I time block access to scrolling social media apps to limit their distractions * 0000-1800 - No scrolling socials * 1800-0000 - Can access I currently use the Stayfocusd Chrome extensions to limit time on my computer and the built-in work mode on my Android phone.

Password managers help you remember strong passwords which prevents simple brute force / reuse hacks. But they also prevent phishing attacks. You'll be so used to autofill that you'll wonder why it's not working - hopefully pausing enough to realize it's not the real site / url.

The primary reason I'm moving off Coolify is to get Infrastructure as Code. If it works today, it should work next week / month / year as well. hamy.xyz/blog/2025-08...

Interesting data: How people use ChatGPT. I personally use AI most for: - Information search - How tos / debugging - Coding references / generation Sometimes images.

After using F# for awhile, raw null / undefined handling feels icky - especially if the nulls are implicit. Option = Some | None feels much better and forces you to handle each case. Reason #6 F# Sucks: Nulls haunt you hamy.xyz/blog/2025-06... #fsharp

Why I'm Moving my Personal Notes from Notion to Obsidian hamy.xyz/blog/2025-09...

AI is like a power tool. * Wielded strategically - large efficiency improvements * Wielded poorly - Might lose a hand To get the most use out of AI, take an active role. Don't vibe code, power code.

I generally like formatting my PR titles like this: [WORKSTREAM][TASKID] TYPE: CHANGE An example might be: [BLOG][HAM-101] Feature: Provide an example of good PR titles How to write good PR titles - hamy.xyz/blog/2025-08...

How to Setup an Ergonomic Workstation - More productivity, less back / neck / wrist pain hamy.xyz/blog/2025-09...

Will removing Python's GIL make it substantially faster? For well-designed parallel workloads: yes. For most general purpose single-threaded workloads: No and in many cases these get SLOWER.

My office goals: * Simple - To setup, change, and replace * Productive - Helps me achieve my goals * Ergonomic - So I can do this long-term My Multi-Computer WFH Productivity Desk Setup as a Software Engineer and YouTuber - hamy.xyz/blog/2025-05...

A simple TypeScript Result type:

The simplest method I've found to brand types in TypeScript is: type Brand<K, T> = K & { __brand: T } This lets you create a brand with simple syntax: type ValidDuration = Brand<Duration, "ValidDuration"> hamy.xyz/blog/2025-05... #typescript

Review your AI's Code - A Simple Process for Building More Robust Systems Faster with AI hamy.xyz/blog/2025-09...

As I've gotten older, my eyes have started rejecting the high contrast, super dark color themes I used to love. - halation - eye strain - headaches Now I find mid-contrast shades of grey to be more comfortable.

IMO Hack is a D Tier language. It's actually not all that bad as a language itself but with basically zero ecosystem outside of Meta I see very little reason for anyone to pick it up. If Hack seems to fit your needs - PHP is almost definitely a better option. #hacklang #php

Because the type system can't tell you what exceptions a function can throw, it's easy to leave cases unhandled. To handle this, exception-based systems often use blunt try / catch alls which lose out on granular control - potentially missing some cases or over catching in others.

I've now spent 3 days tweaking the colors of my Obsidian vault. Sometimes too much customization is a bad thing.

3 Technical Interview Types I saw in my 2025 Software Engineer job search: * LeetCode - Still used frequently, mostly mediums * System Design - Mostly popular questions or related to the company * "Real World" Coding - Build or integrate with an API or dataset hamy.xyz/blog/2025-04...

Throwing exceptions is 14x slower and uses at least 40% more memory than returning exceptions / results as values in C#. C# Errors vs Exceptions Performance Benchmarks - hamy.xyz/blog/2025-07...

What Are Stacked Commits and Why Should You Use Them? hamy.xyz/blog/2025-07...

If Go had more expressive types, it would be an S tier language.

AI unlocks so much personal software. We're going to have 10-100x more software in the world.

Some AI coding tools have built in checkpointing which is useful and often has good DX. But you should still use git for checkpointing code long term. It's the most used code version control and works with most tools.

Eventually AI will get so smart that it will start procrastinating on boring tasks, too.

I realized I had to do smth ab my social media usage when I found myself subconsciously opening IG from muscle memory whenever my mind would wander. I tried quitting outright but that didn't work. Instead I took a layered management approach - time, tasks, and bookmarks.

The biggest thing missing from C# is native unions. There are other things that it should have but that's the biggest blocker for me and how I like to structure logic.

Why I'm Ditching Coolify for Ansible for Deploying my Web Apps hamy.xyz/blog/2025-08...

I am a big fan of data-driven tests. * Easy to see what each test does, how it differs from others * Share test code to avoid accidental logic forking * Requires a LOT less test boilerplate

My Sleep Schedule as a Full-Time Software Engineer with a Newborn hamy.xyz/blog/2025-08...

Common terms in F# that I hadn't heard before coding with it: * Computation expressions * Railway-oriented programming * Partial application * Currying * Monads * Discriminated Unions * Exhaustive pattern matching Reason #7 F# Sucks - You start talking weird hamy.xyz/blog/2025-06... #fsharp

Successfully installed pgbouncer on my DB Server to help manage connections. Took awhile to configure correctly but now that it's setup with Ansible, should work ~deterministically for future deploys.

Common vibe coding failure cases: * Don't provide enough direction * Miss problems as you build * Don't understand your own system These often lead to additional time / effort spent debugging your system if not bricking it entirely. The fix: active power coding.

The cool thing ab my Ansible deploy script is I can now deploy my entire architecture to any servers I have access to - no vendor lockin. There's still a bit of manual effort to rehydrate backups but generally can just target new servers and deploy.

Took longer than expected but I now have an Ansible playbook that spins up: * 1 DB server - Postgres, Redis * 1 Analytics server - Grafana, Prometheus, Loki, Umami * N App Servers - with all my apps behind Traeffik Seeds each with monitoring agents to send to analytics server.

"I should host my apps on my own servers. Won't be too hard, I'll save a lot of money, and won't have vendor lockin. Plus deploying a new app / server will just require a few config changes!" 5 days and 2400 lines of YAML later: "Might've been faster to just use the GUI"

What is Python's GIL? The Global Interpreter Lock ensures only one thread controls the interpreter at a time. It allows safe memory management across threads for Python's reference counting GC This works well for single-threaded code but is a huge bottleneck for multi-threaded

Tailscale is pretty awesome. Easy to setup and now I can do `my-server:5000` in my browser and hit apps running on my server, without risk of it being open to the public.

My understanding of the main difference between popular Linux distros:

CinderBlockHtml isn't the fastest C# HTML generation paradigm but it's very competitive while offering (IMO) a simpler interface. hamy.xyz/blog/2025-07... #csharp

How I'm building software with AI agents: * Project RFC * Task description markdown * AI creates task tracker markdown Iterate through task: * Provide task description and tracker * AI codes * I review + change * Commit and push

While tactical AI best practices for building software will likely shift in the coming months, I think strategic best practices will largely stay the same: * Write good docs * Plan projects * Split into atomic tasks * Review and verify each change * Refactor when useful

Most of my technology posts are evergreen. New versions come out each year, but the info is valid for years. AI is interesting because it evolves so fast that the tactical best practices today will likely be out of date in the next 6 months, certainly the next few years.

My Recording Studio is attached directly to my desk. * Camera * Mic on a boom arm * Key Light This makes it super easy to start recording - no setup / teardown required. My Multi-Computer WFH Productivity Desk Setup as a Software Engineer and YouTuber - hamy.xyz/blog/2025-05...

What is a Result type? Results are ergonomic, universal patterns for representing a type that can be either a Success or Failure. `Result<Success, Failure>` says: * On Success - get Success payload * On Failure - get Failure payload

A code checkpoint is like a save in a game. A project checkpoint is like a quest tracker. They are both useful for staying on track and saving progress with AI agents in their own ways.

TypeScript Branded types are useful for differentiating two types that are structurally similar: * ID types (string, Guid, etc) * raw vs validated payloads (parse, don't validate) hamy.xyz/blog/2025-05... #typescript

Power went out again and my backup battery kicked on fast enough for my desktop to not power off. Pulling ab 150w for workstation and charging accessories so should last 10+ hours.

Two kinds of checkpoints I find useful when building software with AI: * Code checkpoints - git or similar version control * Project checkpoints - A project plan (for humans) and a project tracker (for AI) Code checkpoints save progress, project checkpoints keep you on track.

I think Golang is a C Tier language. I understand why people like its simplicity. But I think it aims for simplicity at micro scale at the expense of simplicity at macro scale. IMO macro is where complexity really rears its head so it's optimizing for the wrong thing. #golang