ilyass elfouih

I will not respect/trust/buy from you. If your brand marketing strategy is evolved around rage/click baits. It’s not marketing, just dishonest. This needs to stop!

Human mistake not an AI 👀

@BenLesh No, can confirm it was not related to bun. Just developer error

Be careful with influencers, they did it to us before with JS frameworks era and they are doing it now. Make your own judgements of what works and what not!

@arbaoui_mehdi This one is by far my best one lol

عشنا وشفنا <div> كتسونطرا ببرومبت.

@kayintveen How do you deal with multi tasking? With worktrees I am more drained than ever! It’s supposed to help us work less isn’t ?

@elfouih worktrees for parallel work is the biggest throughput unlock. once you have 4-5 agents running in parallel branches you cant go back to sequential. the speckit + skills combo for context is what keeps it from becoming chaos

I just realized that I didn’t write a single line of code for more than a week now! Crazy times

@AnassAb01 One of the things that I hate about ´ agentic engineering ‘ is that it’s very difficult to measure the effectiveness of anything you add to your project

@elfouih I'm interested in learning from you what you configured, how and why? If you could share anything I'll appreciate it.

@AnassAb01 And this was done gradually, I only add something if it make sense

@AnassAb01 Basic stuff with claude: - A monorepo with BE and FE - Skills for Jira, FE, BE… - Hooks to run linting/prettier - Agents teams (depends on what I am doing) - Speckit if it’s a new feature - Docker sandbox or dangerously skip permissions - Worktrees (for parallel work)

@AnassAb01 I can see this work for greenfield projects or small apps. But for big entreprise apps with heavy business logic not really. This week was the only week when I was able to pull it off in such environments!

@elfouih I think you have some kind of complicated setup, for me I just fire up my IDE > new chat > write a prompt (give it a research.md and plan.md files) > review the output > accept, edit or reject. That's it!

@AnassAb01 I’ve been using agentic AI for a long time now, TBH I was faster in most cases (except writing tests). The amount of work required to setup skills, agents, spec files, and monitoring them was never productive to me!

@elfouih A weeek ??? Those are rookie numbers you got pump those numbers up 😂

@zakelfassi @messenger That’s very cool, it looks like the only bottleneck now is the human in the loop lol, I bet you wish you can create a new session of yourself

@elfouih @messenger The equivalent of running a 10-12-person eng team… sometimes overwhelming; most days it’s business as usual. Yes, I review everything that passes, in addition to having different harnesses set up just for reviews (of course with a very extensive test surface).

this would've been sci-fi when we were building the @messenger dev platform 10 years ago. now it's just a Wednesday night on the deck.

@zakelfassi Good pattern idea, I’ve been running the meta skill creator manually, but never thought to bake it into every session. My only issue with skills: they don’t always get invoked when they should. Not sure if you’ve seen the same

essay: zakelfassi.com/skdd-skills-dr… repo: github.com/zakelfassi/ski… portable build loop: forgeloop.zakelfassi.com

october 2025. 1:54 AM. typed git init /ClaudeRecursiveSkills/ and something crystallized. by 2:17 AM — three commits later — a skill had created another skill. the recursion logged itself: "can_create_meta_skills": "Yes" one question added to every build loop since: should this become a Skill? if yes: write a SKILL.md. the agent finds it next time. vibe coding has no memory. specs accumulate descriptions. SkDD accumulates capabilities. day 1: 45 min per blog post. day 90: the repo knows your infrastructure better than you do. essay + repo in the reply.

@laytoun Interesting, maybe with spec driven development those requirements can be stored in spec files so they can be accessed later on by agents? I am really curious on how everything will work out eventually

@elfouih Depends how big the debt is and how much its spread. AI still lacks the business requirements that led to some decisions in first and those can bring back the classic issues

Technical debt is increasing as fast code as code generation right now ! The speed offered by AI is outpacing our capacity to audit the resulting code, leading to, probably, a new category of technical debt.

@f10ux3 @laytoun And proud!

@elfouih @laytoun People like you who call themselves "experts" & "consultants" 😂

A software engineer ending his working day, after they survived another round of layoffs, forced to use AI tools in fear of being replaced by AI, watching the stock market flatline and juggling the workload of three teams.

@f10ux3 @laytoun A lah ya khouya lah, who hurt you?

@laytoun The biggest myth in tech is “software engineer” anyone can learn this field and surpass people who call themselves “engineers”, now with ai advancement they re harvesting the results of their arrogance and elitism.

@_iMeriem Guilty of trying, but the moment the model asked for my personal picture I am like Naah, you already know to much

Y’all look nice in that prompt 😁 Not falling for that thread, bsehtkoum 😂