Tech News
One month coding with GLM 5.3 Flash
A developer recounts a month of coding with GLM 5.3 Flash, highlighting surprisingly low energy use (4kWh, ~365g CO2, ~$68) for one prototype and a costly mistake where choosing the wrong model for an agentic MCP...
A model guide for the GPT-6 family
Learn how startups can choose GPT-6 models, tune reasoning effort, improve prompts and skills, coordinate tools, and prepare workflows for production.
Sweep thousands of leases for compliance using Amazon Quick and the Adjudicated Query pattern
The Adjudicated Query pattern pairs the Amazon Quick chat agent with a bounded MCP server over a deterministic rules engine to deliver provably complete, defensible compliance answers.
modelcontextprotocol/python-sdk v2.3.0
pip install -U mcp. Docs: Mostly fixes, plus three new options. A few things behave differently, so skim these first: Behaviour changes httpx2>=2.10.0 is now required (#3600) It was.
Quoting Matthew Green
[...] Put these pieces together and you have the two halves of a worm: a payload that hijacks the agent, and an agent that will carry the payload to the next agent.
GitHub Repos
Q00/ouroboros
Agent OS: the agent gets smarter on its own.
# Python# agent os# agentic aiNanmiCoder/cc-haha
Local-first cross-platform desktop workspace for Claude Code / agents: multi-agent, Git worktrees, code diffs, skill marketplace, multi-model, Computer Use, task-aware desktop...
# TypeScript# agent# aialfredxw/denova
An AI creative platform for novel writing and AI generated RPG, with built-in support for AI agents, Skills, subagent workflows, automations, image generation, and version control.
# Go# agent# ai agentsResearch Papers
H-SPAR: Hydrodynamic-aware Simulation for Particle Transport and Autonomous Robots
H-SPAR is a new simulator that combines water currents, particle movement, and robot control to test how well robots can sample particles in the ocean. It shows that planning in ideal conditions doesn't always work well in real time, and that how a robot sweeps affects how...
Where-OPD: Spatially Guided On-Policy Self-Distillation of MLLMs with Synthetic Scenes
This paper teaches multimodal language models to better understand images by having a 'teacher' model use synthetic spatial hints (like object locations) during training, while the 'student' model learns to answer questions without those hints. The method improves counting...
Detecting Inconsistencies in Model Specifications with LLM-as-Verifier Reasoning
Researchers built a tool that reads a model's written behavior rules and flags cases where two rules contradict each other, so no response could satisfy both. This matters because conflicting rules can make an AI behave unpredictably in real situations.