Serban Petrescu retweeted
Microsoft releases NLWeb
NLWeb uses MCP to make it simple to interact with websites in a standardized way.
Devs can now convert any website into an AI app.
MCP is to NLWeb what HTTP is to HTML.
This went largely unnoticed this week, but it looks like a big deal.
Serban Petrescu retweeted
Cursor now supports 1M context-windows, the most expensive models, and unlimited tool calls with API pricing.
Serban Petrescu retweeted
today we are introducing codex.
it is a software engineering agent that runs in the cloud and does tasks for you, like writing a new feature of fixing a bug.
you can run many tasks in parallel.
Serban Petrescu retweeted
🧵Google DeepMind just dropped a bombshell:
An AI agent that autonomously writes algorithms better than humans.
It’s called AlphaEvolve, and it could completely change how we build software and solve problems.
Here’s why this changes everything👇
Serban Petrescu retweeted
AI SDK 5 concept: loop until condition
Often you want flexible control over when the steps loop stops.
Serban Petrescu retweeted
🦜 LangGraph Studio v2
We're launching a new and improved version of LangGraph Studio. Includes:
- No more desktop app! Run it locally
- Edit prompts and other configuration directly in studio
- Playground integration
- Pull down production traces and run them locally to debug
See the launch video here: youtu.be/Mi1gSlHwZLM
Serban Petrescu retweeted
🔥 🤖 🎥 SURPRISE LAUNCH: GPT-4.1 now available in ChatGPT
if you haven't tried it yet because it's only been available via API, I recommend you try it...
it's different.
🎯 here it is searching the web and explaining how to generate an MCP Server from a natural language spec along with code
Serban Petrescu retweeted
DSPy’s “value prop” is simple:
If you stop writing prompts and instead write Signatures, you gain access to an ever-improving repertoire of algorithms (Modules and Optimizers).
You also future-proof your AI system: you stop tinkering with the stuff (like specific prompts) that will get washed away in a few months—and you focus instead on the information flow fundamental to your application.
DSPy isn’t “hard”. There’s a learning curve to doing proper software design for AI. Skipping DSPy doesn’t make those principles go away.
The principles are below:
Serban Petrescu retweeted
Automating stuff with a human in the loop is 10x harder than it looks.
Automating stuff to where it has no human in the loop is at least 1,000x harder than it looks
Serban Petrescu retweeted
You can now connect GitHub repos to deep research in ChatGPT. 🐙
Ask a question and the deep research agent will read and search the repo’s source code and PRs, returning a detailed report with citations. Hit deep research → GitHub to get started.
Serban Petrescu retweeted
Gemini 2.5 Pro Preview 05-06 scored 77% on the leaderboard, coming in 2nd place close behind o3 (high).
Full leaderboard:
aider.chat/docs/leaderboards…
Serban Petrescu retweeted
Gemini-2.5-Pro-preview-05-06 is now my top coding model.
It beats o3 and Claude 3.7 Sonnet on several of my hard prompts.
One example prompt: "Code simulation of water in a bucket that is rocking back and forth." See how it crushes o3 and Sonnet.
Google, call it Gemini 3!
Serban Petrescu retweeted
new guide published, this time for web development in @cursor_ai
link in comment below
Serban Petrescu retweeted
Introducing Deepwiki MCP! 🔥
- Get fresh Docs in Cursor
- Free, no signup or API Key!
- Fetch a single page or multiple
(Unofficial MCP, couldn't wait!)
Example: outdated Tailwind v3->v4 upgrade vs updated Deepwiki upgrade.
Link in the Comment 👇
Serban Petrescu retweeted
A few things to note:
1. GenAI tools will result in *far far more* software produced than before
2. The more agentic the tool, the more it will generate (vs autocomplete)
3. "Accepted lines of code" does not mean
