Donald Cook

Pull Requests shouldn’t feel like this. We’ve all opened a large PR review and thought: “Well... this is going to hurt.” Sure, small PRs are great, but sometimes you need the full picture. That’s where stacked PRs help (been using Graphite for that, a tool for managing PRs and code reviews). Stacked PRs treat your feature development as a chain of commits / PRs layered on top of each other. Think of it like a Jenga tower: each block (PR) builds on top of the previous one. pr1 → main pr2 → pr1 pr3 → pr2 Now it’s getting even better. I just got early access to Graphite Chat 👇 - I can literally ask it “what’s going on in this PR?” and it gives me the full story, codebase history, past PRs, even outside context. - I can just highlight it and ask “explain this”, instant breakdown. - Flags risks before reviewers even show up. - When someone leaves feedback, it helps me make sense of it and suggests fixes. - And the best part? When CI blows up, I can just ask “why did this break?” and actually get an answer. I’ll even be able to pop open the Graphite Editor right inside Chat to view and edit the codebase. Reviews feel lighter, more collaborative. Exactly how they should.

1,572

141 Comments

See the fastest LLM serving engine, 100% open source!!! If you’re running multi-round QA, RAG, or any long-context workload, you know the pain: slow time-to-first-token and expensive bills. You should take a look LMCache, a next-gen LLM serving engine extension designed to slash TTFT (time to first token) by 3–10x and boost throughput. How? LMCache stores KV caches from any reusable text (not just prefixes) across GPU, CPU DRAM, and disk, then shares them across serving instances. That means fewer recomputations, faster responses, and huge GPU savings. Key highlights: 🔥 Seamless integration with vLLM v1 ⚡ High-performance CPU KV offloading 🔗 P2P KVCache sharing across servers 💾 Storage on CPU, Disk, or NIXL ✅ Supported in vLLM production stack, llm-d, and KServe 📦 One-liner install: pip install lmcache If you’re scaling LLM inference, you should check it! 👉 GitHub link in the comments. #LLM #AI #Inference #OpenSource #RAG #AgenticAI

379

27 Comments

5 ARIA attributes every dev should know (and 2 to avoid) Most developers know ARIA exists, but which attributes actually matter for everyday accessibility work? Before we get started, remember the first rule of ARIA: Don't use ARIA unless you have to. If semantic HTML can do the job, use that instead. But when you do need ARIA, here are 5 ARIA attributes that are actually useful: 1. aria-labelledby - Describe the current element with the text from another element on the page. Better than aria-label because it allows you to reference existing visible text on the page and keep it translatable. 2. aria-describedby - Connects help text, error messages, or extra context to form fields and other elements. Screen readers announce this after the main label. 3. aria-expanded - Essential for dropdowns, accordions, mobile menus - anything that opens and closes. Bonus: pair it with aria-controls to define which element it's controlling. 4. aria-current - Perfect for programatically defining the current page in navigation, active step in a process, or current item in pagination. Much clearer than visual styling alone. 5. aria-live - Announces dynamic updates to screen readers. Use "polite" for status messages, "assertive" for urgent alerts. Two to avoid (or use carefully and thoughtfully): 1. aria-label - It's typically a better choice to choose another way to label elements, with text inside a button tag, a label tag associated with a form field, or visible text associated with the aria-labelledby attribute. Treat aria-label as a last resort when no other method of labeling an element can be used. 2. role - Most of the time, there's no need for that role attribute when you use semantic HTML. Skip adding role="button" on a div - just use a button tag. A proper heading tag doesn't need a heading role or an aria-level attribute. The best accessibility starts with good semantic HTML. ARIA is intended to fill in the gaps, not replace the foundation. What ARIA attributes do you find most useful? Any that trip up your team?

52

14 Comments

New episode with Matt Palmer from Replit We get into Replit's crazy growth since the launch of Agent. And Matt shares useful tips on making videos for devs and writing great docs. Find the full episode on Spotify, YouTube, Apple Podcasts etc.

61

🤖 I love how the discipline of doing things right rarely gets appreciated - until it unlocks possibilities you didn't even know existed... Check out how I can now ask Claude about our team performance 👇 We've been tracking engineering metrics at loveholidays since day one - from team health and code quality to performance data, even building them into data products. It was an investment to ensure our environment and culture was as great to work in as it was performant. Recently, thanks to this investment and the brilliant work of Stuart Caborn, Misha Griffiths and George Malamidis, we were able to connect our data to Claude via MCP. Meaning we can now have conversations to explore six years of performance insights. We've all heard that engagement drives performance, but here is what our data actually shows: Teams with high health scores show: → 40% faster code reviews → 25% higher goal completion → 50% fewer escalated incidents It's a reminder that new technologies don't create value out of nothing. They amplify what you've already built. The invisible work we did for years - tracking data religiously, making it accessible, having organisational patience - that's what made this breakthrough possible. What foundational work is your organisation undervaluing today? #TechLeadership #EngineeringLeadership #AIIntegration

67

15 Comments

If you’re working through someone else’s code and feel completely lost… Welcome. That feeling never totally goes away. But here’s how I tackle it: 🔸 Start from the entry point 🔸 Trace the flow, one hop at a time 🔸 Don’t panic - debug with curiosity and intention Good engineers aren’t the ones who never feel lost. They’re the ones who keep moving anyway.

32

4 Comments

How do you choose what language to use? We’re building Lago with Ruby on Rails. It’s not trendy. Young grads think of Ruby like I think of COBOL. Here’s why we picked it anyway (and my framework for choosing the right language): -First, we knew Ruby best. With 10+ years of Ruby experience, it was the most practical choice. This is important: What enables YOUR team to ship well and quickly? -Second, we didn’t care what everyone else used. Just because everyone else uses Rust doesn’t mean Rust is the perfect choice for you. -Third, your language should match the product you’re building. Ruby is battle-tested for API products. -Fourth, you need talent. Choose a language that ensures you’ll find enough engineers to hire. For our product and team, Ruby checks all of those boxes. Today: Lago handles millions of API calls a day, we’ve upgraded across multiple Ruby/Rails versions, and yes—we’ve added Go and Rust where it makes sense. But if we were starting again in 2025? We’d still choose Rails.

35

11 Comments

You may see folks claiming that they can get "A.I." coding assistants to do exactly what they want by completely and precisely describing the software they require up-front, and hailing the return of Big Design Up-Front. Anyone who's spent more than a handful of hours using these tools will know that this simply isn't true.

103

116 Comments

Let's talk about something that's becoming way too common and completely unacceptable. Why are companies asking QA engineers to test their real products during interviews? Not a dummy app. Not a sandbox environment. Their actual live products. Let's call it what it is free labor disguised as an interview. As a QA professional, your skill set is valuable. You've spent years learning how to detect critical issues, improve user experience, and protect product integrity. That knowledge shouldn't be exploited under the pretext of "assessment." If a company truly wants to assess your ability, they should provide a test environment, dummy app, or simulation something designed for that purpose. Not their production site. Not their paying users. And if they insist otherwise? That's your cue to walk away. Your time, your mind, and your experience have value. Don't give them away freely. It's time QA professionals start setting boundaries. The moment we begin to treat our expertise as valuable, the industry will too. To every QA engineer reading this: V You're not overreacting. You're not being "difficult." You're protecting your craft. Because professionalism is not just about what you can do it's about what you refuse to do for free. #SoftwareTesting #QualityAssurance #QACommunity #Women|nTech

50

12 Comments