It’s almost always KVCache quantization
AI
-
Disaggregated Inference: GPU for Prefill, RDU for Decode
By
–
AI agents spend their time doing two very different jobs: understanding context and generating responses.
— SambaNova (@SambaNovaAI) 15 juin 2026
Disaggregated inference sends each stage to the hardware best built for it.
GPUs for prefill. RDUs for decode. Better performance from both. ⚡ pic.twitter.com/KFXRdUBLWxAI agents spend their time performing two very different tasks: understanding context and generating responses. Disaggregated inference sends each step to the hardware best suited for that task. GPU for prefill. RDU for decode. Better.
-
Cursor accounted for 40-50% of Anthropic revenue early on
By
–
Holy, in its early days, Cursor accounted for roughly 40% to 50% of Anthropic's revenue. And Claude Code was just a reserach project. How quickly everything has changed.
-
Dynamic Linear Attention for Adaptive Token Compression
By
–
"Dynamic Linear Attention" Most long-context linear models compress tokens using fixed blocks or logarithmic schedules, but long texts are not uniform. Stable segments can be summarized, while the
-
Full Analysis of CUA-Bench Benchmark for Computer Use Agents
By
–
Full analysis by @qi_zhengyang
: https://
snorkel.ai/blog/cua-bench
-benchmarking-computer-use-agents-on-professional-software/
… Tasks: https://
cua.ai/cuabench/regis
try/cua-bench-kicad
… -
Fleet turns a chat into an agent for your daily applications
By
–
Describe a task in a short prompt. Fleet devises a plan, takes action, and works with the applications your team uses daily. Turn a chat into an agent in just a few clicks.
-
Index entire Linux kernel in 3 minutes with codebase-memory-mcp
By
–
How do you index the entire Linux kernel (28M lines of code) for an AI agent in 3 minutes? You stop letting the agent read files one by one. There is a fascinating new open-source release called codebase-memory-mcp. It's a code intelligence engine that swaps traditional
-
Google’s 50-page breakdown on AI agent failures and vibe coding shift
By
–
My friend @Saboo_Shubham_
, @AddyOsmani and the team at Google just published a 50-page breakdown on the shift from vibe coding to agentic engineering. It covers the new Software Development Life Cycle with AI Agents. The most interesting takeaway? Most AI agent failures aren't -
Frontier computer-use agent fails on 21 of 25 electrical engineering tasks
By
–
New research: we partnered with @francedot and @ddupont808 at @trycua to stress-test a frontier computer-use agent on real electrical engineering tasks.
— Snorkel AI (@SnorkelAI) 15 juin 2026
25 expert-authored KiCad tasks. 4 passed. 0 build-from-scratch tasks succeeded.
The failure modes are concrete and they point… https://t.co/qHDpHC4jFFNew research: we partnered with @francedot and @ddupont808 at @trycua to stress-test a frontier computer-use agent on real electrical engineering tasks. 25 expert-authored KiCad tasks. 4 passed. 0 build-from-scratch tasks succeeded. The failure modes are concrete and they point
-
Voice control agent builds concept, cast, storyboards, and shots
By
–
Use voice control (very beta), or simply type to have the agent build everything from concept, to cast, to storyboards and generations.
— Pika (@pika_labs) 15 juin 2026
Shots are generated directly from approved boards, and automatically placed in both your timeline and your asset library.
Editing happens via… pic.twitter.com/1JtzZJKLklUse voice control (very beta), or simply type to have the agent build everything from concept, to cast, to storyboards and generations. Shots are generated directly from approved boards, and automatically placed in both your timeline and your asset library. Editing happens via