Yannick

Yannick

@osteel.bsky.social

Freelance Product Engineer at Yellow Raincoat Ltd. Founder of pixelwatcher.io. Organiser of @phpsussex.uk. I post in French once in a while

355 followers126 following739 posts

Yannick reposted

📢 Our next event is up! On Wednesday 11 November, James Turner will share how he’s building and running software with AI tools. A useful temperature check one year into the rise of agentic models. 📅 Details & registration 👇 www.meetup.com/php-s... #php

Seeing software companies use yesterday's events to nudge worried businesses away from PHP... as if other languages don't have lurking security flaws. If anything, the response has reassured me. I feel more confident than ever building software with PHP for my clients.

AI hasn't changed the fact that I can still spend a couple of days on a single PR. No matter how good the latest models are, they still suffer from tunnel vision and it's still on us to know when to take a step back and refactor

I don’t know if it’s just my baseline that’s adjusted, but Open 5.5 felt much faster at launch and I now find it pretty slow again. Just me?

Look I was excited to give Dungeon Crawler Carl a go but that Amazon Kindle exclusivity kinda poured cold water on that

Yannick reposted

European attitudes toward digital privacy and technology regulation are often dismissed in the US as technophobia. But there are distinct, historically grounded reasons why Europe has moved further and faster to regulate Big Tech, writes Enrique Dans, a professor at IE Business School, Madrid.

buff.lyThree reasons Europe learned to distrust Big Tech before America didEuropean attitudes toward digital privacy and technology regulation are dismissed in the US as technophobia. But there are distinct, historically grounded reasons why Europe has moved further and…

Yannick reposted

Is anyone here gainfully employed as a software engineer / developer / programmer / insert equivalent term, who only learned to code after having generated code with LLMs? Or do you know anyone who is?

I want to take stock of my current AI-assisted development workflow and compare notes with others. I mostly use Claude Code these days, locally. The flow looks like this (skills linked at the end):

View full thread

If a lot of UI/UX changes are involved, I like to ask for an HTML artifact, which is basically how I do my own QA (much easier than trying to reach each screen/state manually).

I want to take stock of my current AI-assisted development workflow and compare notes with others. I mostly use Claude Code these days, locally. The flow looks like this (skills linked at the end):

View full thread

* Main implementation * /simplify * /polish (if UX/UI changes) * /review * /analyse (deterministic checks – style, architecture, etc.) * /cover (gaps in test coverage) * /qa (if UX/UI changes – in a browser) * /wrap-up (docs updates, ADRs, PR) * /monitor (makes sure CI is green)

Basically, the agent hands me a green PR, and I review it manually from there. Depending on the changes I ask for, I re-run some of the loop’s steps.

I want to take stock of my current AI-assisted development workflow and compare notes with others. I mostly use Claude Code these days, locally. The flow looks like this (skills linked at the end):

* Open Claude Code in the terminal * Call /tackle skill with a ticket/issue number * That pulls the task, gathers context, and /grill-me about it (a customised version of Matt Pocock’s skill) * I review the final plan, then the loop (mostly subagents):

Yannick reposted

Quick Claude tip: if you're reorganising your local folders, don't forget to tell Claude about it, e.g.: "This is the project's new API folder." Claude will migrate existing memories to the new folder.

I don't know for how much longer we're going to put up with Anthropic being so unreliable. I'm not sure why I'm paying so much to see my work being interrupted twice a day.

I'm at a stage where my main engineering issue is the duration of my agent loops. Any task goes through the same loop, but it's disproportionately heavy for many of them. I need to tailor the loop to the task somehow

Just watched Inside the Manosphere after finishing Adolescence yesterday, while watching my one-year-old son sleeping on the baby monitor. How hard can it be to keep him off the internet till he’s 20?

Agentic work means I work on two things in parallel most of the time now. I can handle 3, but opportunities to do so are fairly rare. I occasionally tried to do more, but looks like 3 is my limit – beyond that, that’s too much context-switching for my brain

I understand some of the feedback around Opus 5 now. It definitely is more conservative, in an annoying way. Random directives it picks from elsewhere now seem to have as much weight as my own instructions.

I’m trying to use Sonnet as my daily driver to save on token consumption, but it’s just too dumb. I end up wasting tokens (and my time) instead

If you're looking for a task for Claude to do semi-autonomously (and that you can monitor from your phone, for instance), try `/doctor`. It does a lot more than it used to, and will probably flag and offer to fix things currently flying under your radar

Claude Code tip: To maximise the number of 5-hour windows I get per day, I set up a dummy routine that runs at 5:30 am. It starts my first session before I wake up, so when I begin work I have a fresh session with a couple of hours left, and can fit three sessions into the day.