Mark Szymanski

Realization of the day: I’m not actually driven by money. I want comfort and to provide for my family, but I don’t care about Ferraris or flexing. What I’m really chasing is status through trust. Being widely known, being respected, having access to other respected people I can directly exchange value and collaborate with.

In light of recent events - My next vibe-coding project: an app to track all the stuff you need to remember and do for a baby Spoiler: it's a lot

@RyanDictation You mean wispr flow on android? Or native android?

@MarkJSzymanski Android dictation still has some weird gaps once you actually use it for real work. If it can’t catch names and punctuation reliably, it feels more like a demo than a tool.

A Million down, infinity to go. Dictating > Typing mjs.link/Wispr-Flow

@ajaydsouza I am yea. It's so much better than the the native android speech to text

@MarkJSzymanski Are you using it across your devices? I was struggling to remember it was Whispr Flow

@WojciechPiskorz @OBSProject

@MarkJSzymanski @OBSProject sounds like a cron for hemes ;P

TIL @OBSProject plugins do not auto update. I had a third-party drop shadow plug-in that's been causing me a lot of graphical weirdness. Turns out, the problem has been solved for probably months, and I just never manually updated it.

Addendum: Obviously, this could have been done without AI. My point is that I was able to do this in literally less than five minutes because I was able to clearly articulate what I wanted to do and bully the LLM into doing the work.

Another incredible random AI feat. I, like many people, have a ridiculous amount of shit on my main PCs desktop. I wanted to access it from my laptop when I'm away from my desk but I didn't want to go through the process of backing up 400 GB of files right now. I needed it quickly. I already had Tailscale set up for my Hermes setup so I was able to very easily, with like two clicks, (actually Claude did all the work for me) expose it as a network drive on my laptop.

Let the clanker do the work for you.

Several variations. -Clients building and deploying their websites from Claude Chat. -Non-technical COOs on the Claude Enterprise Plan building all sorts of hosted web apps. All sorts of stuff. Here are some I documented: youtube.com/watch?v=bnMCIG… youtube.com/watch?v=M_nxUG… youtube.com/watch?v=trGuVW…

@MarkJSzymanski What sort of conversations have you had?

It is remarkably hard to deal with prospective clients that are squarely in the AI-Emboldened phase. If you haven't ran into them yet, buckle up.

Treat AI like a human creative. Give it clear instructions, inspiration, and examples. Don't overthink it.

@thekevingeary "Bev The 'Dev'" ™️

@MarkJSzymanski Bev on steroids?

My Hermes is set up locally. My power is out. Not ideal.

Heads up if you use @WisprFlow: They just added this auto cleanup feature and mine was on HIGH by default. That was way too aggressive for my AI prompting as is was removing a ton of detail. So maybe consider turning it down if you haven't.

TIL: Claude Code saves all of the images that you paste in via the VS Code extension directly to your machine in the transcripts.

This shit is no joke. I've learned so much in such a short amount of time, and now I have agents doing real work for me. More on the workflow specifics soon, but in the meantime, here's everything I learned about Hermes in the past 72 hours.

@paol_taj @claudeai Fair point. I would argue it is still beneficial in a lot of ways but definitely would want an opt-out option

@MarkJSzymanski @claudeai I use it for products and businesses that have nothing to do with one another. It would confuse it and cause greater hallucinations. I use projects to separate my stuff and ask to read another chat within the same project to give it context.

genuine question for the AI crowd: why doesn't @claudeai have unified persistent memory across Chat, Claude Code, and Cowork? same account. same subscription. same Claude. but each surface is a stranger to the others. yes I know about CLAUDE.md. yes I know about third-party memory layers. that's not what I'm asking. why isn't this native? what's the argument for keeping them siloed? feels like the single highest-leverage feature they could ship.