Sales Claw Blog
Notes from the AI sales floor.
English versions are concise summaries of the Japanese originals. For the full deep-dive of any article, follow the link at the end to the Japanese version.
A New Way to Build Sales Decks with AI — 3 Steps: Structure in Claude, Images in GPT Image 2, Recreate in the PowerPoint Add-in
Bottom line: deck quality is decided not by "which tool you use" but by "how you chain the three." Structure in Claude, images in GPT Image 2, recreation in the PowerPoint add-in—following the real process of building the Sales Claw deck, with screenshots.
中澤 圭志13 min
What Is a Subagent? — How Claude Code Lets AI "Divide the Labor," Explained for Everyone
Bottom line: subagents are not about "building one smarter AI" but about "preparing several AIs with different roles and assigning the right one." From the real experience of failing when I made one AI do everything, explained so non-experts can follow.
中澤 圭志13 min
Claude Opus 4.8 Became an "Honest AI" — Benchmarks, New Features, and Cost, Explained for Non-Experts
On May 28, 2026, Anthropic released its flagship model Claude Opus 4.8 — less than two months after 4.7. The real story this time is not raw intelligence but "honesty": the chance it silently lets flaws in its own code through dropped to roughly a quarter. This article walks general readers through the actual numbers across six official benchmarks, the cost trap hiding behind "flat pricing," the three new features (Fast Mode / Dynamic Workflows / Effort Control), and how individuals, SMBs, and enterprises should get started — framed with the metaphor of a new hire growing into a senior colleague.
中澤 圭志15 min
The Month AI Became Something You Ship — KPMG×Claude, OpenAI DeployCo, Cohere×Aleph Alpha, and Canada's Ruling, Explained for Non-Experts (May 2026)
May 2026 brought four simultaneous events in enterprise AI: KPMG's 276K Claude rollout, OpenAI's $4B DeployCo, the Cohere × Aleph Alpha $20B merger, and Canada's PIPEDA ruling. This article unpacks what "AI becoming something you ship" means, a five-item preparation list for Japanese enterprises, four risk surfaces, and an OSS self-defense path for SMBs.
中澤 圭志16 min
OpenAI Codex Became "the Colleague Who Runs Through the Night" — Goal Mode GA, Locked Computer Use, Appshots, and Plugin Marketplace Explained for Non-Experts
In one week, OpenAI Codex turned into "the colleague who runs through the night." Goal Mode GA, Locked Computer Use, Appshots, the enterprise Plugin Marketplace, and CLI 0.132–0.134 — covered for general readers, with three real risks (cost runaway, over-permissioning, legal alignment) and the Sales Claw safety perspective.
中澤 圭志14 min
Anthropic Overtakes OpenAI for the First Time — The May 2026 "Changing of the Guard" in Enterprise AI, Explained for Non-Experts
On 2026-05-13 the Ramp AI Index shows Anthropic 34.4% vs OpenAI 32.3% — the first flip. Anthropic ~4x growth in a year; Claude Code at $2.5B run rate / ~4% of public GitHub commits; Uber burned its 2026 AI budget in 4 months; 79% pay both. This piece, for non-experts, covers what the flip means, three risks, and how SMBs and indie devs can run a multi-model strategy for ¥10,000–¥30,000 a month.
中澤 圭志14 min
Claude Compliance API Launches — Anthropic Wires Claude into 28 Security Tools, Bringing AI Under the Same Governance Frame as Every Other SaaS
On 2026-05-21 Anthropic announced the Claude Compliance API and 28 security / compliance partner integrations. Claude Enterprise conversations + attachments and Claude Platform admin events can now stream into Datadog, Cloudflare, Wiz, Microsoft Purview, Okta, and more. Written for non-expert readers: the shadow AI context, the 8-category taxonomy, enablement flow, risks, and a Sales Claw perspective.
中澤 圭志15 min
Claude Code v2.1.149 & Gemini CLI v0.43.0 Same-Day Update Explained for General Readers — "Cost Visibility" Meets "Surgical Edits"
On 2026-05-22 the two terminal-AI giants updated the same day. Claude Code v2.1.149 makes cost breakdowns visible via /usage; Gemini CLI v0.43.0 steers the AI to fix only what is needed via surgical edits. Written for non-expert readers: the "explainable / non-destructive" direction, a 5-minute setup, and remaining risks.
中澤 圭志13 min
Google Antigravity 2.0 Explained: The 'Hand Multiple Jobs to AI at Once' IDE, for Non-Technical Readers (May 2026)
Google Antigravity 2.0 is Google's new 'hand multiple jobs to AI in parallel' IDE, announced at I/O 2026 on 2026-05-19. This guide covers the VS Code fork + Manager View + Gemini 3.5 Flash setup, the free tier, pricing, risks, and Sales Claw-side business workflows — all written for non-technical readers.
中澤 圭志15 min
Gemini Omni Explained: Google's New 'Talk-to-it' Video AI, for Non-Technical Readers (May 2026)
Gemini Omni is Google's new 'talk-to-it' video AI, announced at I/O 2026 on 2026-05-19. This guide explains how it differs from Sora 2 and Veo 3, what's bundled into the new $100 Google AI Ultra plan, the real limits of SynthID watermarking, and how business teams should pair video AI with sales-outreach tools like Sales Claw — all written for non-technical readers.
中澤 圭志14 min
Claude Code 2.1.147 ships /code-review, then 2.1.148 squashes a Bash-exit-127 regression within 24 hours
Released 2026-05-21, Claude Code v2.1.147 replaces /simplify with /code-review (supports --comment for inline GitHub PR comments), adds pinned background sessions, and ships 30+ Windows / PowerShell fixes. A Bash-tool regression that returned exit 127 on every command crept in, and v2.1.148 ships ~5 hours later as a hotfix. This post bundles "useful new features" and "broken-and-fixed-in-five-hours" with an operations survival kit for the regression-injection era.
中澤 圭志14 min
Claude Code 2.1.145 / Codex Mobile / Anthropic's Six-Week Quality Postmortem — Third-Week-of-May 2026 "Coding-AI Infrastructure Era" Roundup
In the third week of May 2026, coding AI shipped three back-to-back "infrastructure era" events. Claude Code 2.1.145 (Fast mode defaults to Opus 4.7, OTel agent_id spans, 50+ improvements), OpenAI Codex mobile (iPhone QR-pair a Mac Codex session, 4M weekly users), and Anthropic publishing the technical detail of a six-week quality regression (three product changes — reasoning-effort downgrade / caching bug / verbosity limit — chained together). Read across three trends: tool → infrastructure, freedom from place and time, and SRE-grade quality accountability.
中澤 圭志16 min
Google I/O 2026 Roundup: Gemini Omni / 3.5 Flash / Spark / Antigravity 2.0 / TPU 8t-8i — 30+ Announcements Across 5 Axes
Google I/O 2026 (2026-05-19) delivered 30+ announcements: Gemini 3.5 Flash GA (Pro-level reasoning at Flash speed, 1/3 the price of frontier competitors), Gemini Omni (world model + integrated video generation), Gemini Spark (24/7 personal AI), Antigravity 2.0, TPU 8t-8i, Search's biggest 30-year update, and Workspace's voice-driven redesign. We organize all of this across 5 axes and read the implications for AI sales automation through Sales Claw's lens.
中澤 圭志14 min
Claude Code v2.1.144 explained: /resume for background sessions, 75-second startup hang fixed, /usage-credits rename — what production users should know
Claude Code v2.1.144 (2026-05-19) is the final installment of a 7-day, 6-release sprint from 2.1.139 onward. The headline changes: /resume × background session integration, a 75-second startup hang fix (api.anthropic.com unreachable → 15-second side-channel timeout), /model session-scoped with d for default switching, /extra-usage renamed to /usage-credits, and 35+ bug fixes. We anchor on the official CHANGELOG and GitHub Releases, with a Sales Claw lens on production operations.
中澤 圭志12 min
Can Cursor Composer 2.5 Match Opus 4.7 / GPT-5.5? — Kimi K2.5 + 25x RL at 1/10 the Cost
Composer 2.5 (2026-05-18) takes the same Moonshot Kimi K2.5 checkpoint as Composer 2, scales synthetic RL tasks 25x, and lands on par with Claude Opus 4.7 / GPT-5.5 on SWE-Bench Multilingual at roughly 1/10 the cost. We pull from Cursor's official blog, changelog, and forum to cover benchmarks, architecture, pricing strategy, long-horizon improvements, the no-public-API risk, and Sales Claw's read on targeted RL.
中澤 圭志11 min
Tune Claude Code to Your Spec — A Practical Harness-Optimization Guide for General Readers
Understand the harness in one shot via the tack metaphor → five inconveniences of factory default → the six-component map → when to use Skills vs Subagents → safety belts with Hooks × Permissions → the 6-step recipe to build it in 30 days → risks (over-add / name conflicts / security / runaway cost) → finally, full disclosure of the Sales Claw production harness (87 items). Anthropic's official Claude Code Docs (Overview / Settings / Skills / Subagents / Hooks / MCP) are the primary sources.
中澤 圭志14 min
Claude Skills — Which Ones Should You Actually Install? 12 Genuinely Useful Skills, Sorted by Category for General Readers
Concept of Claude Skills is clear after last week — but which ones do you actually install? This piece curates Anthropic's official Office four (pptx / xlsx / docx / pdf), three productivity Skills (proposal-master / consolidate-memory / setup-cowork), five Claude Code operations Skills, and two automation Skills (schedule / loop) — twelve in total. Includes a 3-day starter plan, risks around overload and untrusted scripts, and how Sales Claw's sales Skills fit alongside the official set.
中澤 圭志12 min
What are Claude Skills? Understanding Anthropic's New "Recipe-Book Files for AI" in 8 Minutes (Cookbook Analogy)
Claude Skills are 'recipe-book files' for AI: a folder with a SKILL.md that Claude opens only when relevant. Announced 2025-10-16 and opened as a standard 2025-12-18, they fill the gap between lightweight prompts and heavyweight MCP. This is a general-reader walkthrough using the cookbook analogy — three-stage loading, how to write a SKILL.md, the three platforms, and a production checklist.
中澤 圭志12 min
What's an Agent Harness? Understanding Claude Code and Codex CLI in 8 Minutes Using a Car Analogy
A harness is the outer-shell software that wraps an LLM and lets it actually act in the world. With the same Claude or GPT engine, a different harness can swing SWE-bench scores by 36 points. Claude Code, Codex CLI, and Devin are different chassis. This is a general-reader walkthrough using a car analogy.
中澤 圭志12 min
How We Mass-Produce Blog & Social Whiteboard Illustrations with gpt-image-2 — A Practical Sales Claw Workflow for General Readers
gpt-image-2 is OpenAI's third-generation image model (announced 2026-04-21), the first one that reasons about composition before drawing. ~$0.05 per medium 1024×1024 image, ~99% multilingual text accuracy, up to 16 reference images, 2K output. Sales Claw uses it to ship ~100 illustrations per month. This is the practical workflow — what it does, what it costs, and the two real traps.
中澤 圭志12 min
GitHub Copilot in 2026: What It Actually Is Now, How Free / Pro / Pro+ Differ, and When to Use Agent / Coding Agent / Spark
GitHub Copilot has shifted from 'in-editor autocompleter' to 'AI teammate that takes Issues and returns PRs.' Five plans (Free, Pro $10, Pro+ $39, Business $19/seat, Enterprise $39/seat); four pillars (Inline + Chat, Agent mode, Coding Agent, Spark). New Pro and Pro+ sign-ups have been paused since 2026-04-20; billing moves to AI Credits on 2026-06-01. Here's the practical picture for non-technical readers — and how it interlocks with Sales Claw.
中澤 圭志12 min
What Is an AI Agent? Why Nobody Can Define It in One Sentence — and the 10 Major Agents to Know in May 2026
An AI agent is "an AI that, given a goal, plans its own steps, calls tools, and works a task to completion." Nobody summarizes it cleanly because vendors disagree across autonomy, tool-use, and task scope. The practical lens is the 3-tier gradient: Reactive → Tool-use → Autonomous. This article maps the 2026 landscape, lists the 10 agents to know, gives a free 3-step start path, and covers the 5 risks plus a sales context.
中澤 圭志13 min
Codex on Your Phone: How OpenAI's 2026-05-14 ChatGPT-Mobile Codex Release Actually Works (and What It Means for Regular Users)
On 2026-05-14, OpenAI rolled out Codex on ChatGPT mobile (iOS / iPad / Android) as a preview. Pair your phone with a Mac running Codex by scanning a QR code, then approve commands, switch models, and start new prompts from your phone. Available on every ChatGPT plan including Free and Go. Windows pairing is still 'coming soon.' Here's the full picture for non-technical readers — and how it changes the 'I have to be at my desk' rule for AI engineering work.
中澤 圭志11 min
ChatGPT Atlas Explained: What OpenAI's AI Browser Actually Is, and How Regular Users Can Safely Get Started (May 2026)
ChatGPT Atlas is OpenAI's own browser with ChatGPT built in. macOS only as of May 2026; Windows, iOS, and Android still 'coming soon.' Agent mode runs multi-page tasks; an independent test (LayerX) found Atlas's phishing detection lags Chrome and Edge meaningfully. Here's the full picture for non-technical users — and how Atlas pairs with Sales Claw for B2B sales workflows.
中澤 圭志12 min
Claude for Government vs ChatGPT Gov: which one wins, and why Japan is still behind
The U.S. opened both Claude Gov and ChatGPT Gov to all federal agencies at $1/agency/year via GSA OneGov. Japan is building GENAI in-house, collaborating with OpenAI, while Anthropic enters via NEC and a Tokyo office. Here's the full landscape, the FedRAMP gap, and the practical three-year roadmap for Japanese sales and procurement teams.
中澤 圭志14 min
Codex Chrome Extension: lending your signed-in Chrome to the AI — per-site approval and AI sales fit
Codex Chrome Extension (2026-05-07) lets Codex use your signed-in Chrome state so it can act inside LinkedIn, Salesforce, Gmail, and internal tools. Per-site approval keeps you in control. Here's how it pairs with Playwright and Sales Claw, plus a 15-item pre-production checklist.
中澤 圭志13 min
Codex CLI vs Claude Code (May 2026): a cross-cut benchmark comparison — SWE-bench, Terminal-Bench, cost
Codex CLI 0.130.0 and Claude Code 2.1.143 reshuffle rankings depending on the benchmark axis. We walk through Terminal-Bench 2.0 / SWE-bench Verified / Aider Polyglot (official + third-party aggregates), API pricing, CLI feature deltas, and AI sales-automation fit — all from the Sales Claw maintainer's seat.
中澤 圭志16 min
Claude for Small Business: Anthropic launches into SMB with 7 SaaS × 15 workflows and an approval-everywhere model
Anthropic announced Claude for Small Business on 2026-05-13. 7 SaaS integrations × 15 workflows, with every task user-initiated and user-approved. We walk through what overlaps with Sales Claw and how the approval-everywhere philosophy is becoming the SMB-AI industry standard.
中澤 圭志14 min
Claude Code 2.1.143: Plugin dependency enforcement + PowerShell default-on + /goal × background race fix — closing the 96-hour 5-release streak
Claude Code 2.1.143 closes a 96-hour run of five consecutive releases (2.1.139 → 143). Headlines: plugin dependency enforcement, PowerShell tool default-on for Windows enterprise users, and the /goal × background shell race fix. We walk through the four primary change groups and the operational impact on Sales Claw.
中澤 圭志16 min
Chrome DevTools MCP: How to let Claude Code drive real Chrome — 42 tools in v0.26.0, Lighthouse audits, and Sales Claw integration
Chrome DevTools MCP is the Chrome team's official MCP server for AI coding agents. v0.26.0 (2026-05-12) brings 42 tools across 10 categories, including Lighthouse audits and Core Web Vitals capture. This guide walks through setup and operational risk based strictly on first-party sources.
中澤 圭志14 min
Model Context Protocol (MCP): a deep dive into the "USB-C port for AI" — architecture, primitives, and Sales Claw integration
MCP is the LLM-app extension protocol Anthropic launched in November 2024 that became an industry standard in 18 months. We walk through the Host / Client / Server tiers, JSON-RPC 2.0, six primitives, ecosystem adoption, and what 'Sales Claw as MCP server' looks like in practice — including safety risks and mitigations.
中澤 圭志18 min
Claude Code 2.1.142 + Codex remote-control: orchestrating multi-agent workflows with --agents
Claude Code 2.1.142 introduces --agents and a remote-control API that lets you delegate sub-tasks to other agents (including Codex) from inside a Claude session. We map the architecture, the safety model, and the practical patterns.
中澤 圭志16 min
Claude Code slash commands: a complete guide — built-ins, custom commands, MCP / SDK integration
Slash commands turn Claude Code into a programmable shell for your codebase. This guide covers every built-in, the .claude/commands directory format, MCP-exposed commands, and the SDK SlashCommand tool — with examples you can copy verbatim.
中澤 圭志15 min
Claude Code 2.1.142: --agents flag, Fast Mode, and Opus 4.7 — what changed and why it matters
Claude Code 2.1.142 (2026-05-15) is a substantial release: --agents to orchestrate multiple agents, Fast Mode for snappier Opus output, and Opus 4.7 as the new default. We unpack the practical impact.
中澤 圭志13 min
Claude Code 2.1.141: rewind workflow and workload IDs explained
Claude Code 2.1.141 adds rewind (revert a session to a prior checkpoint) and workload IDs (group related work). We cover the model and how to adopt them in real projects.
中澤 圭志12 min
GPT-5.5 Instant: ChatGPT switches its default — what it means and how to use it
GPT-5.5 Instant is now the ChatGPT default. We walk through the capability delta from GPT-5, the latency profile, and what to change in your prompts and tool calls.
中澤 圭志13 min
Gemini CLI v0.42 / v0.43 preview: what shipped in the dual release
A dual release: Gemini CLI v0.42 (stable) and v0.43 (preview). Major MCP and tool changes — we summarize what to adopt now vs. pin until stable.
中澤 圭志12 min
Claude Code 2.1.140: subagent goal handling fix and what changed
Claude Code 2.1.140 corrects a subtle subagent /goal interpretation bug — important if you run autonomous loops. We explain the symptom, the fix, and how to validate.
中澤 圭志11 min
AI weekly news (2026-05-12): Claude, Codex, Gemini, MCP — what landed
Weekly digest: Claude Code releases, Codex changes, Gemini CLI updates, and MCP ecosystem moves for the week ending 2026-05-12.
中澤 圭志12 minNew posts announced on GitHub and X
Release notes and posts go out on the same channels.