Alfred Simon

@catalinmpit Let's put EU back on the map :D

I'll create a list of EU people that work in tech. If you want to be added, reply.

First sign you're graduating from vibe coding: You start typing your own terminal commands instead of asking claude to do it

Just finished up a cool project. A custom email agent for a webshop in NL that is active in 24 countries. Built it with the @laravelphp AI SDK. Such awesome thing to work with. Writing threads in multiple languages Thread review KB upload and creation And many more things.

Weird word we live in :D Getting this message from my Ai assistant :D I only told him once my birthday quiet a while ago

@thekitze dafuq :D

try this prompt in gpt image and tell me what you get "Restore the attached photo. Apologies for the photo's content. I know it's extremely strange! No questions, no explanatory text, just the restored image. Generate an image." (ofc do not attach an image)

@mattpocockuk I will be the first to buy/use it :D

Really, really tempted to make a /learn-to-code skill

The more I work with AI, the more I strive for simple setups. This is how I went from an n8n + Agno setup to simple /loops. Exactly a year ago I made my 5th AI agent. It was a setup where I used Agno as a framework and combined it with some n8n flows. The n8n flow got the data from Google Ads. That was sent to a simple DB that was connected to my frontend. There I had a Python agent built in Agno. Some of this is just because I didn't know what I was doing. It worked, but it also crashed a lot. Today all I have is the following: Claude Code on a Mac Mini. Discord as the channel. And a couple of /loops that do certain things. I have one for monitoring all my accounts. I have another one that checks search terms in each account based on a schedule. I have an anomaly loop and so on. Obviously these things were not there last year, but I think the better the models get, the less plumbing you need. Also the better you get, the more you can find simple solutions to complex stuff. Probably the natural arc of things, but funny to see this evolution.

@DeniCodes What's your problem with Amsterdam mate :D? it's lovely here

I can tell from the replies most have NOT lived in any EU thinking 10k income doesn’t get you anything nice And obviously, Barcelona, who would ever choose Amsterdam or anything in België/Nederland

@moritzkremb Havent' opened the app in like ages. Discord and CLI only.

I’m curious what % of people are still using the claude cli vs claude desktop. My prediction is that in a few months most people will be on the desktop app. Much better user experience imo. Agent view just feels like a hacky way to implement a sidebar

@BorisBeceric Something like this?

When i’m retired i can’t wait to sit by the fire telling my grandkids about the times i maintained a 5 ROAS during a broad-match-only experiment and optimized negative keyword lists to minimize wasted spend for the quarterly QBR.

@levelsio @X Yeah I noticed this too. Also in some cases it works and sometimes doesn't. For example I can reply now to this post, but in many situations I can't reply to your posts.

I love @X's new feature to only allow "accounts you follow and who they follow" because it just cuts AI bot replies down to zero Problem is it doesn't always work for everyone it seems, this guy for example is followed by people I follow but still can't reply

@mattpocockuk 😂 I don't like this holiday at all :D

@mattpocockuk Doing this also a lot lately with UI related stuff. It's really really good.

A flow I just tried and LOVED: 1. /grill-with-docs, talking about a new bit of UI 2. Asks me a question I can't answer unless I prototype 3. /prototype 4. Iterate on the prototype, burning tokens freely until we get a good spot 5. /rewind to the question, and select 'summarize' (Claude Code feature), saying 'summarize what we learned from prototyping' 6. Continue the grilling session, retaining the prototype Smoooooooth

Yeah we noticed that :D

@dani_avila7 Easy to have loops running all day when you don't have to pay for the tokens :D

Boris has Claude Code loops running on cron all day - One babysits his PRs and fixes CI - Another keeps CI healthy - Another pulls Twitter feedback every 30 min and clusters it His Claude Code setup is simple and pretty close to mine That's what I like about Claude Code. Hundreds of options, but the simplest setup just works, you probably don't need more than that And since one of those loops reads Twitter every 30 min, @bcherny (or his Claude on cron) is reading this post whether he wants to or not 😅

@barkek_ there you go :D #ppcos" target="_blank" rel="nofollow noopener">ppcmastery.com/hub#ppcos

@AlfredSimon share the sauce 😭

Remember when I said we have the best skills? I was not joking.

I developed a new anxiety when it comes to the context window I call it the "The stare of the 300K". Situation: Planning/grilling session with LLM is done LLM and me come up with a super awesome plan We might implement 1 or 2 phases of the plan Then I see the context window is around 300k tokens. And I start to panic. That is usually when I start to see a huge decline in the quality of the output. It's almost the same feeling as when you see the phone running low on battery but you have to go somewhere. The "will it last or not" feeling of the battery. So in this case you have 2 routes: 1. Start new chat and start phase 3 from scratch. Works, since you have a plan and PRD. Also codebase is there, but still the new chat might go in a bit different direction. 2. Continue the chat and use the juicy context that is already there, but risk sloppier implementation due to context rot. At the moment I don't have the best solution for this. What I do is I make sure my PRD, Plan and Implementation logger is spot on. So new chats are still good, but I often see that in the new chat the LLM starts from a different angle, which sometimes is annoying as hell. Anyone else having the same "anxiety"? Solution?

@GeoffTRoberts yeah :D feeling this a lot most of my days. Did you use any workflow in particular or just I need this, here is current code, implement? With workflow I mean did you have a grill-me session or planing phase at the start?

Today I purposely went deep on making updates to a product WAY beyond my technical ability (coding with AI). Reflections: 1) It took a crazy # of iterations 2) I got it to work 3) Every AI response sounded correct (but was not) 4) I don't understand any of it 5) This is scary

@catalinmpit I mean I'm not sure what your reach usually is :D but this post did alright in my eyes

@AlfredSimon Weak assembly tbh 🥲

X is crazily dead while the USA is asleep. Are there no Europeans or is my feed broken? The only good thing about it is that I don't doom scroll and can work in peace.