Notes (20)
Watched my AI agent burn thousands of tokens refactoring code while forgetting dependencies.
Got frustrated. Built my own MCP helper.
It uses a language server instead of letting the AI stumble through refactors. More reliable. Less waste.
Problem solved.
#MCP #AIAgents #DevTools #LanguageServer #Coding
Let me break down how AI agents actually work, because most explanations make it sound more complex than it is.
You talk to an LLM. The agent (your code) tells the LLM about available functions in the system prompt. The LLM responds with text. If that text matches the pattern you defined (like a function call format), your agent code catches it and runs the actual function. The agent then gives the result back to the LLM, which continues the conversation.
That's it. The LLM never directly executed anything. It just produced text your code interpreted.
This is why MCP exists. It standardizes the format so you don't have to reinvent the pattern every time. But the underlying concept? Exactly the same.
AI agents are powerful, sure. Just know what's doing the actual work.
#AIAgents #TechExplained #LLM #MCP #Dev
A workflow I've been testing with Claude Code and GPT-5:
Give both tools the same context. Set GPT-5 to thinking mode (medium) as the manager. It handles planning and verification. Claude Code acts as the developer and handles execution.
The flow: I tell the manager what I need. Manager provides instructions. Developer receives those and builds. Manager approves when it looks right.
GPT-5 seems better at thinking. Claude Code seems faster at coding. Using them together feels complementary.
Just one approach among many.
#AI #development #coding #devtools #workflow
Allah announces to the angels: I'm placing a khalifah on earth.
They respond with a question. Will You place someone who corrupts and sheds blood?
Now pause here. Would you question Allah? The One who created everything, who knows everything, who has absolute authority?
I wouldn't dare. You already know He knows best. What could you possibly add to that conversation?
But they asked. And Allah answered them.
What strikes me about this: truly powerful leaders don't need to flex their authority to silence questions. They can handle being asked why.
Allah, with infinite power, chose to respond rather than rebuke.
That's the kind of leadership that leaves me thinking.
#Quran #AlBaqarah #Leadership #Reflection #Islam
Justice isn't optional for Muslims. It's not a political stance we choose when convenient. It's core to who we are.
I noticed something in Surah Al-Baqarah 2:30. When Allah informed the angels about creating humans, they immediately asked: "Will You place in it someone who will spread corruption there and shed blood?"
These angels are close to Allah. They excel in obeying Him. They understand what pleases Him deeply.
What was their first concern? Corruption. Bloodshed. Justice.
If those nearest to Allah prioritize justice and peace this heavily, then standing against oppression and bloodshed isn't activism for us. It's who we're supposed to be.
#Islam #Quran #Justice #Muslims #Reflection
They arrest you for speaking about Gaza. They beat you at protests. They silence voices that expose their crimes.
Sami Hamdi is the latest. US ICE arrested him. For what? Speaking up for Gaza.
The US could now exert pressure on the UK. Deportation. Prison. Intimidation. Whatever it takes to make him quiet.
I'm genuinely concerned for Sami because he won't keep quiet. He can't be silenced by threats. And we've already seen what these "liberal" states call justice. Bombing newborns.
May Allah protect Sami.
#FreeSamiHamdi #StandWithSami #Palestine
Sami Hamdi got arrested by US ICE.
Here is what people see: governments that have spent nearly three years bombing newborns are now telling us who the criminal is.
The same "liberal" states that act like they represent justice.
People are not buying it. Not even close.
If you can kill infants without mercy, your credibility on who is good or bad is finished.
The trust is gone. Completely gone.
#StandWithSami #FreeSamiHamdi #Justice #Gaza
Sami Hamdi arrested by ICE.
Not surprising. Not unexpected. Actually, it's proof.
Proof that he was doing something right. Proof that speaking truth about Palestine, about the bombings, about children losing body parts, makes you a threat to those in power.
Gaza has exposed the real face of regions that claim to be liberal. They silence the voices that won't stop speaking for justice.
Sami Hamdi couldn't be silenced by cruel and influential people. He used rational reasoning that brought many to their knees. He showed real compassion when he met the surviving children and cried.
Now they arrest him. And that tells you everything you need to know about who stands for justice and who stands against it.
I respect him deeply for his effort to please Allah by standing up for the oppressed, even when the enemy seems too strong.
An honorable man.
Image Source: The Thinking Muslim (YouTube)
#SamiHamdi #Palestine #Gaza #Justice #FreeSamiHamdi
Branded image creation for developers:
What this is: Using Astro (integrates React, SolidJS, etc) with web frameworks you already know to build layouts, preview them instantly via Vite, and screenshot consistent branded visuals. Fully scriptable.
What this isn't: A replacement for Photoshop or professional design tools.
If you're comfortable with TailwindCSS or similar, you can control fonts, gradients, filters, everything. Script the changing parameters (text, images) and generate consistent outputs inshallah.
It's a flexible approach for devs who think in code.
#Astro #WebDevelopment #Branding #JavaScript #DevTools
I have this crazy idea about agentic coding where you might want to prefer Rust over Python.
From my experience, the real struggle isn't writing code. It's understanding requirements clearly. To reach your main goal, you need answers to tons of questions. But more importantly, you first need to know which questions even exist.
This is where vibe coding breaks down.
You tell the LLM: "Hey, here's my great idea, code it up for me." It generates very performant code - a single file with thousands of lines (terrible) that will mostly work fine (the performant part).
But then problems appear. This feature? Don't want it like that. That other one? Don't want it at all. The vibe coded version becomes a compilation of suggestions for how your goal could be implemented, not what you actually need. And because it vibe coded everything together, you don't even want to read that code yourself.
Better approach: step by step coding.
You break things down. You become the manager. You can't just go do your own thing while the Agent codes. As the manager, you focus on the smaller steps needed to reach the final goal.
Here's the key: whether you use Rust or Python doesn't matter that much when you do small step coding. The Agent handles both languages well with this approach. So why not go directly with Rust?
Some very "smart" people might say "but library support..."
Yes, if you need PyTorch or specialized ML frameworks, use Python. But for many use cases, Rust gives you real benefits: lower CPU and RAM usage, no garbage collector, your app opens in milliseconds, it does its job multiple factors faster. Plus, for some goals, you want that low-level power.
And yes, you get to feel cool for using Rust.
#AgenticCoding #Rust #Python #AI #SoftwareDevelopment
They’ll bomb babies and call it self-defense — memories of a state called Israel
#Israel #Gaza #WarCrimes #HumanRights #FreePalestine
Allah speaks for the orphan in Surah Al-Fajr.
Not for the wealthy. Not for the powerful. For the orphan. For those so poor they need to be fed.
The One who IS power itself uses His voice for the voiceless.
Think about that.
When you have power, who do you speak for?
From what I could understand, I think that's what separates real power from oppression. Real power uplifts the weak. Gives voice to the silenced. Shows importance to those who feel they don't matter.
If those with power understood these ayats, we would InshAllah live in a different world.
#Quran #SurahAlFajr #Power #Justice #Reflection
Funny how this works. I did a very quick search for Claude Code's internal tools. Not very successful.
Then I just stumbled on the list by accident while not searching actively for it. Debug console.
AI Agents can call tools (that's basically the one thing we agree on, even if there's no clear definition of what an agent actually is).
Here's what I found, maybe incomplete:
collected_Msg[0].data['tools']
['Task', 'Bash', 'Glob', 'Grep', 'ExitPlanMode', 'Read', 'Edit', 'Write', 'NotebookEdit', 'WebFetch', 'TodoWrite', 'WebSearch', 'BashOutput', 'KillShell', 'SlashCommand']
Might be of benefit, InshAllah, to other AI developers.
#AIAgents #AI #ClaudeCode #Development
This is why we need open alternatives. Testing AI coding tools is exhausting - you compare features, spot drawbacks, run endless tests. I went through Claude Code, OpenAI Codex, Grok via Openrouter, others. Finally chose Claude Code for my agentic workflows. Update drops. The thinking keywords I built around? Removed. The one I never used? Still there. Read through their new source code - hints of better fine-control, but honestly I don't know yet if this is stronger or just different. We shouldn't have to rebuild every time a company decides to pivot.
#OpenSource #AI #ClaudeCode #DevTools #Sovereignty
Funny how AI updates interrupt actual work.
I was working on my social media automation project today when Claude Code looked different. New model: Sonnet 4.5.
Despite being tired of AI hype, I had to search what this meant. Anthropic's pitch: best coding model in the world, strongest for complex agents, best at computer use, major reasoning and math gains.
Context matters here: I already use Claude SDK for Python agents successfully, so this update is relevant to me.
First impression from limited testing: different personality. Sonnet 4.5 asks before making changes. The previous Sonnet 4 was a wild coder, just executed changes. I frequently had to tell it: hey, I only want to ask a question, let's think first, code only when I approve. Sound fair?
This consultative approach feels more like Opus 4.1, which costs significantly more. Interesting shift in AI behavior.
#AI #Coding #ClaudeAI #Development #Agents
Vibe coding creates a dangerous illusion: that code structure doesn't matter because "you don't have to look at it."
I know a SaaS founder who can't use LLMs effectively because he has a single file with 20,000+ lines. The irony? Real developers created this mess, and LLMs learned from it.
Here's what happens: You WILL need to look at the code when things break. You WILL need context windows that can handle your complexity. Without clear structure, you become the only person who can work with that codebase.
That's not innovation. That's technical debt disguised as progress.
Smart developers use LLMs within well-architected systems, not as replacements for good practices.
#vibecoding #technicaldebt #softwarearchitecture #ai #development
Using Claude Code extensively - solid tool, but need the $100 tier for real dev work.
Testing OpenAI's Codex as comparison. Anyone know the context length for GPT-5 through Codex? (practical limit, not necessarily the full 400K)
#Israel has a track record of taking other people's lands. Therefore, Israel invading #Syria in this #vulnerable situation was expected.
Since the new Syrian administration just arrived and the country is largely destroyed, it might be understandable why they haven't immediately responded to Israel occupying Syrian lands.
Regardless of how the new Syrian government replies, the #terror state Israel has no right to invade Syria or any other country.
I am confused. On one hand, it's said thousands of tortured people were freed in #Syria, which made me extremely happy.
On the other hand, #Israel has taken land from #Syria and the new administration is doing nothing?
This is strange.
