Martin-Zack Mekkaoui
6.3K posts
Martin-Zack Mekkaoui
@martinzack
Doing stuff @AWSCloud
Katılım Ağustos 2008
713 Takip Edilen854 Takipçiler
Les limitez d’utilisation de Claude c’est pire qu’un rollercoaster
ClaudeDevs@ClaudeDevs
Usage limits are up, effective today we're: 1) Doubling Claude Code's 5-hour limits for Pro, Max, Team and seat-based Enterprise plans 2) Removing peak hours limit reduction on Claude Code for Pro and Max plans 3) Substantially raising our API rate limits for Opus models
Français
@mitsuhiko @ubmit One of the greatest advantage of Vienna maybe. I agree on the convinience of Abus folding one, just in some place of the world they won’t last one single night
English
@martinzack @ubmit To annoying to use. I had this lock up more expensive bikes. For where I’m using it, it’s fine.
English
New toy for commuting and it’s fun. I feel like gravel bikes for all the hate they get are a great compromise if you don’t want to own multiple bikes.
English
@mitsuhiko @ubmit Easiest thing to break and steal your bike. Get yourself a Kryptonite and their insurance
English
@ubmit It’s an abus folding lock. This one is pretty old and the precise model is no longer manufactured. But there are many options. amazon.com/ABUS-Bordo-600…
English
@thdxr @vercel Hey! We open-sourced ara few weeks ago. It’s built to be npm/pypi for AI github.com/ara-registry/s…
English
Martin-Zack Mekkaoui retweetledi
@lucasmeijer Grosssprachmodellentscheidungsparalysenfaulheit
Deutsch
Martin-Zack Mekkaoui retweetledi
We have to close source because AI finds out security issues is a pretty weak argument honestly.
Bailey Pumfleet@pumfleet
Open source is dead. That’s not a statement we ever thought we’d make. @calcom was built on open source. It shaped our product, our community, and our growth. But the world has changed faster than our principles could keep up. AI has fundamentally altered the security landscape. What once required time, expertise, and intent can now be automated at scale. Code is no longer just read. It is scanned, mapped, and exploited. Near zero cost. In that world, transparency becomes exposure. Especially at scale. After a lot of deliberation, we’ve made the decision to close the core @calcom codebase. This is not a rejection of what open source gave us. It’s a response to what risks AI is making possible. We’re still supporting builders, releasing the core code under a new MIT-licensed open source project called cal. diy for hobbyists and tinkerers, but our priority now is simple: Protecting our customers and community at all costs. This may not be the most popular call. But we believe many companies will come to the same conclusion. My full explanation below ↓
English
2016: I need to be able to code on an airplane, I can't use an IDE in a browser
2026: I can't code unless I'm connected to an AI model in the cloud
English
@tpschmidt_ Or you can just use AutoMode and you have virtually no controllers to manage
English
It makes the infrastructure feel (almost) invisible and invincible!
Devs just request resources in their manifests and the compute appears 🥰
English
Karpenter is great because it ignores the idea of groups entirely. It looks at exactly what your pods need and talks to the AWS API to spin up the right instance at that moment 🏗️
The process is straightforward:
1. A pod can't find a home
English
@bcherny @yoavgo Actually Bedrock does have a CountTokens API, just not supporting new models yet docs.aws.amazon.com/bedrock/latest…
English
@yoavgo 👋 yep we do this for Bedrock, Vertex, and Azure, since they don’t have a token-counting API available yet. When using Anthropic API we use the token-counting endpoint directly
English
TIL claude-code sometimes calculates the token count for a context by sending the request to haiku model and observing the resulting token count field in the response.
English
Martin-Zack Mekkaoui retweetledi
Today we're releasing a tool we've done for ourselves and been working on for a while: zml-smi.
A mix between nvidia-smi and nvtop. No install, runs out of the box on NVIDIA, AMD, TPU and Trainium.
Releasing today. Enjoy!
ZML@zml_ai
Introducing zml-smi, a universal diagnostic and monitoring tool for GPUs, TPUs and NPUs. Real-time insights into the performance and health of your hardware. Read & Download: zml.ai/posts/zml-smi/
English
Martin-Zack Mekkaoui retweetledi
ft.com/content/f62751… UK risks losing defence tech start-ups to relocation amid funding delays
English
In order to properly support AWS Neuron and provision machines easily for the team, I put Skypilot in place with a service account.
Which got my account flagged for fraud.
And now I can't create a VM.
"Under review" they say.
*sigh*
English
Martin-Zack Mekkaoui retweetledi
Martin-Zack Mekkaoui retweetledi
MCP sucking is a harness problem, not an MCP problem
MCP unlocks behavior that is fundamentally impossible to get via CLI or APIs
Bad auth, too much context usage, all get solved with an execution layer - your agent writes code to progressively discover and call tools
Garry Tan@garrytan
MCP sucks honestly It eats too much context window and you have to toggle it on and off and the auth sucks I got sick of Claude in Chrome via MCP and vibe coded a CLI wrapper for Playwright tonight in 30 minutes only for my team to tell me Vercel already did it lmao But it worked 100x better and was like 100LOC as a CLI
English
Martin-Zack Mekkaoui retweetledi
I've just released a new version of typeagent, a Python library I've been working on since mid last year --more and more using Claude-- that implements memory for agents.
Not originally my idea, I mostly ported the TypeScript version by Steve Lucco and Umesh Madan. This release was improved a lot by Bernhard Merkle.
To install, use "pip install typeagent". Changelog: github.com/microsoft/type…
English
Martin-Zack Mekkaoui retweetledi
In other life, they would be living in Berlin Mitte and drive their kids to kindergarten with one of those cargo bikes
English
I built a CLI tool to speak to your favorite agentic coding client (Claude Code, Kiro) using the new Mistral Voxtral Mini 4B, 100% local github.com/mekza/speak-to
English
@pbeyssac Faut lire l’article avant de dire des âneries boss man
Français
Incroyable, la récente panne d'AWS était due à un agent IA qui a fait n'importe quoi.
Qu'on ne nous dise plus *jamais* que les infrastructures numériques US sont gérées avec un sérieux et une sécurité inégalés.
rat king 🐀@MikeIsaac
amazon's internal A.I. coding assistant decided the engineers' existing code was inadequate so the bot deleted it to start from scratch that resulted in taking down a part of AWS for 13 hours and was not the first time it had happened incredible ft.com/content/00c282…
Français
Biggest incident management improvement action for this outage is, if an outage correlates to a recent deployment, rollback first, ask questions later.
English
Supabase Feb12 outage highlights:
- Accidentally enabled blocking of public traffic for all VPCs
- Same AWS account used across services allowed one service’s change to impact all
- Preprod didn’t catch cause not similar to prod
- No canary to detect external connectivity issues
English
@LelloucheNico @juliencdt @Fabien_Mikol Peut être que le plus gros problème actuel est le traitement de ce genre d’infos par des journalistes qui n’ont peut être pas le baggage technique suffisant pour comprendre pleinement un sujet (et qui veulent juste sortir un papier)
Français
@Fabien_Mikol Confondre le service et le modèle = le plus gros problème actuel dans le traitement grand public. Il faut dire que des Google avec leurs Gemini et Gemini n'aide pas.
Français
C'est dingue que Le Monde continue encore de faire des articles sur les capacités de ChatGPT sans même préciser quel modèle est évalué. En l'occurrence ici c'est ChatGPT-4o, qui commence déjà à dater. Et pourtant c'était précisé dès l'abstract de l'étude !
lemonde.fr/pixels/article…
Français