Manish Kumar

OVERRATED: running tons of agents in parallel; working on too many things at once; perpetual context-switching; opening lots of low-quality PRs that may never land. UNDERRATED: using one or two agents at a time; focusing on the task in front of you; thinking deeply; finishing stuff; making your code works in prod.

1/9 Software Engineers in Ontario stuck on the "no Canadian experience" wall, listen up. There's a weekly meetup in Toronto that quietly turns newcomers into hired engineers. It's free. It runs every Tuesday. And most immigrants have never heard of it. Let me tell you about Civic Tech Toronto. 🧵👇

It is hard to improve as a software engineer when your main coping strategy is distraction. I have seen a lot of engineers (including me) who say they want to get better at systems, debugging, cloud, observability, performance. But the moment things get a bit uncomfortable, they escape. Open YouTube. Jump to another course. Switch from Go to Rust to Python to Kubernetes. Read 20 threads. Watch 3 conference talks. Do everything except sit with the actual confusion. That is the trap. Real improvement needs sustained contact with what you do not understand yet. Take observability for example. Most people would say they “know” logging, metrics and tracing. But ask them to debug a real production slowdown and things start falling apart. API latency spikes. CPU looks normal. Error rate is low. Customers are still complaining. Now what? This is where distraction usually begins. You start reading random blogs on OpenTelemetry, Grafana, Jaeger, Prometheus, distributed tracing. Looks productive. Feels productive. But you are still avoiding the real work. The better way is more uncomfortable. Pick one small service. Add structured logs properly. Add request IDs. Add latency metrics. Add error counters. Add one trace through the critical path. Now send traffic. Break one dependency. Add retry. Watch what changes in logs, metrics and traces. Now you are learning observability for real. It's not only about consuming content. It's is essentially about being in contact with the part that is still weak. You will notice things like: - logs tell you what happened - metrics tell you where the pattern is - traces tell you where time went That one hands-on exercise will teach more than 50 “complete observability roadmap” posts. Most people are stuck because they have trained themselves to escape discomfort too fast. As a software engineer, your growth starts the day you stop using distraction as fake progress and start sitting with the messy part long enough to understand it. Thanks for reading!

Human cognitive friction has long been acting as a regularizer for a lot of digital infrastructure. It made software APIs less terrible and codebases less complex. Now LLM disintermediation is causing this effect to fade, which in turn will cause runaway software complexity.

Save 45%, only on April 19th! Today, you can save on CUDA for Deep Learning by @elliotarledge and other related titles: hubs.la/Q04cF2Lh0 Written for the latest NVIDIA hardware, the book builds a deep understanding of CUDA (Compute Unified Device Architecture) fundamentals that will stay relevant as chips upgrade and evolve. CUDA delivers direct control, debugging power, and acceleration at the GPU level that can’t be matched by other types of optimizations.

A man grows up when he realizes that his parents are mortals, he is all he has, and that he was born alone and will die alone. In 100 years we will all be dead and forgotten. No one will know who you are in 2150, and we all know it's only a matter of time till 2150 arrives. Every possession you have, the house you live in, the money in your bank will be a possession of someone else who will not know who you are. Every time I see someone stressed about some meaningless thing, I am reminded of this: Remember to LIVE while you can.

ADHD tip: don’t fight your weird energy. If you feel like cleaning at 11pm, clean. If you feel like studying at 6am, study. Ride the wave instead of waiting for “the right time.”

an extremely high level strategy for getting unstuck is to commit to the stuckness try to be 10x more stuck. act out the persona of the perfectly stuck person. indulge the stuckness. savour it. enjoy every moment of it paradoxically, by leaning into stuckness, you create relaxation. you take all the energy that was going to shaming & beating yourself up and you free it. now that energy can go towards fun. when you’re having fun, things start moving I have given this strategy to a handful of clients and used it myself and it has never failed. the only way out is through

We gave the iOS simulator Chrome DevTools. Performance, networking, and logs from your app - in every test report. Slow launch? Failing request? Poor performance? Find out why!

my therapist told me this and it hit me: “healing is so hard because it is a constant battle between your inner child who is scared and just wants safety.... your inner teenager who is angry and just wants justice.... and your current self-who is tired and just wants peace."

Apple introduced the Accelerate framework many years ago, so it is available almost on every version of Apple platforms. It is a high-performance and energy-efficient way of doing computations using the vector-processing capabilities of the device. swiftwithmajid.com/2025/05/13/opt…

Swift’s concurrency system has serious smarts: 6 “awaits” are squashed down to just 3 jobs enqueued when the scheduler sees it can avoid a hop. We can verify by creating a custom executor that logs whenever the actor enqueues a job:

If you care at all about AI compute, you must read this.

Many folks avoid actors due to a proliferation of “awaits” that can introduce subtle race conditions as multiple threads can interleave and access the actor’s data. But did you know you can squash many “awaits” down to a single one, eliminating race conditions in the process?

❓How do I show placeholder text while streamed properties load in SwiftUI? 💡Make properties optional and use the nil-coalescing operator (??) to display default text until the streamed response populates the values. 👉Free #SwiftUI picture book at bigmtn.studio

❓Can I use if/else statements in SwiftUI prompts? 💡Yes! The Prompt builder has internal functions like buildEither that handle if/else conditions, making it perfect for screens with toggles and conditional logic. 👉Free #SwiftUI picture book at bigmtn.studio

If you use AI to build SwiftUI apps, this is for you. Visual-first SwiftUI reference for AI tools like Claude Code, Codex, and more. learnandcodewithenid.com/agent-kit

SwiftUI Tip 💡 Use projectionEffect() to create a skew effect on a view. → learnandcodewithenid.com

PyTorch is probably the most important framework used to train LLMs. You’re expected to know autograd, optimizers and nn.Modules inside out for any ML role. We’ve released a PyTorch sheet which takes you from basics to applying attention mechanism. tensortonic.com