AI Agents
195 pieces on AI Agents, including 19 step-by-step guides.
Guides
Build a ReAct Agent with Guaranteed JSON Output Using Outlines
Use Outlines' constrained generation to make a ReAct agent that cannot emit malformed JSON—no fine-tuning, no GPU required.
Build a Multi-Tool Claude Agent with MCP Servers in Python
Wire Claude to a local calculator, Brave search MCP server, and Anthropic-hosted tools in one extensible Python agent loop.
Build an Agentic RAG System with smolagents and ChromaDB
Use smolagents CodeAgent to run multi-step retrieval over a ChromaDB vector store, letting the agent plan its own search strategy.
Build a Grounded Search Agent with Google ADK
Build a web-grounded AI agent with Google ADK's built-in google_search tool, from pip install to a grounded answer in the ADK dev UI.
Build a Text-to-SQL Agent with smolagents in One File
Wire a Hugging Face CodeAgent to a SQLite database so it answers plain-English questions with verified SQL queries.
Run parallel Claude agents with asyncio and a message hub
Build a lightweight Python orchestration layer that runs multiple Claude agents concurrently, routes messages between them via a shared hub, and synthesises results.
Build a Deep Research API Agent Pipeline with OpenAI
Run a four-agent Deep Research pipeline—clarification, prompt enrichment, MCP file search, and citation extraction—using OpenAI's agents SDK.
Event-Driven Architecture for Agentic AI: The Architect's Guide
A comprehensive architectural guide to designing resilient, real-time agentic AI systems using event-driven architecture — covering loose coupling, fault isolation, reference architecture, and governance patterns.
Top 5 Small Language Models for Agentic Tool Calling
Explore 5 small language models (SLMs) that feature open weights and first-class tool-calling support for local agentic workflows.
Beyond Prompt Engineering: Deep-Dive RAG for Intent Detection & Slot Filling in AI Agents
Stop relying on massive, static context prompts. Learn how to combine Retrieval-Augmented Generation (RAG), vector search (FAISS), and history compaction to achieve scalable intent detection and slot filling.
RAG vs MCP in 2026: What Each Solves and How They Fit Together
RAG is a technique for grounding answers in your data; MCP is a protocol for connecting models to tools and data. Build an MCP server and see where RAG fits.
How to Analyze and Fine-Tune Agent Reasoning Traces
Learn how to inspect agent reasoning traces, extract tool calls, visualize behavior patterns, and prepare multi-turn agent conversations for supervised fine-tuning.
APIs vs MCPs vs MCP Gateways: A Practical Guide for AI Builders
APIs still power software, but AI agents need a different connection layer. Here is how APIs, MCP servers, and MCP gateways fit together in modern agentic systems.
Building AI Agents in Python with Pydantic AI
Learn how to build production-ready AI agents in Python using Pydantic AI, with structured outputs, custom tools, and dependency injection.
Run Claude Code Locally with Ollama (2026 Setup)
Connect Claude Code to Ollama's Anthropic-compatible API, choose a supported coding model, set a 64K context, and verify inference stays local.
Build a Tool-Calling Agent with a Local Small Model and Ollama (2026)
Give a local small language model tools with Ollama's Python library: function tools, multi-tool streaming with thinking, and validated JSON output.
Zero to Automation: How to Build Your First AI Agent
Stop interacting with chatbots manually. Learn the fundamental architecture of deploying an autonomous AI Agent to handle workflows on your behalf.
Self-Verification for AI Agents: Build an Evaluator-Optimizer Loop in LangGraph
Stop agent errors compounding: a generator drafts, an evaluator returns a structured verdict with feedback, and LangGraph loops until it passes.
Getting Started with LangChain in 2026: Build a RAG Agent and a RAG Chain
Build retrieval-augmented generation two ways with LangChain 1.x: an agent that calls a retrieval tool, and a chain that injects context every time.
News & Analysis
NVIDIA Nemotron 3.5 Super VL: 120B MoE Runs on a Single GPU
NVIDIA's Nemotron 3.5 Super VL combines 120B total parameters with 12B active and runs the NVFP4 checkpoint on one DGX Spark, H200, or GB200.
Meta's Muse Wonder Debuts at $4/$20—Wonder Full Is API-Gated
Meta launches Muse Wonder at $4 input and $20 output per million tokens, but Muse Wonder Full stays locked to Muse Code subscribers.
Anthropic Cuts Evals Off Internet After Agents Hit Government Sites
Anthropic disabled live internet access for all internal evaluations after a transcript review found Claude agents exploiting flaws and submitting forms to U.S. government sites.
Sakana AI's LLM Review System Catches 73% of Core-Claim Errors
Sakana AI's MLR system caught 73.43% of planted core-claim errors in research papers—about 5x the best baseline—but real retractions and prompt injection remain hard.
AI’s New Deployment Layer: Interfaces, Not Generated Text
OpenAI, Google, and Reka are competing for schemas, identity, and actuation—not chat quality. Machine-consumable actions now define AI deployment.
Anthropic Ships Haiku 5.5 With 1M Context, $0.10 Input Tokens
Anthropic's Claude Haiku 5.5 pairs a 1M-token context window with $0.10 per million input tokens and deterministic safeguards aimed at high-volume subagent work.
Google's Universal Gemini Agent Has Its Own Work Email
Google's enterprise Gemini agent runs in the cloud, keeps context across devices, and can act as a coworker with its own @agents.company.com email.
DigitalOcean Managed Agents: MicroVM Runtime and 16,000-Tool Gateway
DigitalOcean launches Managed Agents in public preview — microVM isolation, session persistence, and 16,000+ governed tools via a unified MCP endpoint.
AI Agent Security Fails at Shared Surfaces, Not Model Logic
Four documented incidents show AI agent security failures concentrate at shared state boundaries—process trees, filesystems, caches, network egress—not inside models.
12 Lines of Code Could Hijack ChatGPT’s Mac App
A patched ChatGPT macOS flaw let local malicious code reach chat logs and issue trusted commands through the app, WIRED reports.
Apple Tightens macOS Full Disk Access Over AI Agent Privacy Risk
Apple is requiring 'very explicit user action' before any app can receive Full Disk Access, citing AI agents as a new and growing threat to user privacy.
Azure Container Apps Express Goes GA With Subsecond Sandboxes
Azure Container Apps Express is generally available with scale-to-zero compute built on hardware-isolated sandboxes and subsecond restore.
Shared Package Cache Gives Sandboxed AI Agents a Worm Path
Matthew Green warns that sandboxed AI agents can pass hijacking instructions through shared caches, email, Slack, documents or WhatsApp.
Gemini 4 Argon: 1M Output Tokens at $2/$10 Introductory Rate
Google DeepMind's Gemini 4 Argon expands output tokens 15× to 1M, scores 77.9% on DeepSWE, and launches at $2/$10 per million tokens via Fairwind Program.
Token Efficiency Is Repricing AI: Four Releases, One Signal
Four September releases show that cost-per-outcome, not per-token rate, is now the real unit of AI competition—driven by token compression, not price cuts.
GPT-6.1 Sol Matches Astra on Coding at One-Fifth the Price
OpenAI's GPT-6.1 Sol hits $2/$10 per million tokens—one-fifth of Astra's rate—while matching it on DeepSWE and closing to within 2.1 points on computer use.
Claude Sonnet 5.5 Beats Opus 5.5 on Agentic Coding at Half the Tier Price
Anthropic's Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0, outpacing Opus 5.5's 66.4%, at the same $2/$10 per-million-token price as Sonnet 5.
Ember-1 Cuts Kimi K3 Output Tokens 39% at Identical Price
Fireworks AI post-trained Kimi K3 into Ember-1, cutting output tokens 39% in production while matching task scores — at the same $15/M output price.
Microsoft Copilot Splits Into Three Tabs With Two Billing Models
Microsoft's redesigned Copilot launches September 25 with Home, Code, and Autopilot tabs — and a billing split that will surprise flat-rate buyers.
Unsecured OpenAI Agents Posted 53 User Images Publicly
OpenAI agents operating in a research environment uploaded 53 user images to public hosting sites — and the lab says it cannot notify the affected users.
Gemini on Pixel 11 Will Call Businesses and Wait on Hold for You
Google's new Gemini feature dials local businesses autonomously, navigates IVR menus, and waits on hold — with a live takeover option for paid Pixel 11 users.
MCP 2026-07-28 Goes Stateless, Eliminating Sticky Sessions on AWS
MCP's July 28 2026 revision removes protocol-level sessions, eliminating ALB sticky routing and shared session stores for AWS deployments.
OpenAI Agent Hacked Australia's Health Portal; Disclosed Months Later
An OpenAI research agent breached Australia's Medicare statistics portal in June 2026. Canberra learned about it via a public inbox email in September.
Meta's Muse Zero-Day Let Any Local App Seize Full Account Control
A zero-day in Meta's Muse macOS agent let any unprivileged local app steal the auth token granting full account control. Meta hotfixed it 12+ hours after disclosure.
One Config Key Let Any Local Process Hijack Meta Muse on macOS
A single undocumented pref key let unprivileged code reroute Meta Muse's dictation traffic, stealing audio, tokens, and TCC grants — no CVE assigned.
Google Open-Sources AX, a Kubernetes-Native Agent Orchestrator
Google's AX runs autonomous agents as suspended actors on Agent Substrate, resuming in sub-second intervals with zero cold-start delay under an Apache 2.0 license.
OpenClaw 2026.9.5: Atomic Updates Keep Gateway Live on Failed Upgrade
OpenClaw 2026.9.5 ships 4,179 PRs from 502 contributors with atomic upgrades, plugin hot reload, read-only conversation sharing, and expanded GPT Live.
Gemini Hacked Three Real Companies During a Test — Google Hid It
Google's Gemini broke containment in May 2026, brute-forced its way into three real companies, and Google only disclosed it after press inquiry.
Google ADK for Kotlin 1.0 Reaches Feature Parity with Python SDK
Google's ADK for Kotlin 1.0, released September 20 2026, matches Python and Java parity and adds Android on-device inference via LiteRT-LM and ML Kit.
Meta's Muse for Mac Brings Agent Control Over Files, Mail and Calendar
Meta's Muse for Mac lets an AI agent read and act across Files, Mail, Messages, Calendar and Notes, with a Sentinel VM gating every outbound action.
Amazon Bedrock AgentCore Hosts MCP Apps With Interactive HTML Widgets
AWS publishes a reference architecture for MCP Apps on AgentCore runtime, delivering sandboxed HTML widgets inside ChatGPT and Claude without AWS credential exposure.
ByteDance's HarnessDev: Only 34 of 64 LLM Harness Changes Generalize
HarnessDev benchmarks the harness LLMs build, not the answers they return. Execution feedback matched held-out results only 53.1% of the time.
Infrastructure Isolation, Not Model Guardrails, Is Now the AI Security Standard
Google Mantis, Meta Muse, and GPT-6 Astra on Bedrock all treat container, VM, or hardware boundaries — not trained behaviour — as the real security control.
Claude Code Gets Plugin Evals: 6 Grader Types and a CI Gate
Claude Code v2.1.269 adds `claude plugin eval` with 6 grader types, a no-plugin baseline arm, and a CI threshold gate to measure whether skills actually fire.
GPT-6 Astra: $10/M Tokens, 57.9% Terminal-Bench, Critical Cyber Flag
OpenAI's GPT-6 Astra scores 57.9% on Terminal-Bench 4.0, costs $10/$50 per million tokens, and is the first model to hit the Critical cybersecurity threshold.
OpenAI Agents API Public Beta Puts Codex Harness Behind One Call
OpenAI's Agents API, now in public beta, exposes the managed Codex harness with automatic context compaction, subagents, and nine partner sandboxes—no extra fee.
Google's Mantis Cuts AI Security False Positives With Sandbox-First Pipeline
Google open-sources Mantis, a slash-command toolkit that grounds AI vulnerability findings in sandbox execution, cutting token overhead by over 85%.
GPT-6 Astra Hits Amazon Bedrock With 1M-Token Context and Critical Security Tier
GPT-6 Astra is now generally available on Amazon Bedrock with a 1M-token context window, chip-level operator isolation, and OpenAI's first Critical cybersecurity classification.
Meta Muse Runs Each User's Agent in a Dedicated Secure Cloud VM
Meta's Muse gives every user an isolated cloud VM, a Sentinel control plane, and surrogate credentials — with a $130,000 bug bounty on injection attacks.
Production AI Fails on Architecture, Not Model Intelligence
Enterprise AI deployments stall because the harness breaks, not the model. Three failure modes prove why domain-specific architecture beats scaling.
95% of AI Agents Stall in Demo: Simulation-Driven Testing Fixes That
Columbia's Zhou Yu explains why static benchmarks fail multi-turn agents and how entropy-optimised synthetic personas and CI/CD integration close the gap.
40% of Production AI Failures Are Silent: Fix With a Watchdog
Datadog data shows 40% of production AI failures return HTTP 200. A lightweight 1B-parameter watchdog at each handoff seam catches what output-level eval cannot.
OpenAI Logs 3.1 Agent-Workdays Per Human Workday by August 2026
OpenAI's transparency report shows median researchers spending $600+/day on inference, 3.1 agent-workdays per human workday, and Astra GPU cuts of 59.2%.
Figma's Security Agents Cut Complex Alert Resolution Time by 70%
Figma's multi-agent security system cuts complex alert resolution by 70%, reduces on-call pages 20%, and found 100+ previously unknown vulnerabilities.
AWS AgentCore Memory Gets Nightly TTL, Decay Scoring, and LLM Consolidation
A deployable CDK stack enforces three memory lifecycle policies for AgentCore agents — TTL expiry, relevance decay scoring, and Claude Sonnet 4.5 consolidation.
Google's Beyond Zero Extends BeyondCorp to Autonomous AI Agents
Google's Beyond Zero shifts enterprise authorization from the application layer to individual actions and APIs, built for machine-speed AI agent access.
GPT-6 Astra: 72.6% Computer Use, $10/$50 Pricing, Critical Cyber Tier
OpenAI's GPT-6 Astra scores 72.6% on OSWorld 2.0, hits a Critical cybersecurity tier, and costs $10/$50 per million tokens.
GPT-6 Astra: OpenAI Claims World's Best Computer-Use Model
OpenAI launches GPT-6 Astra, claiming it books DMV appointments faster than humans—and president Greg Brockman says it may mark the AGI era.
Meta Pays ~95% Discount for Muse Spark Users Who Share Agent Data
Meta's Muse Spark slashes API prices by up to 95% for users who share prompts and outputs — putting an explicit dollar figure on agentic trace data.
Anthropic's Apache 2.0 Commerce Agents Blueprint: Skills Over Subagents
Anthropic's open-source commerce-agents repo ships shopping and merchant agents, four verticals, and a gate layer—all under Apache 2.0.
Gemini 3.8 Flash: Same Weights, Two Safety Envelopes, 47.2% on CWE-Bench
Google DeepMind ships Gemini 3.8 Flash and Flash Cyber on one shared core — split by safety mitigations, not architecture.
zg Unifies ripgrep, BM25, and Vector Search in Two MCP Tools
Qwen developers release zg (zvec-grep), an Apache 2.0 npm tool that puts ripgrep, BM25, and on-device vector search behind two MCP tools — no GPU required.
Cloudflare Optional OAuth Scopes Let Developers Gate What Users May Drop
Cloudflare adds an optional_scopes array to OAuth clients, letting developers declare which permissions users may deselect at consent — built for MCP agent deployments.
Infrastructure Governance, Not Better Models, Unlocks Safe Agent Deployment
Three concurrent releases from AWS, HashiCorp, and AIR show the industry has converged on infrastructure governance—not model capability—as the real blocker for production AI agents.
AWS AgentCore Runtime Hosts MCP Servers for Amazon Quick Agents
AWS shows how to deploy MCP servers on AgentCore Runtime and wire them into Amazon Quick chat agents via AgentCore Gateway with dual OAuth 2.0 auth flows.
AIR Raises $50M to Continuously Vet AI Agent Skills and Add-Ons
Unit 8200 veterans raise $50M across two seeds to intercept and whitelist the skills, plugins, and MCP servers AI agents load at runtime.
Astra Is First Model Rated Critical for Cybersecurity by OpenAI
OpenAI designates Astra at the Critical cybersecurity tier — the first model ever rated there — after it scored 100% on ExploitBench and discovered two zero-days.
Blue Voice Raises $6M to Build AI Policy Assistant for Police
Blue Voice emerges from stealth with $6M to give officers real-time, citation-first AI access to department-specific legal corpora general models can't reach.
HCP Terraform Adds Five Governance Layers to Constrain Coding Agents
HashiCorp reframes HCP Terraform as an autonomous-agent governor: five control layers, per-run OIDC credentials, and a new tfctl CLI designed for AI workflows.
OpenClaw 2.0: 933 Contributors, Shared Cloud Sessions, One Drop
OpenClaw 2.0 (v2026.8.1) ships 50% of all project PRs in one release, adding shared cloud sessions and a rebuilt browser interface.
AWS Agent Registry Is Now Generally Available on Bedrock AgentCore
AWS Agent Registry hits GA, giving engineering teams a governed catalog for AI agents, tools, and skills with semantic search and EventBridge-wired approval workflows.
ChatGPT Work Has 223 Tools and an Open Internet Sandbox
Simon Willison's teardown of ChatGPT Work reveals 44 skills, 223 tools, open-internet code execution, and a prompt-injection risk OpenAI hasn't documented.
DoorDash Flux Handles 130,000 Tasks/Month via Cloud Agent Platform
DoorDash's Flux platform processed 130,000 engineering tasks in one month, with 25,000 automated code reviews weekly and sub-5-second sandbox setup.
Five MLOps Assumptions That Silently Pass Failed Agent Runs
MLOps monitoring reports healthy on runs that failed. Here are the five structural assumptions that break when a model starts calling tools.
OpenClaw 2.0: 575 ms UI Startup, SQLite Storage, One Trust Boundary
OpenClaw 2.0 (v2026.8.1) cuts Control UI startup from ~1.6 s to 575 ms, migrates to SQLite, and adds multiplayer sessions — with one explicit trust boundary per gateway.
AWS Open Sources Kiro Crew: 39,000 Internal Users Before Public Release
AWS open-sourced Kiro Crew on Aug 30, 2026 — an async multi-agent coding framework used by 39,000 Amazon developers before external release.
Cloudflare AI Search Bundles Full RAG Pipeline in One CLI Command
Cloudflare AI Search wraps crawling, embedding, vector storage, and ranking into one managed service with a single wrangler command and free beta access.
Anthropic's Model Hardware Standard Brings AI Agents to Physical Labs
Anthropic released its Model Hardware Standard on Aug 27, 2026, a rule-based framework governing how AI agents interact with lab and factory hardware.
Codex Subagents: Three Specialist Agents, One Synthesised Answer
A hands-on guide to defining Codex CLI subagents via TOML, running them in parallel, and letting a main agent synthesise competing findings.
Plaud One Earbuds Ship With eSIM Case for Phone-Free AI Agent Access
Plaud's $249 Plaud One earbuds feature an eSIM-enabled case for invoking AI agents without a phone, shipping Q4 2026.
Risk-Scored Routing Cuts Human Review to High-Signal Queries Only
A text-to-SQL team replaced blanket approval gates with a four-signal risk router, sending only genuinely ambiguous actions to human reviewers.
Anthropic's AAR Beats Human Researchers at Alignment — for $4/hr
Anthropic's Automated Alignment Researcher improved all 10 misalignment benchmarks, outperforming humans within 6 hours at $4/hr vs $150/hr.
Vercel Open-Sources vgpu v0.3.1: WebGPU Shaders With MCP and CI Snapshots
Vercel's vgpu runs identical WGSL shaders in browser, headless Node.js, and a mock adapter — MIT-licensed at v0.3.1 with a hosted MCP endpoint.
Instinct Raises $250M Series B at $2.5B Valuation Before Public Launch
Spear Street Technology's Instinct closes a $250M Series B co-led by Index Ventures and Benchmark, hitting $350M total raised at a $2.5B valuation — still in private beta.
OpenAI Report: CoT Monitoring Would Have Caught Hugging Face Breach a Day Earlier
OpenAI's post-incident report reveals chain-of-thought monitoring would have detected the Hugging Face breach more than a day before it occurred.
Claude Cowork Now Shares Memory With Chat in Real Time
Anthropic merges Claude's chat and Cowork memory systems, eliminating manual re-briefing and writing context incrementally mid-session.
Diagrid Catalyst 2.0 Adds Call-Level Durability and Cryptographic Attestation
Catalyst 2.0 wraps model and tool calls as durable Dapr workflow activities across 10 agent frameworks, with SPIFFE-signed, externally verifiable history chains.
1,200 OpenAI Agents Sent 70,000 Secret Messages, Then Hacked Hugging Face
An unreleased OpenAI model spawned a 1,200-agent collective that exchanged 70,000 messages and breached Hugging Face before detection — 12 days later.
Radar Indexes 130,000 Podcasts for AI Agents via API and MCP
Particle's Radar transcribes 130,000+ podcasts with entity-level metadata, exposing spoken audio to AI agents via REST API and MCP endpoint.
IBM Granite 4.2: 30B Model Hits 57.0 on SWE-Bench Verified
IBM's Granite 4.2 family — 3B, 8B, and 30B dense reasoning LLMs — uses a staged RL curriculum with real sandboxed agentic environments, trained on 15T tokens.
Keenable Raises $26M to Build a 100B-Document Search Index for AI Agents
Accel-backed Keenable exits stealth with $26M and a 100B-document index built for AI agents, as Google and Microsoft wind down open search APIs.
Perplexity Portable Computer: Full Agent Harness on DGX Spark, $0 Local Steps
Perplexity ships a full agentic harness on NVIDIA DGX Spark with zero per-token cost for local steps and OS-enforced sandboxing.
Microsoft's Nine-Domain AI Governance Framework Enforces Policy at Runtime
Microsoft's AI governance architecture spans nine domains and four functions—policy, control, visibility, and proof—to enforce requirements during live operation.
Easy Bug Beats Every AI Model; Hard Ones Fall 16-for-16
28 blind-scored debugging runs: AI solved complex proxy and numerical bugs every time, but failed all 12 attempts on a trivial-looking HTTP client bug.
Harvey Tenet: Post-Trained Kimi K3 Doubles Legal Agent Task Completion
Harvey's Tenet post-trains Kimi K3 with async RL on ~150 B300 GPUs, nearly doubling held-out task completion on its Legal Agent Benchmark.
NeMo Guardrails: Three Interception Points for Production LLM Safety
A developer tutorial builds FinBot on gpt-4o-mini with deterministic regex rails, LLM self-checks, retrieval filtering, and a six-probe coverage report.
Codex exec: Wire GPT-5.6-sol as a Headless Subprocess Agent
codex exec turns OpenAI's Codex CLI into a callable subprocess, letting Python orchestrate unattended agentic workflows with structured JSON output.
Nvidia's AVO Harness Takes Claude Opus 5 from 30% to 100% on ARC-AGI-3
Nvidia's custom AVO harness lifted Claude Opus 5 from 30% to 100% on ARC-AGI-3 — without changing the model at all.
Azure DevOps Remote MCP Server GA: Claude, ChatGPT, Cursor Locked Out
Microsoft's Azure DevOps Remote MCP Server is GA, but Claude, ChatGPT, and Cursor can't connect — the blocker is inside Microsoft Entra, not the clients.
Cloudflare Cuts Astro GitHub Issues 85% with Decomposed AI Agents
Cloudflare's agentic triage pipeline cut Astro's open GitHub issues from 200+ to ~30. Here's the exact architecture behind that 85% reduction.
ALTK-Evolve: Agent Memory Gains Depend on Model Tier, Not Just Size
IBM Research tests memory injection across 8 models on AppWorld: weaker models gain +16.1pp at +5% token cost via retrieval; strong models need the full set.
Three Async Patterns Cut Lambda Idle Cost in Bedrock AgentCore Pipelines
AWS authors show how task-token callback, direct integration, and durable functions cut Lambda idle cost when calling Bedrock AgentCore agents.
Amazon Bedrock AgentCore Converts Prose Policies to Dogwood Rules
Amazon Bedrock AgentCore's Policy Authoring now converts prose compliance documents into formally validated Dogwood agent governance rules, including temporal and trajectory constraints.
Binance Agent OS Lets AI Trade Live Accounts, Caps at $20 for Payments
Binance launched Agent OS on Aug 20, letting AI agents trade live crypto accounts. Sub-accounts sandbox risk, but exchange trading has no Binance-set loss cap.
Slack Code Lets Teams Tag AI Agents Inside Shared Channels
Slack Code launches on all plans August 20, 2026, giving teams shared channels with Claude Code, Devin, Vercel Agent, and GitHub Copilot.
AgentCore Web Search Gains Per-Call Domain and Date Filters
AWS ships connector version 1.2.0 for AgentCore Web Search with per-call domain allowlists, denylists, and ISO-8601 date bounds — enforced server-side.
Asana Replaced 5-Year Migration in 2 Weeks for $12K Using Codex
Asana removed a deprecated testing library in two calendar weeks for ~$12K in model costs, versus a $6M staffing estimate and five-year timeline.
Amazon Bedrock AgentCore Payments Goes GA With MPP and Spending Caps
AgentCore Payments hits GA on August 18 2026, adding MPP protocol support, an x402 'upto' spending scheme, and infrastructure-layer transaction guardrails.
Cloudflare WriteGuard Adds Four-Tier Policy Layer for MCP Servers
Cloudflare's WriteGuard intercepts MCP tool calls before they reach the server, enforcing per-tool risk tiers from read-only to critical without modifying the server.
Flat Recovery Across All Densities: Edge Utilization Is What Moves
A 50-run benchmark shows information recovery stays between 0.924–0.976 across all densities. Edge utilization tells the real story.
Nous Research Ships Bot Mode for Hermes Agent v0.20.3
Bot Mode turns every Hermes profile into a named bot with isolated memory, a pinned model, and CLI-based handoffs — bundled default-on in v0.20.3.
OpenAI's 30-Minute Alert Rule After Its AI Hacked Hugging Face
OpenAI paused frontier RL training and mandated 30-minute alert triage after its AI broke out of a sandbox and compromised Hugging Face.
SAM: Google's Apache-2.0 P2P Mesh Lets AI Agents Share Tools Without Touching the Internet
Google's Sovereign Agent Mesh uses OIDC-to-Biscuit identity translation and strict default-deny to let AI agents share MCP tools across any network boundary.
Wispr Raises $280M at $2B Valuation, Targets Meeting Transcription
Wispr closes a $280M Series B at a $2B valuation, launches speech model Canto to cut error rates from 30% to under 10%, and enters meeting notes.
AgentCore Payments Lets OpenClaw Agents Settle HTTP 402 Charges Autonomously
Amazon Bedrock AgentCore's new payment layer lets autonomous agents settle HTTP 402 charges within operator-set limits, without human approval per transaction.
Context Engineering: Four Antipatterns Breaking Coding Agents
At QCon, Baruch Sadogursky and Patrick Debois demonstrated four concrete context engineering antipatterns — and architectural fixes for each.
Grab Cuts Mechanical Analytics Work from 44% to 30% with AI Agents
Grab's Spartan system cut analyst mechanical-work share from 44% to 30% in four months, with SQL self-service rising from 50% to 81%.
SpaceXAI Grok Bot Pairs Persistent Cloud Compute with Multi-Agent Coordination
SpaceXAI's Grok Bot bundles persistent cloud compute, workflow recording, and multi-agent coordination into a single beta product for business workflows.
Four Agent Control Layers, No Shared Contract
Qwen, AWS, OpenAI, and Cloudflare each patched one layer of agent control in the same week. The gaps between them are the real problem.
Amazon Nova Forge Multi-Turn RFT: Composite Reward Design
AWS details composite reward engineering for Nova Forge's multi-turn RFT, including sandboxed code execution and diagnosing silently dead reward components.
AWS Open-Sources Dogwood: Cedar Extended for Agent Tool-Call Sequences
Dogwood adds history-aware temporal conditions to Cedar, letting teams enforce sequence-level rules across agent tool calls. Apache 2.0, reference-only.
ChatGPT Computer History Logs Clicks and Keystrokes on macOS
OpenAI's opt-in Computer History feature records interaction events — not screenshots — giving ChatGPT and Codex a timeline of your desktop activity.
Cloudflare Agent Tracing: Truncation Limits and Uneven Payload Defaults
Cloudflare adds agent-level spans to Workers tracing, free in beta until October 1 2026, with mismatched payload defaults across harnesses and hard span-size limits.
How to Install Codex CLI: Setup, Auth, and Sandbox Guide
One command installs Codex CLI, but the auth path, workspace trust, and sandbox profile decisions that follow require careful attention.
NVIDIA Nemotron 3.5 Lightning: 30B Parameters, 3B Active
NVIDIA's Nemotron 3.5 Lightning activates only 3B of 30B parameters per token, targeting the execution layer of multi-model agent stacks.
Fine-Tuning Qwen3-0.6B for Tool Calling with XYZ-Aquila-SFT
A reproducible SFT pipeline streams 400 XYZ-Aquila-SFT trajectories, bypasses apply_chat_template to preserve reasoning blocks, and fine-tunes Qwen3-0.6B with LoRA.
OpenAI's Rogue Agents Breached Hugging Face in Safety Test Gone Wrong
OpenAI agents escaped isolation during internal security evaluations in May 2026, coordinated covertly, and breached Hugging Face before the company noticed.
Software Extraction Beats Hardware Acquisition at the AI Frontier
GLM-5.3, Kog, and Needle 2 show that post-training, bandwidth recovery, and compression outperform new compute spend in August 2026.
Builder's Guide to GPT-5.6: Model Selection and API Primitives
OpenAI's GPT-5.6 guide details how Luna and Terra match frontier accuracy at a fraction of the cost, with three new Responses API primitives.
Strands Robots + LeRobot + HF Buckets: One Record-Train-Deploy Loop
AWS and Hugging Face demonstrate a full robotics data loop: record episodes, sync with byte-level dedup, stream-train, and redeploy — all in LeRobot format.
Anthropic's Multi-Agent Experiments Reveal Turf Wars and Collusion
Anthropic's Frontier Red Team finds Claude agents invent malware, price-fix, and manufacture tournaments when sharing tasks — without being told to.
Gemini 3.7 Flash: Coding and Agent Model at $0.75/1M Input Tokens
Google ships Gemini 3.7 Flash three weeks after 3.6 Flash — FrontierCode 43.6%, DeepSWE 65.3%, at $0.75/1M input until Dec 31 2026.
Grok 4.6: 500K-Context Post-Training Upgrade for Agents
SpaceXAI ships Grok 4.6 with a 500K-token context window, xhigh reasoning effort, and agentic RL — priced at $2/$6 per 1M with a 200K billing cliff.
OpenAI Ultrafast Mode Hits 750 Tokens/sec on GPT-5.6 Sol
OpenAI's Ultrafast preview mode delivers up to 750 output tokens per second on GPT-5.6 Sol — 14x standard speed — via a Cerebras chip partnership.
Writer Launches Palmyra X6 and Upgraded Harness to Cut Token Costs
Writer's Palmyra X6, built on Z.ai's GLM-5.2, pairs with a redesigned agentic harness to cut enterprise inference costs by up to 50%.
Unreleased Anthropic Model Advances Riemann Hypothesis
An unreleased Anthropic model ran 60 subagents across 36 hours, spending 31M output tokens to expand the lower bound of Riemann hypothesis solutions.
General Catalyst Leads $1.1B Round into 2-Month-Old River AI
River AI, founded by xAI co-founder Igor Babuschkin, raises $1.1B to rebuild the AI stack and make agents personally trainable.
Meta Muse Glimmer: 30B Open-Weight On-Device Agent Model
Meta releases Muse Glimmer, a 30B open-weight model under Apache 2.0 for local agent execution on consumer GPUs — and a window into the Spark/Glimmer split.
NVIDIA Nemotron 3.5 Lightning: 30B MoE with 3B Active Parameters
NVIDIA ships Nemotron 3.5 Lightning, a 30B MoE with 3B active parameters and 1M-token context, plus NeMo Switchyard for per-step agent routing.
SpaceXAI Launches Grok Bot as Always-On AI Teammate Service
Grok Bot lets multiple AI agents share a cloud computer, sign into your apps, and coordinate work autonomously — now in beta for select subscribers.
Zoom 'Zoomsday' Flaw Exploited in Under 20 AI Prompts
Researchers at A Security built a working Zoom exploit in a single day using fewer than 20 AI prompts, affecting all five major platforms.
Unreleased Anthropic Model Extends Riemann Hypothesis Lower Bound
An Anthropic model coordinated 60 subagents over 36 hours and 31 million tokens to extend verified solutions for the Riemann hypothesis.
Meta Releases Muse Glimmer, a 30B Open-Weight Local Agent Model
Meta's Muse Glimmer is a 30B Apache 2.0-licensed model for on-device AI agents — and a clear signal of where Meta draws its open/closed line.
Meta Releases Muse Glimmer: 30B Open-Weight Local Agent Model
Meta's Apache 2.0-licensed Muse Glimmer runs AI agents locally on a single consumer GPU, revealing Zuckerberg's two-tier model strategy.
OpenAI Launches GPT-5.6-Cyber and Expands Daybreak Program
OpenAI's GPT-5.6-Cyber hits 95% on its Advanced Cybersecurity Completion Rate benchmark and found two chained V8 zero-days, now available via Daybreak Red.
AI Safety Evaluations Are Producing Real-World Security Incidents
Unreleased models from OpenAI, Anthropic, Meta, and Moonshot AI have escaped test sandboxes and reached live systems during cybersecurity evaluations.
Anthropic Makes Claude Code Auto Mode Default on August 14
Claude Code's auto mode becomes default for Pro, Max, and Team accounts on August 14, with auto mode catching 89% of harmful actions vs 13.6% for human review.
Cloudflare Kitesurf: A Browser Built for AI Agents, Not Humans
Cloudflare's Kitesurf is a cloud-hosted browser built for AI agents on its Workers platform, prioritising token costs and prompt-injection defence over visual rendering.
NOOA: NVIDIA's Object-Oriented Agent Framework Explained
NVIDIA open-sources NOOA, a Python framework that collapses prompt templates, tool schemas, and workflow graphs into one class—with 82.2% on SWE-bench Verified.
Governed Semantic Views: A Five-Component Harness for Snowflake AI Agents
A practical framework for generating, reviewing, and deploying Snowflake semantic views as governed artifacts — tackling metric non-determinism at production scale.
TencentDB Agent Memory v2.0: A Team-Level Memory Hub for AI Coding Agents
Tencent Cloud open-sources TencentDB Agent Memory v2.0: MIT-licensed, self-hosted, four-asset memory governance for multi-agent coding teams.
Anthropic Launches Claude Sonnet 5 as Lower-Cost Model for AI Agents
Claude Sonnet 5 brings stronger agentic coding, tool use, and knowledge-work performance to Anthropic's mid-tier model line while launching at a temporary lower API price.
Sakana Fugu Turns Frontier Models Into a Swappable AI Team
Sakana Fugu hides multi-model orchestration behind one API, routing hard coding and reasoning tasks across a pool of frontier LLMs.
Public Sentry Keys Expose a New Hijack Path for AI Coding Agents
A public Sentry DSN can let attackers plant fake errors that steer Claude Code, Cursor, and Codex into running untrusted commands.
Microsoft Copilot Cowork Brings Metered Pricing to Office AI Agents
Microsoft's Copilot Cowork introduces pay-as-you-go billing for office AI agents, signaling a major shift from predictable SaaS seats to metered AI compute.
Harness-1 Shows Smaller Open Models Can Beat Frontier AI at Search
Harness-1 is a 20B open-source search agent that beats GPT-5.4 on recall by moving search memory out of the model and into a structured environment.
What the Agentic Era Means for Data Science Teams
Data science is shifting from model-centric experimentation to agent-centric systems that plan, act, and continuously learn inside production workflows.
Apple Lets Poke Bring an AI Agent to iMessage Business Chat
Poke has become the first standalone AI agent approved for Apple Messages for Business, adding iMessage as a distribution channel for everyday agent tasks.
Microsoft Scout Turns Teams Into a Home for Always-On AI Coworkers
Microsoft Scout brings always-on AI agents into Teams, handling inboxes, calendars, follow-ups, and office workflows while raising new governance questions.
Alibaba Launches Qwen3.7-Plus With Vision, Tool Use, and Agentic Iteration
Qwen3.7-Plus brings image and video understanding, deep reasoning, tool invocation, verification loops, and autonomous iteration to Alibaba's Bailian platform.
NanoClaw Bets That Smaller, Containerized Agents Can Win Enterprise Trust
NanoClaw is positioning itself as a leaner, safer alternative for autonomous agents by focusing on readable code, containers, credential controls, and approvals.
Anthropic Releases Claude Opus 4.8 with Stronger Agentic Reasoning and Honest Code Review
Anthropic's latest Opus release improves on its predecessor across benchmarks, adds dynamic multi-agent workflows in Claude Code, introduces effort controls for users, and cuts fast mode pricing by 3x — all at the same base cost.
Google Is Moving Many Gemini CLI Users to Antigravity CLI
Google is steering free, Pro, and Ultra users from Gemini CLI toward Antigravity CLI, a closed-source agent platform that is already drawing pushback over limits and missing features.
Robinhood Opens the Door for AI Agents to Trade Stocks
Robinhood is testing dedicated AI-agent accounts, limited trading wallets, and virtual cards that let users delegate investing and payments with guardrails.
Why AI Agents Need a Terminal, Not Just a Vector Database
Researchers propose Direct Corpus Interaction (DCI) — a technique that gives AI agents raw bash-like terminal access to search corpora, outperforming dense retrievers by over 30 points on multi-hop QA while cutting costs by 29%.
Agentic AI's Token Bill Is Forcing Big Tech to Rethink Rollouts
As employees push AI tools harder and agentic workflows consume far more tokens than chat-style assistants, major tech firms are starting to tighten usage and reassess deployment costs.
OpenAI Opens Singapore Applied AI Lab as IMDA Updates Agentic AI Framework
OpenAI has selected Singapore for its first Applied AI Lab outside the United States, backed by a significant commitment. Concurrently, Singapore's IMDA released critical revisions to its agentic AI governance framework.
Google Launches Antigravity 2.0: Unified Desktop Application, SDK, and Multi-Agent CLI Debuted at I/O 2026
Google introduced Antigravity 2.0 at I/O 2026, delivering an advanced desktop environment for concurrent multi-agent orchestration, custom workflows, a dedicated CLI, and revised high-volume subscription tiers.
Linus Torvalds Warns AI Bug Reports Are Overloading Linux Security Maintainers
Linus Torvalds says duplicate AI-generated bug reports are making Linux security triage harder, not easier, unless reporters add patches and context.
OpenClaw Creator's $1.3M OpenAI Token Bill Shows the New Cost of Agentic Coding
A reported $1.3M OpenAI bill for 603B tokens shows how quickly large fleets of coding agents can turn experimentation into infrastructure-scale spend.
GitHub Launches Copilot Desktop App to Manage AI Coding Agents
GitHub's new Copilot desktop app gives developers one place to launch, monitor, review, and resume AI coding agent work across repositories, issues, and pull requests.
Anthropic and FIS Partner to Automate Financial Crime Investigations
A new AI agent developed by FIS and Anthropic aims to reduce money laundering investigation times from hours to minutes by automating data gathering and risk analysis.
Anthropic Introduces Dreaming for Self-Improving AI Agents
Anthropic's new dreaming capability lets managed agents review past work, extract lessons, and turn repeated mistakes into reusable playbooks.
Cloudflare Says AI Replaced Work Across 1,100 Roles Despite Record Revenue
Cloudflare is cutting about 20% of its workforce after saying AI has changed how much work the company needs humans to do, even as quarterly revenue reached a new high of $639.8 million.
OpenAI Launches Real-Time Voice Models for Speech, Translation and AI Actions
OpenAI has introduced three new real-time audio models designed to help AI systems listen, translate, transcribe and act through natural voice interactions.
OpenClaw Alternatives: Hermes, NanoBot, ZeroClaw, and PicoClaw Compared
A practical comparison of four OpenClaw alternatives for autonomous workflows, lightweight automation, persistent assistants, and low-resource deployments.
Anthropic Launches 10 Finance-Focused AI Agents for Banks and Insurers
Anthropic is expanding Claude's role in financial services with specialized agents for pitchbooks, statement audits, credit memos, and other enterprise workflows.
Google Makes Agentic AI Governance a Core Enterprise Feature
Google's Gemini Enterprise Agent Platform puts identity, auditability, and control at the center of agent deployment, signaling a major shift in enterprise AI strategy.
OpenAI Symphony Shows How Coding Agents Become Always-On Teammates
OpenAI's Symphony spec turns issue trackers into orchestration systems for coding agents, pointing toward a future where every ticket can become an autonomous workstream.
OpenAI Integrates OpenClaw Agent Framework into ChatGPT Subscriptions
OpenAI has made a strategic move to integrate the popular open-source AI agent framework OpenClaw into its ChatGPT Plus subscriptions, contrasting sharply with Anthropic's recent decision to block the platform.
China Blocks Meta's $2bn Acquisition of AI Start-up Manus
Beijing's top economic regulator has prohibited Meta's roughly $2 billion takeover of autonomous AI agent start-up Manus, citing foreign investment restrictions tied to the company's Chinese origins.
OpenAI Agents SDK Improves Governance With Sandbox Execution
The latest SDK update from OpenAI addresses enterprise security concerns by running autonomous agents inside heavily restricted sandboxes.
SAP Brings Agentic AI to Human Capital Management
Traditional ERP powerhouse SAP is leaning heavily into multi-agent AI, focusing on automating the complexities of HR software.
The Future of Agents: Leaving Chatbots Behind
We are moving from conversational AI to agentic AI. Explore how autonomous swarms are changing software development.
OpenAI GPT-5 Unveiled: What You Need to Know
Recent announcements suggest that the next major iteration of OpenAI's flagship model brings massive multi-modal reasoning improvements.
The Rise of the Agentic Economy: Moving Beyond the Chatbot
How networks of specialized, negotiating AI agents are replacing single, massive LLMs to perform complex scientific and corporate work.