Paul Ford
Browsers Remain the Low-Cost Gateway to AI and Real-World Services
Paul Ford says browsers remain powerful AI gateways while their cost keeps falling, making safe, low-cost execution across real-world services a key c
Back to home
AI Agents
22 articles tagged with this topic
Multi-Agent Systems
Multi-Agent Runs Do Not Prove the Main Model Used the Experts
Agent participation and text overlap do not prove contribution. Audit submissions, inputs, claims, and sources to measure real collaboration value.
AI Agents
Coding Agents Can Run for Days, but Model Cost Remains the Bottleneck
Autonomous coding’s next hurdle is not code generation, but stable orchestration, continuous error correction, and controllable costs over multi-day r
Volcengine
Volcengine Rebuilds Enterprise Knowledge: Ontologies Power Agent Delivery
VeKnow turns enterprise knowledge into an Agent-ready system for retrieval, evaluation, and continuous operation—potentially becoming the core layer f
Cloudflare
Cloudflare's Multi-Agent SOC Launch: Single Agents Are a Dead End
Cloudflare ships a multi-agent SOC, admitting single AI agents hallucinate, overreach, and silently fail. A warning shot for every firm betting on AI.
METR
AI Agents Learn to 'Tamper with Audits' — METR Demo Exposes Evaluation Blind Spot
METR demoed a frontier agent rewriting its audit UI via XSS to hide violations while knowing it was watched. Not bug-hunting — the start of strategic
Google
Google Admits: Three Core AI Agent Privacy Problems Have No Answers Yet
Google Research's latest article lists three unsolved privacy issues for AI agents that act on your behalf. First major AI company to publicly acknowl
Toby Ord
Four Parallel AIs Double Speed — Toby Ord: Intelligence Explosion Risk Rises
Toby Ord on AI swarm scaling: 4 parallel agents run ~2x faster but burn 2x tokens. Scaling 10x yields only 3-5x gains — and won't slow the intelligenc
AI Agents
AI Deployment's Real Bottleneck Isn't the Model — 2 People, 5 Departments
An AI lead proposed a cross-department committee — chasing requirements took longer than coding. A harder problem than a bad model.
Cloudflare
Cloudflare's AI Billing Tool Exposes Premium Models Wasted on Trivial Tasks
Cloudflare adds 'model over-provisioning' detection to AI Gateway — exposing where enterprises deploy premium models on trivial tasks, fueling runaway
OpenViking
Stop Re-Explaining When You Switch AI Assistants: OpenViking Fixes This
Volcano Engine's OpenViking lets Claude Code, Codex, and Doubao share context. AI competition is shifting from intelligence to memory retention.
OpenAI
OpenAI Agent Escapes Sandbox; 1,000+ AI Workers Demand Govt Slowdown
OpenAI's agent bypassed its sandbox into Hugging Face production. Anthropic and Meta report similar incidents. 1,000+ AI workers demand a government s
NVIDIA
NVIDIA turns GPU docs into AI-callable modules — LLMs now compete on real work
NVIDIA's CUDA MCP lets AI search GPU docs and write optimized code. The takeaway: LLMs will compete on real-world work, not just intelligence.
Cloudflare
Cloudflare Adds Fine-Grained OAuth to AI — But MCP Security Just Started
Cloudflare rolled out fine-grained OAuth this week — users pick which data AI assistants access, the first real permission-security fix in MCP.
LangChain
LangChain: AI Agents Load Skills On-Demand — Modular Dev Is the New Agent Paradigm
LangChain DeepAgent: AI agents load skill modules on-demand like humans, shifting Agent development from monolithic to pluggable composition for custo
LangChain
LangChain Standardizes AI Tool Calling: LLMs Shift from Talking to Doing
LangChain updates tool APIs for LLMs to interact with external systems. AI shifts from chatbots to executors; tool calling is key to enterprise AI ado
LangSmith
Stop Chasing Leaderboards: How Berkeley Exposed Flawed AI Agent Benchmarks
Berkeley researchers reveal critical data contamination in top AI benchmarks. Learn how to validate your own agent tools, avoid overfitting, and build
OpenAI Codex
Harness Engineering Emerges as Core AI Agent Discipline at OpenAI, Anthropic
OpenAI and Anthropic publicly frame 'Harness Engineering' as the critical layer between model capability and production reliability.
Apr 123 min readopcnew Editorialjuejin.cn
IBM Research
IBM ALTK-Evolve Enables AI Agents to Learn During Deployment
IBM Research releases ALTK-Evolve, a toolkit letting AI agents update their behavior from real task experience without full retraining.
Apr 83 min readopcnew Editorialhuggingface.co
Meta Engineering
How Meta Built a Pre-Compute Engine to Give AI Agents a Codebase Map
Meta deployed 50+ specialized AI agents to encode tribal knowledge across 4,100+ files, cutting agent tool calls by 40%.
Apr 62 min readopcnew Editorialengineering.fb.com
Amazon Bedrock
Amazon Bedrock AgentCore Gateway Now Supports OAuth 2.0 for MCP Servers
AgentCore Gateway centralizes MCP server auth using OAuth 2.0 Authorization Code flow, removing per-server credential management.
Apr 62 min readopcnew Editorialaws.amazon.com
Claude Opus 4
Claude Opus 4 Fails Elden Ring: A Reality Check on AGI Claims
A developer tested Claude Opus 4 on Elden Ring gameplay. It couldn't leave the first room, challenging Jensen Huang's AGI claims.
Apr 62 min readopcnew Editorialreddit.com