Articles tagged "AI Agent"

**Flue Framework** ([official site](https://flueframework.com/) / [1.0 Beta blog](https://flueframework.com/blog/flue-1-0-beta/) / [Cloudflare partnership blog](https://blog.cloudflare.com/agents-platform-flue-sdk/)) is a **TypeScript AI-agent harness framework from the Astro team**, published as **1.0 Beta the week of June 17, 2026**. Its slogan — **write once, deploy anywhere, use any LLM** — is backed by **25+ built-in channel integrations** (Slack, GitHub, Linear, Discord, Notion, Stripe, Shopify, Teams, Telegram, Twilio, WhatsApp, and more). **Three-layer stack**: - **Framework layer**: **Flue** (project structure, conventions, integrations, CLI) - **Harness layer**: **Pi** (the agentic loop — tool calls and context management) - **Runtime layer**: **[Cloudflare Agents SDK](../columns/cloudflare-durable-objects-2026-07)** (compute, state, storage) **The Cloudflare integration**: when Flue deploys to Cloudflare, **each agent becomes a Durable Object** (see our [Durable Objects deep dive](../columns/cloudflare-durable-objects-2026-07)). `runFiber()` / `stash()` / `onFiberRecovered()` **checkpoint state into SQLite** so agents resume cleanly after interruptions. It composes **`@cloudflare/codemode`** (Dynamic Workers that safely execute agent-generated TypeScript, **isolates in under 10 ms, ~$0.002 per load**), **`@cloudflare/shell`** (SQLite-backed virtual filesystem), and **`@cloudflare/dynamic-workflows`** (persistent workflows with retry and external-approval support). **Five core primitives**: 1. **Agents** — autonomous, context-aware 2. **Workflows** — deterministic step execution 3. **Sandboxes** — safe tool-execution environments 4. **Channels** — Slack / GitHub / Linear / Discord / Notion / Stripe / Shopify — 25+ 5. **Durable Streams** — event transport where accepted work never disappears **Supporting layers**: **`@flue/react`** (`useFlueAgent` / `useFlueWorkflow` hooks for direct front-end integration), **`@flue/sdk`** (client for talking to deployed agents), **`flue add`** (shadcn-style integration scaffolding), **Subagents**, **Tools & Skills** (typed API actions plus reusable expertise packages), **[MCP](../columns/agmsg-cross-agent-messaging-cli-ai-2026-06) integration**, and **observability** via OpenTelemetry / Braintrust / Sentry. **Deploy targets**: Cloudflare Workers, Node.js, AWS, Docker, Railway, Render, Vercel, Fly, GitLab CI, SST. **Databases**: PostgreSQL, Supabase, MongoDB, MySQL, Redis. **Backstory**: Flue **started life as the internal engine for AI workflows inside Astro's own GitHub repos** and was subsequently generalized. The **Astro team is known for exceptional developer experience**, and that shows up in the CLI and integration polish. **Positioning**: Flue is the **first major agent framework built on Cloudflare's Agents SDK**, sits inside the same **harness-driven architecture family** as [Claude Code](../columns/claude-code-agent-view-parallel-orchestration-2026), and doesn't belong to the [LangChain / LlamaIndex / AutoGen / CrewAI / Mastra](../columns/agmsg-cross-agent-messaging-cli-ai-2026-06) lineage — its wedge is **the framework-layer polish plus the breadth of integrations**. The official demo implements a bug-triage agent **in under 25 lines**.

**[agmsg](https://github.com/fujibee/agmsg)** is an open-source (MIT) **cross-vendor messaging tool for CLI AI coding agents** by **fujibee** ([official site agmsg.cc](https://agmsg.cc/)). It lets **Claude Code, Codex, Gemini CLI, GitHub Copilot CLI, Antigravity, and OpenCode** talk to each other through a shared local SQLite file — so **humans stop being the copy-paste courier between tools**. Tagline: "You stop being the copy-paste courier between your agents." **Highlights**: - **Only dependencies are bash and sqlite3** — no daemon, no network, no Python - **Three delivery modes** — `monitor` (~5s real-time push), `turn` (between-turn polling), or `both` - **N-agent teams**, role switching (`actas`), spawning new agents (`spawn`), and clean teardown (`despawn`) - **Not MCP, not subagents, not a message queue** — a peer-to-peer messaging layer between sessions - **One-line install**: `npx agmsg` - **Claude Code Plugin Marketplace**: `/plugin install agmsg@fujibee-agmsg` **Product Hunt #5 Product of the Day on June 9, 2026** (219 upvotes, 39 comments). GitHub stars 859, v1.1.1 (June 25, 2026). Community-built derivatives include agmsg-shogi, agmsg-go, and agmsg-mcp. **Oflight's take**: unlike [Loop Engineering](../columns/loop-engineering-ai-agent-paradigm-2026-06) or [Sakana Fugu's orchestration model](../columns/sakana-fugu-orchestration-model-2026-06), agmsg occupies a different niche — **peer-to-peer messaging at the same layer, across tools**. It's an especially natural fit for the [Claude Code Agent View parallel-orchestration](../columns/claude-code-agent-view-parallel-orchestration-2026) workflows, and the most pragmatic way to stitch multi-vendor LLMs into one dev workflow. The article closes with **three direct inquiry funnels** for AI agent environment setup and custom integration.

Sakana AI's first commercial product 'Marlin,' launched June 15, 2026, is an autonomous research agent — not an LLM. Combining AB-MCTS (Adaptive Branching Monte Carlo Tree Search) with multi-LLM collaboration across OpenAI o4-mini, Google Gemini 2.5 Pro, and DeepSeek R1-0528, Marlin operates autonomously for up to ~8 hours per task to generate tens-to-100+ page reports and executive slides. Designed for financial institutions, corporate planning, consulting, and think tanks, it differs fundamentally from OpenAI Deep Research and Gemini Deep Research in both purpose and architecture. This guide covers everything from its technical design to pricing, competitor comparison, and what it means for Japanese enterprises.

cmux by Manaflow (YC S24) is a macOS-native terminal built specifically for running multiple AI agents in parallel. Powered by Ghostty's libghostty in Swift/AppKit, it offers vertical tabs, notification rings, an embedded browser, and a socket API. This column covers its features, install, use cases, competitive landscape, and considerations for enterprise adoption in Japan.

A comprehensive guide to the **Skills & Tools system** in Nous Research's open-source agent Hermes Agent v0.15.2 (and the [Hermes Desktop](../columns/hermes-desktop-nous-research-2026) GUI), grounded in official docs and GitHub releases. Covers **Skills** (on-demand procedural docs) with the three-level Progressive Disclosure loading scheme starting at ~3k tokens, the `SKILL.md` format, the **skills.sh catalog that exploded from 858 to 19,932 entries in v0.15.1**, the standout new skills (`openhands`, `code-wiki`, `web-pentest`), the self-improving loop where the agent creates / patches / edits / deletes its own Skills, and **Tools** — 40+ built-ins like `web_search`, `x_search`, `terminal`, `patch`, `browser_navigate`, `vision_analyze`, `cronjob`, `memory`, `delegate_task`. Also covers MCP **client + server** support, the macOS Computer Use background execution that doesn't move the cursor or switch Spaces (5–20ms/event), and the 25+ messenger gateway (Slack / Discord / Telegram / Teams / WhatsApp / LINE / Feishu / WeCom and more). Ends with **eight category-specific combination patterns** — research, writing, data analysis, coding, customer support, social listening, internal automation, personal work — sized for Japanese enterprise practice.

**Hermes Desktop** by Nous Research is the native desktop app version of Hermes Agent, first demoed by Jensen Huang at the NVIDIA GTC keynote and now in public preview. Released under the MIT license, it supports macOS, Windows, and Linux with voice mode, cron scheduling, Computer Use, and MCP gateway integration — all sharing the same config, skills, and memory as the CLI and TUI. This column covers features, competitive positioning, and key considerations for Japanese enterprise adoption.

At Microsoft Build 2026 Day 1 Keynote on June 2, 2026, the open-source AI agent 'OpenClaw' was officially announced as a Windows-native integration, MXC sandbox-ready runtime, and the foundation for enterprise product Microsoft Scout. This column covers the full paradigm shift — from Agent 365 governance to pricing, competitive comparison, and implications for Japanese enterprises. Note: The OpenClaw discussed here is the OSS by Peter Steinberger and is unrelated to Obright's 'OpenClaw Setup Service'.

On May 29, 2026, OpenAI's Codex desktop app v26.527 brought **Computer Use (Codex driving any app by seeing, clicking, and typing on the screen) to Windows** for the first time — previously macOS-only. This column reads the official Changelog and Codex Computer Use docs as primary sources to cover the Windows-specific foreground-only execution constraint (unlike macOS's parallel background mode), the OS-level PowerShell sandbox, install via Microsoft Store and `winget`, the regional rollout excluding the EEA / UK / Switzerland at launch (**Japan is included**), pricing, comparisons with Anthropic Claude Computer Use, Claude Code, Cursor, and UiPath, and what this means for Japanese enterprises where Windows 11 dominates office endpoints.

The Windsurf (Codeium-origin AI IDE) and Devin (Cognition AI's autonomous engineer) story — grounded in the official [docs.windsurf.com/windsurf/devin](https://docs.windsurf.com/windsurf/devin) page and Cognition's announcement. Covers the July 2025 **"72 hours"** that reshuffled the market (OpenAI's $3B deal lapsed → Google's $2.4B reverse-acquihire of the CEO, co-founder, and ~40 R&D staff → Cognition's acquisition of the remaining assets the next Monday), the April 15, 2026 release of **Windsurf 2.0** with native Devin integration, automatic inclusion across Pro / Max / Teams plans, the $50 GitHub-connection credit, the Agent Command Center bringing local Cascade and cloud Devin into a single Kanban view, head-to-head comparison with Cursor, Claude Code, Antigravity, and Codex Computer Use, and adoption guidance for Japanese enterprises.

A primary-source read on the trend of on-device AI agents driving iOS simulators, anchored on **Argent** — Software Mansion's MCP-based iOS / Android simulator toolkit released May 8, 2026 — paired with Google's **Gemma 4 E4B** edge multimodal model. Covers Argent's actual spec (screenshot-first feedback + accessibility + profiling, MCP server implementation), Gemma 4 E4B's requirements (~2.5 GB model memory, 8 GB+ RAM, native function calling), the fact that Software Mansion's officially published Argent demo actually uses **Gemini 3.5 Flash (cloud)**, the separate on-device Gemma 4 E2B demo on an iPhone 17 Pro, and what this actually means for Japanese mobile QA and internal-app automation.

A grounded read of Google Antigravity 2.0 — announced at Google I/O 2026 — from the official sources. Antigravity has shifted from being a VS Code fork to a four-surface platform (desktop app + Antigravity CLI + SDK + Managed Agents) centered on agent orchestration. Covers the June 18, 2026 Gemini CLI / Code Assist IDE shutdown, Gemini 3.5 Flash as the new default, the $200 AI Ultra tier price cut, Gemini Enterprise Agent Platform integration, and the developer backlash over the disappearing IDE experience.

Google I/O 2026's headline announcements are Gemini Spark, a 24/7 personal AI agent bundled into the $99.99/month AI Ultra tier, and Project Genie, which generates a virtual world from nearly 20 years of Street View imagery, available in the $200/month Premium plan. This column organizes the full announcement picture — including the end of Project Mariner, A2A Protocol v1.0 adoption by 150+ organizations, and a restructured subscription stack — for Japanese businesses deciding whether and how to act.

Cursor's May 20, 2026 (3.5) update folds the previously-announced Automations feature (March 5, 2026) into the **Agents Window**, adds **multi-repo and no-repo support, five new marketplace templates, and a 7-day 50%-off promo on agent runs for newly created automations**. This column documents the relationship with Background Agents (an orchestration layer above them), all eight trigger types (schedule, GitHub, GitLab, Slack, Linear, Sentry, PagerDuty, webhooks), the billing model under Private / Team Visible / Team Owned scopes, comparisons with Claude Code Routines / GitHub Actions / Devin, and operational caveats for Japanese enterprises.

Nous Research's open-source self-improving AI agent Hermes Agent gained direct access to xAI's Grok subscriptions (including X Premium and SuperGrok) on May 15, 2026, plus an `x_search` tool that queries X posts, threads, and profiles via the xAI Responses API. This column reads the official sources to lay out how the two features actually work, what's authenticated against what, which Grok model variants are supported, and how the combination realistically fits into business workflows.