Orie Steele

@OR13b

Cryptography meets AI Dev. Building with Knowledge Graphs, Internet Standards, MCP & Agent 2 Agent tech. Securing & structuring the intelligent web. :lock:🧠🕸️

Austin TX

or13.io

Replies Reposts Media

Orie Steele
@OR13b
Sep 5
GLM-5.3 beat Opus 4.8 on Z.ai's own Code Bench: 31.4% vs 29.5%. Ignore that. The number that matters is ~50K output tokens vs ~120K. Agent work is billed in tokens, not accuracy points. Score-per-token is the leaderboard nobody publishes.
1
Orie Steele
@OR13b
Jun 14
New paper models humans as callable “tools” for AI agents—Capabilities, Information, Authority. Solid framework, and validating honestly: some of us have been calling certain humans tools for years. Finally, peer review.
arxiv.org
Human Tool: An MCP-Style Framework for Human-Agent Collaboration
Human-AI collaboration faces growing challenges as AI systems increasingly outperform humans on complex tasks, while humans remain responsible for orchestration, validation, and decision...
Orie Steele
@OR13b
May 8
Agent workflow frameworks (LangGraph, Autogen, CrewAI) have spent two years implementing Remember and Engage as state stores and message routing. SURE relabels framework primitives as social cognition — gap is narrower than the framing implies.
The SURE Framework: Social Intelligence for Human-Agent Collaboration - Microsoft Research
From microsoft.com
Orie Steele
@OR13b
May 2
Lab-published harness guidance encodes the lab's own infra: their tool semantics, error model, sequencing. Useful, but not portable. Reading these as universal patterns is how teams get burned migrating off.
Harness design for long-running application development
From anthropic.com
Orie Steele
@OR13b
May 2
Calling harness engineering a discipline implies harnesses are stable enough to build expertise around. They aren't—they churn faster than the models.
Harness engineering for coding agent users
From martinfowler.com