matt g

@nurijanian Interesting, I will need to give it a try. Thx :)

@mat_gajda idk maybe superpowers weren’t sticky enough and I saw the Every guys iterate on CE and honestly the CE skills and subagents just kept triggering at the right time without asking

my favorite engineering skills for AI: - Compound Engineering: github.com/EveryInc/compo… - Ryan Singer's shaping skills: github.com/rjs/shaping-sk… - Matt Pocock's skills: github.com/mattpocock/ski… I switched from Superpowers to Compound Engineering as they perfected the plugin over time, and I'm pretty sure I still only use like 10% of it

I think we should start talking more about being supported by an llm rather than replaced. Take a look on this article nolanlawson.com/2026/05/25/usi…

#javascript → #python cheatsheet for async: Promise.all → gather setTimeout → sleep (seconds, not ms!) fire & forget → create_task top-level IIFE → run Wrote up the mental model shift here: mattgajda.com/blog/2026/05/2…

@stan_info Still lots of traffic is also made by regular people using ChatGPT daily. So it happens gradually

there are ~2M software engineers in the US there are maybe 40 million (and growing with vibe coders) in the world and it creates significant compute constrain on GPUs imagine when entire world moves into agentic AI we either need to get a lot more efficient intelligence, per byte or i don't know

Built a MacWhisper-style offline STT tool in ~1 hour with Codex (plan + Script Kit docs as context). Script Kit = UI+mic, whisper.cpp = local transcription mattgajda.com/blog/2026/01/0…

We are entering an era of reading code rather than writing it. I wonder if, in the long term, this will actually expand our attention spans beyond what they are now in a world dominated by Instagram, TikTok, and the like.

@encrypted Is it AGI?

wait... you did what?!

@andyhennie Same

@mat_gajda i love the fact that people think i'm rage-bating. never my intention 🤭 read my other replies.

I just created a sidebar Dock for MacOS, and it feels amazing to use. Every app has a vertical sidebar these days, why doesn't the OS itself have it?

@AutisticOvrflow @amuldotexe More work for developers soon 😄

@mat_gajda @amuldotexe Agreed :/

I do not know about you people but I think because of LLMs - every software I am writing is now more complex + tedious - everything is far more nuanced and detailed, because I know what is one-shot low-bar enough will not be differentiated enough - am doing far more complex PRDs & code than I did in previous years, and need to hold far more context in my head - am taking so many more judgements per day than I did previously

For decades, the hardest part of being a programmer was writing the software in the first place. In the age of agentic coding, that has flipped: the hardest part is reading code you didn’t write and deciding whether to trust it. And careful reading, with all the hidden assumptions and trade-offs that aren’t obvious until you’ve sat with them, can genuinely be harder than writing the thing from scratch. mattgajda.com/blog/2026/05/1…

@AjeyGore I remember a time when people said microservices were too big and that we should split them into nanoservices😅

You don't need microservices and queues from day one! you actually don't need them for before massively scaling, can you please make sure that you can run things in ONE SINGLE VM, you will save money, time and thank me later.

@AutisticOvrflow @amuldotexe Same, plus I think because of it people are even less aware while reviewing code

@mat_gajda @amuldotexe It became tedious, I even sometimes have to review 100% AI generated PRs for a client and it's always super low quality code imo

@steipete Do you use worktrees?

I'm late to the party, but cmux is great. github.com/manaflow-ai/cm… current split: codex mac app: knowledege work, learning, reading cmux + codex cli: coding

@fraxool Sounds like a dopamine loop, you finish one more things to feel this kind of happiness after finish something. addictive 😅

As a solo Shopify developer since 2020, people could easily assume that AI made my work lighter. Ironically, in some ways, it’s been the exact opposite for me. Over the past few months, I’ve started struggling with sleep for the first time in my life. My brain just doesn’t seem able to switch off anymore. Before the AI era, I could actually disconnect. I could travel, enjoy weekends, step away from work without constantly thinking about the next feature or project. Now everything moves so fast that I feel permanently “on”. Even on weekends, I end up reopening my laptop, launching Codex or Claude, and finishing “just one more thing”. I genuinely love building with AI and I’m excited by what’s happening right now, but the pace and constant possibilities have changed the way my brain works. Instead of working less, I feel like I’m thinking about work more than ever. And honestly, it’s exhausting sometimes. I’m more tired than before, my sleep is getting impacted, and I’m realizing how difficult it has become to truly disconnect. Curious to know if other solo devs or entrepreneurs are feeling the same lately.

Full write-up: mattgajda.com/blog/2026/05/2…

Even when the shell isn't real, the environment has to be. HOME, cwd, and the language flags aren't ornamental. They're the contract that makes the emulation usable. Skip them → you'll chase bugs that only reproduce on someone else's machine.

I was wiring up a bash-js sandbox in Next.js and hit a weird question: If bash-js is a JavaScript emulator (no real shell), why does it need HOME? new Bash({ cwd: WORKSPACE_PATH, env: { HOME: '/home/user' }, ... }) Turns out: a lot of reasons. 🧵