Opus 4.7 has a unique combo of being quite smart but lazy AF at the same time. Given enough context, it can do wonders. But if smth is missing, I get a lot of "good old" ChatGPT 3 vibes with the hallucinations and false claims/assumptions. Opus won't try to find info on its own.

I've been traveling this week, and ended up in a rental apartment with unsecure WiFi (no password). So I used my phone as a hotspot. On Friday evening, the mobile network started to struggle and went down to 3G, so I had a unique opportunity to test the web of 2026 with 3G internet.

I've never experienced a UUID v4 collision myself, but folks in this HN thread shared their stories. A great reminder that "statistically almost impossible" means "might happen still" news.ycombinator.com/item?id=4806...

In case somebody missed this with all the AI thingy and world news. The 2nd edition of the best (imo) software engineering book has been recently released. Thanks for the gem, @martin.kleppmann.com

I really love Apple products as a user: iPhone, MacBook, iPad, OS, etc. But I really hate Apple products as a developer: XCode, Dev Account registration & app submission process, etc. How come these 2 things coexist?

The Return of the King blog.jquery.com/2026/01/17/j...

I've started working on a new product and decided to pick up a good old Java & Spring stack instead of fancy things. And you know what? It feels great to be back! ☕

It's a bummer that the session limit can be reached in the middle of the answer. Understandable from the cost perspective, but a bad UX nevertheless. A great example of a hard engineering task to solve

Bought a used Lenovo laptop and installed Fedora on it to test forq.sh, as it’s a Linux-first tool. It’s been a while since I touched non-server Linux. Surprised how fresh Fedora looks! Great to see that Linux is becoming more user-friendly for non-tech folks.

Software engineering is full of trade-offs. Take SQLite for example: only 1 writer can modify the DB at the same time. Coming from the Postgres world, this sounded like a no-go. But, hey, it was a walk in the park to implement the FIFO ordering in forq.sh due to that nature.

Forq - SQLite Queue Service

Forq - SQLite Queue Service. The missing middle between embedded libraries and enterprise solutions.

forq.sh

If your workflow contains builds for ARM architecture, prepare your patience 🧘

Claude strikes back 🤖

I ran out of monthly credits in Cursor and decided to give the "grok-code-fast-1" model a try, as it is free for now. As you can see, it didn't go well. Not sure about developers, but it can already replace the Lorem Ipsum thingy for sure 🙂

DHH has just announced that Campfire is open-source from now on. Pretty cool! See here: github.com/basecamp/onc...

When Claude convinces you to use a particular product despite the initial idea to go with the competitor, it's a good sign of nice docs and credibility. DaisyUI it is, then. Great job, @saadeghi.bsky.social and @daisyui.bsky.social contributors!

So, I've finally tried HTMX. Funnily enough, here is what excited me the most: If you leave your browser tab open for 42 years, it will stay the same. No refreshes, no crashes. The page is just stoically waiting for you to be back. Long-forgotten joy after a decade of SPAs!

I manually submitted talktolocals.ai to 31 directories, and here are the stats 2 months later 👇 We have a clear winner, so far: TAAFT. But 2 launches are still pending: BetaList and ProductHunt. Let's see who will win 🙃 I'll post new stats in a month, so stay tuned and don't miss out.

Reached my very first milestone with talktolocals.ai The next screenshot will be from the Polar dashboard. I hope 😄

The number of bots parsing GitHub is insane. Here is a screenshot from my newly created repo with 0 stars and followers. Do you see the similar stats for your repos?

Just noticed that my most used LLM instruction in the agent mode is "Let's only discuss this, do not modify the code" Helps with LLMs' eagerness.

Came across this post on Hacker News today. It's quite ironic to see how professional devs in Microsoft are forced to use the tool that is not good enough yet. But somebody up the ladder can report that AI creates 80% of PRs now. Check for yourself: www.reddit.com/r/Experience...

Digging through a box of forgotten relics, I stumbled upon a pre-Spotify era gem. Punk rock is today's soundtrack 🤘

When you vibecoded your tariffs

Have you ever heard of Brooks's law? "Adding manpower to a late software project makes it later." From 1975, but hasn't aged a day. From 1975, but hasn't been understood yet.

Check out the coolest dark theme on the web! 🔥 What are your top picks?

Google explains recursion with a hidden Easter egg

"Keep it simple" in a nutshell: instead of worrying about the complexity of scaling, they "outsourced" this challenge to Google for free. Ingenious! www.levels.fyi/blog/scalin...

How Levels.fyi scaled to millions of users with Google Sheets as a backend

Our philosophy to scaling is simple, avoid premature optimization

levels.fyi

I started my journey in tech with a role that was a hybrid of PM and QA: - figure out what needs to be done for the product - compile a list of requirements - send it to devs - test the feature they built With LLMs and Cursor, it feels like being back to where it started 🙂

In 2025, choosing a programming language for the project has a new dimension: AI awareness of it. And since LLM quality relies on the amount of data it was trained on, it leads to surprising results: Java > Kotlin JS/TS > Swift C > Zig Python > Rust Interesting, isn’t it?

Coming from (relatively) big tech, I always think about scalability when building something. Would it be able to handle a million users? The truth is that most indie apps end up having 0 users. And tech is not the reason for that. A reminder to myself: think, but not overthink.