Building products to make your life simpler. Ex VP of Eng, AI, BMW self-driving. 💻 https://nitter.cf/t.co/OKj0H3oBWN 🏞 https://nitter.cf/t.co/9Hc8cSrWjT 🐰 https://nitter.cf/t.co/0E3NmviBqD
Germany
Joined October 2009
- Tweets13.7K
- Following1.2K
- Followers27K
- Likes19.4K
Pinned Tweet
I built SimplePost to give ChatGPT and Claude tools to publish on social media.
Not another AI writing tool — your existing AI can prepare, preview, schedule and publish.
It’s been growing fast recently. Time to tell the story over the next few days.
simplepost.social
Vlad Haltakov retweeted
I'm done with dashboards.
I just ask my agent to review the numbers and it comes back with the summary through the connector. That's not a nice-to-have anymore, that's the product.
I have zero desire to get a MacBook with a touchscreen.
I mostly use it with external monitors, mouse, and keyboard, and I don't really see much useful use cases for it.
I'm done with dashboards.
I just ask my agent to review the numbers and it comes back with the summary through the connector. That's not a nice-to-have anymore, that's the product.
Vlad Haltakov retweeted
You can see who has been benchmaxxing most after @ArtificialAnlys released their v5 benchmark.
Looking at you Kimi K3.1 Pro, Gemini 4 and Grok 4.7
Updated 2 of my ChatGPT plugin extensions and the review took only 1 hour.
This used to take weeks before...
Good job @mxstbr and team!
You can see who has been benchmaxxing most after @ArtificialAnlys released their v5 benchmark.
Looking at you Kimi K3.1 Pro, Gemini 4 and Grok 4.7
Vlad Haltakov retweeted
If I was a mathematician...
I'd try to get early access to one of the advanced math models used to generate the 700+ articles, use it to understand the latest advancements in my field, and then use it to advance the field much faster than I could before.
What am I missing?
If I was a mathematician...
I'd try to get early access to one of the advanced math models used to generate the 700+ articles, use it to understand the latest advancements in my field, and then use it to advance the field much faster than I could before.
What am I missing?
Vlad Haltakov retweeted
Grok @bot is actually really good.
It's super fast, has a ton of connectors that just work, and gives me live access to X without paying for the API.
I'm ditching OpenClaw as my default bot. Sorry, OpenClaw.
Grok @bot is actually really good.
It's super fast, has a ton of connectors that just work, and gives me live access to X without paying for the API.
I'm ditching OpenClaw as my default bot. Sorry, OpenClaw.
Vlad Haltakov retweeted
AI agents killed my flow state. I've found a way to get it back.
Before AI, one of my favorite parts of programming was getting into the zone. Good coffee, some background music, and hours inside one problem. Agents broke that rhythm. I'd prompt one, check another, review some code, then switch to a different feature. Coming back meant reconstructing what I was doing. I got a lot done, but after a few hours I was mentally exhausted.
What works for me is mixing tasks that need different kinds of attention. A substantial feature where I think deeply and review code. Deployment or configuration work where the agent guides me through the steps. Something lighter and visual. While one agent works, I move to the next task. The switching feels ordered, and I can get into the zone again.
The mistake is trying to hold every task's full context in my head. My brain has limited RAM. Overload it, and every switch feels like swapping to disk: remembering where I left off and loading the problem again. Letting agents hold the details frees up my attention for reviewing work and making decisions.
For a deployment, I agree on a plan, then ask the agent to track progress and guide me. When it needs my input, I want a short explanation of the situation, clear tradeoffs, and the decision it needs. A wall of text forces me to reload the whole task. I still give instructions and correct the agent, and I trust my judgment more on decisions that depend on the broader context. Its job is to give me enough information to decide without making me reconstruct everything that led there.
I want the speed of AI agents and the feeling of being completely lost in the work. I'm convinced I can have both. How has working with agents changed your flow?
Hey @mxstbr, MCP Events are a great idea, but I'm really struggling to get a good UX for a consumer app...
I have a mobile app for creating workout plans with AI and tracking in the gym. I have a plugin extension (Simple Muscle) and all work good.
Now, when I complete a workout, I want ChatGPT to automatically analyse it and give me feedback - perfect for MCP events!
The problem is that to set it up, my user needs to jump through many hoops even after they already connected the plugin to the app.
1. They need to trigger the activation of the events separately (cannot do it when connecting)
2. I need to ask them to copy the activation prompt
3. I give them a button to jump to the ChatGPT app
4. They need to remember to switch to Work
5. Paste and run the prompt
6. Wait, wait, wait... (and hope they didn't do it already in another chat because it will get rejected)
And finally, when they complete the workout, they receive the analysis in a chat called "Enable workout feedback" not in a new chat.
I see how it is easier to reuse the automation mechanic you already have, but I think event subscriptions should have a better way to be enabled, ideally directly when the user connects the plugin and this should be deterministic, not have the agent "think" about what it should do. But maybe I'm also missing something...
Anyway, if you think it is helpful I can also do a short screen recording of the whole process.
AI agents killed my flow state. I've found a way to get it back.
Before AI, one of my favorite parts of programming was getting into the zone. Good coffee, some background music, and hours inside one problem. Agents broke that rhythm. I'd prompt one, check another, review some code, then switch to a different feature. Coming back meant reconstructing what I was doing. I got a lot done, but after a few hours I was mentally exhausted.
What works for me is mixing tasks that need different kinds of attention. A substantial feature where I think deeply and review code. Deployment or configuration work where the agent guides me through the steps. Something lighter and visual. While one agent works, I move to the next task. The switching feels ordered, and I can get into the zone again.
The mistake is trying to hold every task's full context in my head. My brain has limited RAM. Overload it, and every switch feels like swapping to disk: remembering where I left off and loading the problem again. Letting agents hold the details frees up my attention for reviewing work and making decisions.
For a deployment, I agree on a plan, then ask the agent to track progress and guide me. When it needs my input, I want a short explanation of the situation, clear tradeoffs, and the decision it needs. A wall of text forces me to reload the whole task. I still give instructions and correct the agent, and I trust my judgment more on decisions that depend on the broader context. Its job is to give me enough information to decide without making me reconstruct everything that led there.
I want the speed of AI agents and the feeling of being completely lost in the work. I'm convinced I can have both. How has working with agents changed your flow?
Vlad Haltakov retweeted
I know how to code, but how do I teach my son?
AI is changing the rules of programming, so I’m trying two things with my 8-year-old. We sit down together and use AI to vibe-code his little personal website. While the AI is doing its thing, we switch over to Scratch and build a simple game ourselves, step by step.
I want him to get comfortable building with AI, but also learn the basics so he understands how things actually work.
I don’t know if this is the right approach. How are you teaching your kids?
The EU wants to stamp the text you spellchecked with AI the same as the AI slop others generate. Because they have no idea about the technology, but they can force the AI labs to do it.
Removing it can be simple and 100% private: simpleunmark.com
Watermarks have limits. They’re often undetectable, especially in short passages. Rewriting or translating text can completely remove the watermark.
That’s why, for now, only approved researchers will have access to our detector, so they can help us evaluate and improve the technology.
We’re treating text watermarking as an ongoing area of research. We’ll keep testing and improving, with feedback from users, developers, policymakers, and researchers.
openai.com/index/eu-text-pro…
Based on the fantastic open source library by @guillaumemeyer
github.com/guillaumemeyer/wa…
I know how to code, but how do I teach my son?
AI is changing the rules of programming, so I’m trying two things with my 8-year-old. We sit down together and use AI to vibe-code his little personal website. While the AI is doing its thing, we switch over to Scratch and build a simple game ourselves, step by step.
I want him to get comfortable building with AI, but also learn the basics so he understands how things actually work.
I don’t know if this is the right approach. How are you teaching your kids?
