead in Node.js? Every trace you capture changes your application's behavior. Every context you propagate costs performance. But when do the insights outweigh the cost? Adding observability is the Heisenberg observer effect in action: "What we observe is not nature itself, but nature exposed to our method of questioning." 🔬 Your monitoring changes what you're monitoring. The question is: how much, and is it worth it? This Tuesday (August 13), we're hosting Bryan English to dive deep into what async context really enables, the real performance cost of distributed tracing, when observability overhead becomes a problem, and patterns that minimize the impact. Bryan has spent years building observability tools and knows exactly where async context shines and where it becomes a burden. From cross-service request flows to context propagation gotchas, he'll share hard-earned lessons from the trenches 💡 This discussion is perfect if you've ever struggled to trace requests across services, debugged mysterious memory leaks or spurious errors, wanted to understand traffic flow in high-volume apps, built observability tooling, or questioned if your tracing is worth the cost. This isn't theory, it's practical decision-making. When should you instrument? What's too much overhead? How do you balance visibility with performance? We'll tackle the real questions that impact your production systems 🌙 Observability doesn't have to be a black box decision between "all the traces" or "performance first." Join us to find the sweet spot for your applications 🚀 🗓️ Tuesday, August 13 🔗

57

1 komentář

Most people use vibe coding like this: - Write a bunch of code - Hope it works - Fix what's broken later - Repeat But that's like building a house: - Without blueprints - Without measurements - Hoping everything fits Here's the better way: 1. Start with tests Your tests are your blueprint. They guide every move. 2. Commit constantly Tiny changes. Micro commits. No big bang moments. 3. Pair up Two minds. One keyboard. Zero blind spots. 4. Follow the red-green cycle: - Write a failing test - Make it pass - Clean up Repeat forever. Why this works: - You always know it works - You catch issues early - You move faster (yes, really) - You sleep better at night The old way is dead. This is how pros build. Start today.

62

43 komentářů