Viktor Bezdek

This has nothing to do with my work, but I’m still going to share it here. About 10–15 years ago, music was a big part of my life. I spent nights creating electronic tracks with zero ambition beyond the act itself. No releases, no audience, no plan. Just making music because it felt necessary. Then priorities shifted: career, responsibilities, life. Unfinished tracks were exported, backed up, and forgotten for a decade. Last spring, while cleaning old boxes, I found an external drive. Inside: those sessions, half-finished ideas from a different version of myself. I almost closed the folder. Instead, I pressed play. Curiosity turned into more. Finishing tracks helped decompress from work. Late evenings or early mornings became spaces with no expectations, no outcomes to optimize, no one to impress—like meditation, but different. Just closing a long-open loop. That’s how a long-abandoned project became a finished album. Today, it exists virtually and physically: on streaming platforms like Apple Music, Spotify, and others, as well as pressed on vinyl. Something started over a decade ago, disappeared for ten years, and returned unexpectedly. I’m sharing because many know me only professionally. This is another side: unfinished things, delayed gratification, and the value of not letting old ideas die because life got busy. If curious, I’ll share links in comments. If you have an old hard drive with unfinished work on it, maybe it’s worth opening that box. Vinyl: elasticstage.com/soundcloud/rel… Apple Music: music.apple.com/us/album/singu… Spotify: open.spotify.com/album/0IlX35vz… Deezer: deezer.com/us/album/88212… #music #vinyl #vinylrelease #electronicmusic #independentartist #creativity #sideproject #longgame #unfinishedbusiness #creativeprocess #fromthearchives #rediscovery #passionproject #lifebeyondwork #closingloops

@NoCodeWithAI @claudeai It does

@claudeai Bro still does not pass the car wash test Lol !!!

Introducing Claude Opus 4.7, our most capable Opus model yet. It handles long-running tasks with more rigor, follows instructions more precisely, and verifies its own outputs before reporting back. You can hand off your hardest work with less supervision.

@borekb I don’t understand why? Limited potential for contributors or you hate performance?

možná nepopulární názor, ale za mě je to technologický red flag

@borekb Není to těmi emoji? IMO na fix je třeba použít neproporcionální font a nějakou z odpovídajících šířek mezery * emoji. Případně použít monospaced font, který podobne mezery podporuje, nebo umí emoji ve správném rozměru. Imo to není easy fix :)

Cca před týdnem jsem hledal něco, co po agentech zarovná jejich svislítka, a nic pořádného neexistovalo. Ten problém je reálný 😄.

Still amazed what can be shot on night sky from my back yard

@ArthurFieldbury That’s a good read. Worth your time. Cheers!

@viktorbezdek Nice collection. Bookmarked it. Reading "A Mental Model for Claude Code" now. Thanks.

Here’s my continuously growing collection of awesome open source AI agent plugins, skills, subagents, connectors, orchestrators, and collections for engineering, product, entrepreneurs, and any general knowledge worker that wants to do more with less. Works with most CLI agents, Cursor, Claude, Cowork github.com/viktorbezdek/a…

@sukh_saroy AI

Everyone is hyped about Claude Cowork… but barely anyone knows how to actually use it to replace real work. I collected 300+ mega prompts that turn Claude Cowork into a full-blown productivity engine. Comment "AI" and I’ll DM you everything.

@steipete @OpenAI @openclaw Congrats 🤘🏻

I'm joining @OpenAI to bring agents to everyone. @OpenClaw is becoming a foundation: open, independent, and just getting started.🦞 steipete.me/posts/2026/ope…

The aesthetic holds up, the codebase is cleaner than most community projects have any right to be, and the contributor docs actually make you want to open a PR :) Kudos on that. Also hit me closer to home than expected. I have my own half-built thing at promptarena.dev that never quite got there. You know how it goes. Seeing someone push through that phase and ship something people actually use? That’s not nothing. If I had one thing to flag: new users land on the homepage and get… three buttons. Beautiful buttons, but still. One gentle “start here” for total beginners would close that gap without touching the vibe. But it gave me an idea what would be good addition to uxatlas.io. Some guidelines for dos and donts for various kinds of projects. Thanks for the inspiration. Keep building. Great project 👍🏻

Curated a list of standout GitHub projects. Personally vetted, high-signal, no filler. If you want quality repos without sifting through noise, this is the shortcut. github.com/viktorbezdek/a…

UXAtlas has 93 research methods, with clear, actionable guides. No login or paywall. Pick the right UX approach fast, skip trial-and-error, focus on research that actually moves the needle. uxatlas.io

The counterintuitive scaling playbook nobody wants to hear: Cut 80% of the work. Wait years for top 1% talent. Let small teams radically outperform large ones. Doing less unlocks exponential growth because focus and discipline beat activity theater every time. The hardest part isn't the execution. It's having the courage to say no to almost everything and the patience to wait for the right people instead of filling seats with warm bodies. Most orgs optimize for looking busy. The winning move is optimizing for impact per unit of complexity.

The thing that clicked for me: don't show users how sure the model is. Show them what to do given how sure it is. Behavioral tiers beat raw confidence scores. Calibrated hedging beats false authority. And sometimes the right move is hiding uncertainty entirely — when the stakes are low and the user gets instant feedback anyway.

Most AI interfaces make the same mistake: they present guesses with the visual authority of facts. Your model is 61% confident. The user sees a clean paragraph that reads like truth. Nobody flags the gap. I wrote about what actually works when your outputs are probabilistic ↓ uxatlas.io/articles/desig…

@bdmorgan @geminicli @googlecloud Nice to meet you 🤝

I think it's time I start speaking up a bit more on this platform. I lead engineering efforts for @geminicli and Gemini Code Assist at @googlecloud . If I promise to post pithy thoughts and opinions a bit more, let's see if I can get a few followers... Help me out @ntaylormullen , @JackWoth98 , @LyalinDotCom , @SriThreePO, @geminicli

@_Sironyeka Hello

@sherifgjini Stuck on 350, but unfortunately majority unverified. Good job. You are clearly doing something correctly.

Just crossed 1,200 followers! 🥳 A week ago I was stuck at 900. Yesterday morning I hit 1,000, then gained 163 followers in just 10 hours! Thank you all for following along 🙏

The fastest teams I've seen aren't heroic. They're relentlessly boring: small batches, fast feedback, clear priorities. Everyone else is chasing the latest framework or some "revolutionary" process. The teams that actually ship? Heads down trimming scope, tracking what matters, saying no to almost everything. Not glamorous. Just effective.

Hey @claudeai, tell me why I hit the token quota with Claude Code in ~90 mins with Opus 4.6 on Max 20x for nearly 200 EUR, and never hit limits with @OpenAI Codex 5.3 Max thinking on a 20 EUR subscription. Results are equally good. I really like Claude models, but the cost/benefit ratio stops leaning in your favor.