Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-02-21 · source-backed
Each link below shares sources, entities, or timing with this story.
PCAS: Policy Compiler for Secure Agentic Systems — The first paper to provide measured enforcement results for agent policy compliance (48% to 93%). Uses dependency graphs and Datalog-derived policy language with a reference monitor intercepting all actions. Three case studies...
When the meter's running hot, the obvious move is a cheaper model that's actually good. Mistral shipped one. Devstral 2 (123B, modified MIT) scores 72.2% on SWE-bench Verified. Devstral Small 2 (24B, Apache 2.0) hits 68.0%. Both carry 256K context. Mistral claims 7x cost effic...
arXiv 2608.00765 compresses retrieved docs into query-conditioned visual representations, sidestepping the trade-off where hard compression is query-aware but weak and soft compression is strong but needs costly offline encoding. Beats both baselines across varying retrieval d...
Agent-Orchestrated Adaptive RAG (arXiv:2606.05658) finds agentic enhancements are not universally beneficial. Dynamic query decomposition gained +0.17 MRR on a structured DevOps benchmark but degraded ranking precision on a multi-hop benchmark, and the self-reflective loop onl...
Anthropic: Prompt Caching as Production Architecture — The most significant engineering disclosure from Anthropic this year. Claude Code's static-first prefix structure, Plan Mode as callable functions, Tool Search with defer_loading, and compaction maintaining identical syste...
Two numbers from this paper should change what you do with your .claude/skills directory this week. First: 65.7% of the benefit from agent skills comes from procedural anchoring. Explicit knowledge injection accounts for 4.5%. Second: expand the skill pool from 5 items to 100,...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.