Activity Monitor - Window - GPU history is your friend

😂😂😂 I needed to fix my lawn tractor, so I watched ~10 YouTube tutorials on how to bead seat a tractor tire using flammable aerosol. Now my YT-recommendations are mostly monster trucks, rifles, MAGA and country music.

Funny how the biggest progress has been in vibeCODING. Not vibe law, not vibe logistics or vibe HRing — coding. Coders be like: “let’s automate our own extinction first”

Crushed a half marathon this Sunday - through relentless rain!! 🇱🇻🇱🇻

The stuff you can do with AI now is actually insane: -Add ClaudeCode to a GitHub repo -Mention it in an issue "hey @сlaude fix this" -Go grab a coffee -Come back to a ready PR My only problems: -How the hell do I keep up -We're all cooked

Me: "This method turned out kinda complicated, help me refactor" Cursor: "Nah, it's all clear. Nothing to simplify". 🤨🤨

Tried using AI to fill out a vendor questionnaire. "As a large language model Gemini For Workspaces cannot fill blanks in your Google sheets" Wait, WHAT?!! As we use Cursor, ClaudeCode, MCPs and Agents daily - we forget how far behind "AI for normies" is

I'm gonna show this picture to anyone worried AI will ruin filmmaking.

Looking at Claude Code's pricing, it immediately hits you just how much money Cursor is *burning*. Overdrive mode - just bleeding money. All just to grab more market share. P.S. And honestly, not even sure Claude Code is profitable either. Probably not.

AWS is so annoying. You get a 4-core, 16GB server - but nooo, some "m7ax.whatever" will be 3X faster - with *less* cores, *less* RAM and half the price. Because f**k you, that's why. Here's a 10-mile-long list of instance type descriptions, good luck figuring it out.

Wow, found an awesome solution on reddit: sparecores.com/servers They run s-core/m-core benchmarks on _all_ cloud servers (not just AWS) and publish a table Tick "AWS", filter by CPUs/RAM, sort by benchmark results, CPU-GHz, mem bandwidth etc etc. Awesome tool, bookmark it.

Yaak Yaak API Client is my new favorite REST client P.S. Ironically, made by the same guy who built Insomnia, sold it, now making a new one 🤣🤣

Enjoying my annual MTB vacation

Tried switching from Cursor back to VSCode + GitHub Copilot. Copilot's still behind. Tab completions got a tiny bit better, reasoning and agent-mode file edits are still worse. Back to Cursor - even if it's lagging on upstream VSCode features/extensions.

Just tried jailbreaking AI in my own SaaS product and... Uh oh! It kinda worked

Our SaaS listed in Microsoft Azure "App Gallery" as a SAML/SCIM option. Trying to publish a simple update, here's the timeline day 1: submitted an update day10: no response, raised a ticket with MS partner center day20: no response, raised new ticket in Azure portal day30: no response

In this new AI world, it's not the coding or fancy MCPs that blow my mind - it’s how search has changed. Like, you ask a weird, rare question: "Does the visor from the Airoh Commander 2 helmet fit the 2016 Commander?" And it’s like: "brr brr… not really! sources: x,y,z" 🤯

After ~5 hours comparing inference hosting costs for small LLMs like Gemma/LLaMA, I'm seriously tempted to just buy a damn 24GB Radeon card and run it under my damn desk.

The wildest part of the AI race - no one has a real moat. Any edge vanishes in months via reverse engineering, talent poaching, or clever hacks. Not even hardware is safe: Nvidia meets Google TPUs, then comes Groq with LPUs. Can't decide if that's terrifying or a blessing.

As a founder, nothing makes me more nervous than watching someone review our product on YouTube. Nail-biting, sweating - the full panic mode.😱😱 Sometimes I even ask someone on the team to watch it first, like it's a horror movie and I need a spoiler before I go in.