3/ Start with one file. CLAUDE.md in your project root. Build commands, your folder layout, conventions, the "always do this" rules. Claude reads it at the start of every session. This one file does what those seven were trying to do.
SOFTWARE
-
NVIDIA Metropolis VSS 3: Video Search and Summarization with 16 New Skills
By
–
NVIDIA Metropolis Blueprint for video search and summarization (VSS) 3 is here.
— NVIDIA AI (@NVIDIAAI) 24 juin 2026
Now your coding agent can analyze massive live streams and libraries of videos with a simple natural language prompt. Here's what's new:
– 16 new agent skills: Search, summarize, alert, report,… pic.twitter.com/UojjUu8orkNVIDIA Metropolis Blueprint for video search and summarization (VSS) 3 is here. Now your coding agent can analyze massive live streams and libraries of videos with a simple natural language prompt. Here's what's new: – 16 new agent skills: Search, summarize, alert, report,
-
360 Cameras: Preparation for 3D Gaussian Splatting by Epic
By
–
Guess what 360 camera support is building up towards? 3d gaussian splatting of course! Only a matter of time before epic rolls it out. https://t.co/eP7fLoDTd6
— Bilawal Sidhu (@bilawalsidhu) 24 juin 2026Guess what, the support for 360 cameras prepares what? 3D Gaussian Splatting, of course! It's only a matter of time before Epic deploys it.
-
Voyager 1.6 divides AI deployment into compile and run phases
By
–
Deploying AI models to embedded targets usually means a full SDK on a device that doesn't necessarily have room for it. So Voyager 1.6 splits that workflow in two. You compile your model on a development machine using axelera-devkit, which produces an .axm file. Then you run it
-
New Siri will replace small Google searches for hotkeys
By
–
New Siri is going to eat up so many of the small Google searches. All of the “how do i…” “whats the hotkey for…” Just works.
-
Brain reads threads and provides a complete deliverable
By
–
Users can mention Brain on any task, doc, or discussion to read the entire thread, extract from connected apps like Google Drive, GitHub, and Slack via MCP, and then provide a complete deliverable. Try it.
-
Closing the Agent Loop with LangSmith and Context Hub
By
–
"Most agents don't learn, they just leave traces."
— LangChain (@LangChain) 24 juin 2026
In 12 minutes, @jakebroekhuizen breaks down how to actually close the loop.
Surface issues with LangSmith Engine
Write memory updates back to Context Hub
Let the agent actually improve between runs
If you're thinking about how… https://t.co/aIfMwT33hM pic.twitter.com/QcjKez5SQ4Most agents don't learn, they just leave traces. In 12 minutes,
@jakebroekhuizen
explains how to actually close the loop. Surface issues with the LangSmith engine
Write memory updates to the Context Hub
Let the agent -
Vector Search Perf module for slow AI retrieval diagnosis
By
–
2/ Next problem → sluggish AI retrieval. Scaling it up is notoriously difficult. This Vector Search Perf module walks you through diagnosing slow queries with Atlas Metrics. PLUS actually managing memory sizing and quantization in full production > https://
fandf.co/4b6Q3uW -
Seedance 2.5: 30s generations with up to 50 references
By
–
Wow. Seedance is really good at turning greyboxed 3d references into final quality pixels. Like really good.
— Bilawal Sidhu (@bilawalsidhu) 24 juin 2026
Seedance 2.5 coming with 30 second generations and up to 50 (!) references is gonna be mad. Try the workflow Reid is running below. pic.twitter.com/OtbSzfCFhZWow. Seedance is really good at turning grayed-out 3D references into final-quality pixels. Really good. Seedance 2.5 is coming with 30-second generations and up to 50 (!) references — it's going to be insane. Try the workflow Reid runs below.
-
Scoped agent access to stack felt surprisingly clean
By
–
gave an agent access to my actual stack today
— Nico (@nicos_ai) 24 juin 2026
no scary “it can see everything” moment. scoped permissions, every action logged
felt illegal how clean it was https://t.co/xeAVrovr1D pic.twitter.com/NyPzYIQWf3gave an agent access to my actual stack today no scary “it can see everything” moment. scoped permissions, every action logged felt illegal how clean it was