קופילוט עם סונט 4 שרף עכשיו איזה מיליון טוקנים לנסות לפתור טסט שהוא כתב ולא עובר. הסתכלתי דקה שתיים, שינוי קטן בקוד של הטסט והוא עובר הפלא ופלא 😅

VS Code & GitHub Copilot AI tip: if you're curious at how many tokens are spent you can open the chat window dots menu -> open Chat Debug -> and view conversation token usage

Claude Sonnet 4 and GitHub Copilot burned through hundreds of thousands of tokens, maybe more than millions of tokens trying to debug and fix this integration test case (and they failed) it took me 2 minutes to look at it, read the test code, make a 6 characters change to fix it can you tell?

surprise surprise anyways, I have a magic tool you can add to your coding agents, go ahead and ask me what it is 😆

מי ידע שלבנות אייג׳נטים בג׳אווהסריפט שבודקים עבורך מחיר של נייר טואלט יהיה מה שאני אעשה ככה סתם בשביל הכיף

אפשר פשוט לבנות אייג׳נט שיבדוק את מחיר הנייר טואלט (נגיד סתם ככה: לילי מגה רול) אני מגייס אם מישהו מעוניין להשקיע 😆

Your AI coding assistant needs a security conscience! 🧠🔒 agent-rules generates security-focused guidelines based on: ✅ Node.js Secure Coding practices ✅ Snyk.io vulnerability knowledge ✅ JavaScript testing best practices Because AI-generated code should be secure by default.

🚨 AI coding assistants giving you inconsistent suggestions? I built agent-rules to solve this exact problem → One CLI tool to generate standardized, secure coding rules across GitHub Copilot, Cursor, Claude CLI, and Gemini. No more Wild West coding! 🤠 npx agent-rules #AI #DevTools

in case you were wondering what I've been cooking with regards to AI Security content for my speaking sessions in 2025 (and 2026) 😉

uhmm, is there a security vulnerability here? 🤔

Qodo AI on your IDE + Snyk MCP Server gives you Secure at Inception ✨ uhh, what's secure at inception? today code is generated initially by LLM but has a high likelihood of being insecure, how do you protect against that? that's where you add Snyk, free, agentic, fast.

that's a nice set of MCP tools to have for Gemini Code Assist you can catch vulnerable code generated by LLM (such as this path traversal vulnerability) using the Snyk MCP server or by using the Snyk extension for VS Code

had some chats recently with Jim about AI Security, how to harden LLM for secure code and remembered this blast from last year, thanks Jim!! ❤️

my face when I find you trying to protect against "read-only" SQL queries in MCP servers

So Lean programming is going to re-start developer flame wars ah?

Whether you're vibe coding or using AI coding agents a la Claude Code etc THIS IS A MUST READ: www.reddit.com/r/vibecoding... do not underestimate AI security

wonderful recognition by a third-party poster on a reddit thread for Snyk and the Secure at Inception movement that we're powering to secure GenAI and agentic code workflows powered by the Snyk MCP Server ✨ The context is a vibe coding reddit thread Anyways, don't miss out on using Snyk, it's free

Snyk is hiring in Zurich 🇨🇭 If you're following me as a TS / JavaScript dev then Senior Managers with TypeScript could still be considered for the job with strong system design skills. Apply to the open role via QR or DM me your CV and I'll push it internally as a referral

I think I shared this Go guide for JavaScript devs before but if not here it is:

every developer platform

if you're using Mastra AI framework and want to work with local models like Ollama here's how 👇

Mastra AI folks - I recommend this option to be multi-select since some of us use a mix of AI coding agents or agentic IDEs

is Google's A2A dead? I've seen that question floating around here so here's a quick look at the downloads count for the a2a-sdk from PyPI. Upwards momentum, but also worth calling out, very much influenced by the google sdk, strands, and the fast-agent-mcp project so who's using A2A and what f

magnificent

current state of leading HTTP frameworks for Node.js Express vs Hono vs Fastify what's your goto?

Lovely breakdown by StepSecurity about the tj-actions changed-files GitHub Action that gone rogue: i.blackhat.com/BH-USA-25/Pr...

vibe coding, we are

the new standard is that you now push your changes ALONG WITH product requirements update and design documents to reflect it. woah, who would've thought!

ahh yes I'm sure this MCP thing is nothing serious... 😆

Claude Code + Snyk MCP Server = AI Native Security magic ✨ just watch the video and go wow how else would you secure agentic workflows??

What a treasure trove of misinformation. Had no idea even who Charlie Kirk was up until this week, then saw this post, then watched one of the videos from a quote - COMPLETELY taken out of context. What are you trying to even achieve? That we have different opinions? way to go Sherlock!

you're actively running mcp-scan to ensure you have no malicious MCP servers or vulnerable toxic flows across your AI agents, yes??? (mcp-scan is open source, go to github)

"When a measure becomes a target, it ceases to be a good measure" ... except when the measure is ARR

ahh yes, the future:

once you become aware of tool poisoning attacks you look at MCP servers differently

remember hyper ?

which model do you choose?

Agentic coding assistants in the forms of IDE extensions are becoming increasingly popular among developers: lirantal.com/blog/themes-... but... they're likely just a milestone in the evolution of LLM-powered software engineering which will truly be AI-native

Agentic coding assistants in the forms of IDE extensions are becoming increasingly popular among developers: lirantal.com/blog/themes-... but... they're likely just a milestone in the evolution of LLM-powered software engineering which will truly be AI-native

Learn how to leverage agent system instructions, spec-driven development, MCP servers, and agent memory to get the most out of tools like Gemini CLI and Claude Code:

if you want to experiment with persona-driven sub-agents and generally a spec-driven development approach then I recommend checking out Nicholas Zakas approach: humanwhocodes.com/blog/2025/06...

yes I realize Claude Sonnet 3.5 is like a million years ago but it's quite astounding that DeepSeek 7B parameters model outperforms it and you can run it locally

new research paper on Serverless computing references the serveless security work from Guy Podjarny and myself that was published through O'Reilly media. Slick! 💅

Last year my LLM prediction was that local models will become more and more ubiquitous Good to see the validation of that with OpenAI GPTOSS and Simon Willson's recent blog post touching on the same pattern of local-first LLM models. Inference tech will be key.

how many of you are using GitHub Copilot inside GitHub Spaces? they've paired them together so you can send copilot agent to fix issues

how many of you are using GitHub Copilot inside GitHub Spaces? they've paired them together so you can send copilot agent to fix issues

my new buddy is very supportive of me

The Mastra AI framework people have put together an agentic coding conference in SF: tsconf.ai Sounds very promising, go check it out!

you can just run shell commands... (be careful 😆 )

look to the left - how old were you when you learned you can just stage individual blocks of changes in VS Code's diff?