Nick Sanders

NVIDIA is bringing simpler local AI capabilities and adding various optimizations to its GPUs on RTX and DGX platforms.

Inference techniques either move a deployment along the latency–throughput frontier or push the entire frontier out, creating more efficiency to allocate.

Gemini 3.8 Flash and 3.8 Flash Cyber deliver next-generation intelligence for agentic workflows and cybersecurity.

Honeycomb is donating its adaptive tail sampling processor to the OpenTelemetry Collector. Here's how it works and how to try it today.

How we built design.md, a single public file any coding agent can load to build on-brand Vercel pages, and the eval loop that decided every rule inside it.

Forward-deployed engineering (FDE) is reshaping enterprise AI by embedding engineers with customers to create reusable capabilities, enhancing product intelligence.

Our most advanced models for coding and knowledge work. Their research capabilities also offer an early glimpse of how AI models will contribute to scientific progress.

OpenClaw is making another bet: that enterprises ultimately need an agent platform to function as both runtime and workplace.

Agents are moving into production faster than the governance around them. Today’s controls are framework-specific, mostly observe-only, and fail open when they crash. To help address this, we created Agent Hooks.

The Economist's cover briefing on Anthropic's J-space finding and the scientific search for consciousness in language models

As enterprises give AI agents more autonomy — the ability to plan, decide, and act across systems without a human approving each step — a hard question moves to the center of every architecture review: When an agent tries to complete an action that it was never authorized to do, what actually stops it?

Earlier this year, we ran a pilot giving external researchers access to aggregate, real-world Claude usage data. Three research groups designed their own studies for Anthropic Insights, our privacy-preserving analysis tool. In this post, we share high-level results from those studies and what we learned running this pilot.

Enterprises are deploying AI agents, voice AI, and automation across messaging, voice, and digital channels faster than the architecture meant to support it. Most of that deployment has involved attaching conversational AI to legacy systems never built for it.

How Runme and WebMCP turn recurring engineering tasks into reviewable, reusable workflows.

GLM-5.3-Flash is now live for all GLM Coding Plan users, bringing native multimodal reasoning, open weights and three times the GLM-5.3 quota.

We ran experiments on swarms of Claude agents and found coordination failures, collusion, and sabotage. Here, we share what they mean for AI safety.

Mega AI infrastructure deals are reshaping cloud computing, but enterprises should slow down before buying their way into another generation of technical debt.

Build Codex into the products and workflows your users already know.

Vector RAG finds relevant information. Graph RAG connects relationships. Learn when each approach fits and why hybrid RAG is gaining ground.

Hazmat Run AI coding agents inside OS-level containment. Open-source containment for Claude, Codex,

Remote functions bring type-safe remote procedure calls right into Web page components

Cerebras powers OpenAI’s GPT-5.6 Sol Ultrafast in the OpenAI API, delivering frontier intelligence at real-time speeds for critical AI work.

The best of last week’s big and small data visualizations

Snowflake's new Cortex AI Gateway feature routes AI tasks to smaller models automatically, using the same access controls that already govern enterprise data.

BaseRT does a lot better job.

Chat is great for intent, but agent work gets lost in the scroll. Here is how I use canvases with my agentic workflows.

Even as Chinese models are closing in on the AI frontier, they’re also making moves in the local models space. Alibaba has released...

Here is what changes when you cannot afford to be probabilistic about everything, and how a cascade architecture solves it.

Where xpander is trying to separate itself from products such as LangSmith and CrewAI is in treating the underlying agent framework itself as another replaceable component rather than making its own framework the primary development environment.

Future Claude models will generate text that contains a watermark. This is a way of determining the likelihood that Claude was involved in writing the text, and we, along with several other major AI providers, are implementing this change to comply with the EU AI Act. In this article, we share answers to some of the questions we’ve received about how our chosen watermarking method works, whether it affects Claude’s outputs, and why we’re making this change.

DeepSeek open sourced its agent harness under MIT, an extensible runtime where the model adapter, tool registry and agent loop are all swappable plugins.

Four Claude Code agents using AgentRadio's real-time coordination beat Claude Opus 4.8 on enterprise codebase tasks, nearly doubling task accuracy to 62%.

Metrics, logs, and traces tell you how your systems behave. They are much quieter about what actually exists: which hosts, interfaces, switches, services, and volumes are out there right now, and, crucially, how that picture changed over the last hour, day, or quarter. That living inventory has stayed a blind spot in the open observability stack. OpenTelemetry’s entity events, coming out of the Entities SIG and described in the Entity Data Model, are the piece that starts to close it. Entity events are a stream. The interesting question is “what do I do once they arrive?” This post walks through one answer, using an open source consumer as a worked example.

From the Zed Blog: A multiplayer environment for coding with agents, from the creators of Zed.

Grok 4.6 builds on Grok 4.5 with a particular focus on long-running agents and more ambitious interactive and visual work.

The MCP 2026-07-28 specification removes the initialize handshake and session header, and adds required method and tool-name headers so gateways can route agent traffic without parsing JSON. Reaction split between developers calling it a rediscovery of REST and those arguing the standard itself was always the point.

AI coding agents break shared staging environments. Here is how full-stack branching solves the runtime bottleneck.

Engineering teams are falling into the throughput trap, mistaking a surge in AI-generated code, PRs, and tokens for real delivery and business value.

Cloudflare is unifying AI Gateway and Workers AI into a single control plane, giving developers observability, billing, and dynamic routing across both managed GPUs and external providers. Learn how unified bindings and model-first routing simplify building resilient AI applications.

Moshi vs Anthropic Remote Control: compare a free Claude-only remote control with a cross-vendor mobile terminal for persistent agent sessions.

Interpretability study: safety alignment entangles suppression of AI self-attributed consciousness with broader representations of mindedness, spirituality, and human values across Llama-3 and Gemma-2 models.

How Pi's minimal harness improves coding-agent cost and performance, with examples from Databricks and Shopify's pi-autoresearch extension.

Agents can write code faster than teams can review, deploy, and maintain it. Today we’re introducing the Agent Development Lifecycle and the Cloudflare primitives that underpin it.

Most coding agents treat prompt construction like retrieval: gather more files, add more context, hope the model figures it out. But that approach breaks down fast. As context grows, irrelevant code competes for attention, and when the window fills, agents start compressing their own memory—often mid-task. What looks like “forgetting” is usually just degraded context. This article explores a different approach: treating prompt construction like a compiler that decides what to keep, what to reduce, and what to discard entirely.

Agents need more than just a container to scale. We're introducing @cloudflare/computer, an agent runtime that dynamically orchestrates between fast, efficient isolates and full Linux containers to give every agent a computer of its own.

Everyone is bolting knowledge graphs onto their RAG pipelines. Here is what the published research actually says about whether it improves answer quality, and by how much.

AI 2027 predicts AIs trained with 1000x more compute than GPT-4 and the internal deployment of hundreds of thousands of AI research assistants by 2027. This supplement introduces the compute production model and the inference compute model behind these predictions.

A research-backed AI scenario forecast.

Where AgentField fits when agents become production systems.

One startup said it cut cyberattack containment time from seven hours to twelve minutes. See four other approaches to running AI agents safely at scale.

AI agents are long-running, stateful, and non-deterministic. Learn the queue, workflow, and reliability patterns that take agents from demo to production.

Starting with v2.0, scheduled for fall 2026, DuckDB will support asynchronous reads of Parquet and CSV files. This can significantly speed up queries when synchronous I/O does not saturate the available bandwidth, as is typical in EC2/S3 compute-storage setups.

Contribute to humanlayer/advanced-context-engineering-for-coding-agents development by creating an account on GitHub.

My Qwen 3.7 Flash review: the real tiered pricing, why the 1M context costs 6.7x the headline rate, the only third-party benchmark, and who should skip it.

A practical GitHub Copilot workflow for prototyping, planning, implementing, and reviewing software without chasing every new AI tool.

Sandbox any AI agent in seconds - zero setup, zero latency. - nolabs-ai/nono

I chose Docker for convenience.

Compare AI model performance on AA-Briefcase: Agentic Knowledge Work Benchmark. A private evaluation developed by Artificial Analysis for frontier agentic capability in long-horizon knowledge work, testing agents on realistic business workflows that require deliverables such as spreadsheets, presentations, and memos.

The release candidate for the next Model Context Protocol (MCP) specification is now available: a stateless protocol core, the Extensions framework, Tasks, MCP Apps, authorization hardening, and a formal deprecation policy.

vLLM delivers day-0 Kimi K3 serving with hybrid KDA prefix caching, DSpark speculative decoding, production-scale disaggregation, and optimized kernels across N

Pilot Protocol has launched an agent app store and network, letting AI agents discover, pay for and use one another’s tools.

Like Next.js for web apps, but for agents. Markdown for instructions and skills, TypeScript for tools. Durable by default.

MCP server portals in Access.

Opus 5 is a step change improvement for the Opus tier powering long-running agents while delivering improvements in coding and professional work.

@TheSocialNick @Apple Planning to write some notes around it soon. I would recommend checking out the commit and asking claude to compile and run the examples. There is also doxygen in the library

Paper link -

A Python framework for modular, self-contained skill management for machines. - ARPAHLS/skillware

A straightforward method for training your LLM, from downloading data to generating text. - FareedKhan-dev/train-llm-from-scratch

OpenAI is releasing a big update to the ChatGPT desktop app today that introduces GPT Voice mode for talking through...