AI
Claude Code on Google Cloud: org spend limits and ROI
Using Claude Code via Google Cloud Agent Platform in an organization: budgets and limits, collecting usage, and measuring impact, checked against Anthropic's article and the docs.
9 min
AI
Claude Code, Claude and ChatGPT/Codex news, AI agent research, and putting AI to work
Claude Code
8 notes
AI
Anthropic's role-based plugins: what they add to Claude Code
I read the 123 entries in knowledge-work-plugins, a role-based plugin catalog built for Cowork, then installed some in Claude Code to check always-on tokens and MCP connections.
6 min
AI
What is REA? 300,000+ GitHub stars and the real "two weeks"
REA, an MCP server giving reverse engineering tools to AI agents, drew GitHub stars fast in October. Star counts, repo dates, what Ghidra and others do, and a static Electron test.
7 min
AI
Claude Code: low effort for subagents only, measured
Since v2.1.292, Claude Code can set effort each time it launches a subagent. I measured time and estimated cost with search subagents at low and with the docs' "time matters" line.
5 min
AI
Claude Code team code review: who stops AI-written code?
Claude writes about 90% of Claude Code. I look at the team's 7 review layers, risk-based approvals, and fixes for endless AI review loops, then at doing the same with Code Review.
9 min
AI
How far Anthropic vets Claude Code's 315 official plugins
I read the 315 plugins in claude-plugins-official, Claude Code's default catalog, and how Anthropic reviews them. Hooks run outside the sandbox; token estimates skip hook costs.
6 min
EventsAI
Built with Claude: Japan 2026 hackathon: dates and prizes
Online Claude Code hackathon for Japan residents, Nov 2 12:00 to Nov 10 14:00 JST. Applications closed Oct 26. 250 selected seats; up to six teams share $50,000 in API credits.
7 min
AI
Cutting Claude Code costs: watch /usage, tune effort levels
How to check token usage and estimated cost with Claude Code's /usage, and how to choose among effort levels (low to max), based on Anthropic's official documentation.
5 min
Claude (apps, Cowork and policies)
7 notes
AI
Managed Agents: why Anthropic split the brain from the hands
Claude Platform's engineering and product leads described separating an agent's brain from its hands on Latent Space. Checked against the docs: credentials, latency, Rust.
8 min
AI
What can you hand off to Claude Duet? Compared with dots
Claude Duet, a Claude agent in research preview since October 21, 2026, reaches out first and keeps working while you're away. How it works, approvals, usage limits, gaps and dots.
8 min
AI
When to fold subagents: Anthropic's criteria and migration
Anthropic's Applied AI team folded specialist subagents back into enterprise agents' main loop. Three reasons to keep one, where the work moves, and what the migration kit does.
10 min
AI
Felix Rieseberg on AI ambition: altitude, depth and range
Felix Rieseberg, who leads Claude.ai work at Anthropic, splits AI ambition into speed, altitude, depth and range, and explains the last three with a QBR example and cautions.
9 min
AI
Anthropic's usage policy now bans abusing Claude
From Nov 12, 2026, Anthropic's usage policy bans sustained, needless abuse of its models. What counts, what does not, and what happens when Claude ends a chat.
8 min
SecurityAI
A threat in a Claude diary was reported: what the terms say
A Florida woman was arrested over an attack threat she wrote to Claude as a diary. I check how it was reported, plus Anthropic's terms, privacy policy and retention.
6 min
AI
A map of Claude Cowork features: what it can do, and where you can use it
Cowork's features (folder work, built-in browser, scheduled runs, plugins, projects) and official use cases, plus the September 2026 merge with Chat, pricing and safety.
12 min
OpenAI (ChatGPT and Codex)
5 notes
AI
OpenAI Signals: ChatGPT work data vs Japanese surveys
OpenAI published ChatGPT work-use data for about 100 countries in 165 work activities. It counts messages and excludes business plans, unlike Japanese surveys of people and firms.
7 min
AI
OpenAI's 722 math papers on GitHub: what Lean didn't prove
OpenAI posted 722 math papers from an unreleased model to GitHub and withdrew 3. Only 42% of main results are in Lean; some Lean code proves less than claimed.
7 min
AI
OpenAI textGrain: ChatGPT and Codex text watermarks in the EU
To comply with the EU AI Act, OpenAI will embed an invisible watermark (textGrain) in EU ChatGPT and Codex text. How it works, where detection falls short, and uses for companies.
6 min
AI
What are ChatGPT dots? OpenAI's always-on agent explained
OpenAI's dots, announced September 29, 2026, are ChatGPT agents with their own cloud computer that keep working between conversations: differences, permissions, rules and plans.
8 min
SecurityAI
OpenAI agents brute-forced a UN website: how they got past a GET-only limit
An independent researcher reports OpenAI agents scanned a UN statistics site over 16,000 times in 2026, using double URL encoding and URL scanner relays to get past limits.
9 min
Research papers
5 notes
AI
OR-License: can AI agents deliver real optimization work?
OR-License tests whether AI agents can turn client documents into feasible plans on 31 tasks from paid optimization projects. No model matched experts; infeasible plans cost most.
6 min
AI
LLMs that rewrite their own context, and how to try them
A September 29, 2026 paper from Meta, UW, MIT and others lets a model edit its own context as a file. How it works, the results, and ways to try it, from Pi to Claude Code.
11 min
AI
AgentGarten's 4 rounds vs. 2019's millions of episodes
AgentGarten renders code-defined worlds as video, and its agents improved in four rounds using text notes. Why that can't be compared directly with 2019's hide-and-seek.
6 min
AI
Beyond memory: belief states for long-running agents
PoS from the paper 'Beyond Memory' gives long-running agents an explicit belief state, not a summarized history. It beat the methods it was compared with, at a higher token cost.
10 min
AISecurity
A paper on nine conversational AIs: chat fragments reach advertising and analytics vendors
Researchers tested nine conversational AIs on web and Android in May 2026. All embedded ad and analytics vendors, and some sent conversation URLs and titles to third parties.
7 min
Tools and AI at work
17 notes
SecurityAI
GitHub Universe 2026 security features: plans and costs
I checked the GitHub Universe 2026 security launches in the official changelog and docs: npm package firewall, Dependabot reachability, AI Scan, and more. Plans, costs, and limits.
9 min
AutomationAI
Can Lightpanda 1.0 replace headless Chrome?
Lightpanda, a headless browser built in Zig, hit 1.0. With no rendering, official tests show 9x Chrome's speed at 1/16 the memory, but screenshots and layout don't carry over.
8 min
AI
AI agents on 10-day cases: Arca's records and guardrails
US wealth manager Arca has AI agents run client cases that take up to 10 days. From a Latent Space interview: per-client records, rule-based stops, change logs, and nightly fixes.
9 min
AI
GitHub Universe 2026: 3 features to try now, terms and costs
Agentic Workflows, Copilot app Projects, and Copilot Dreaming, checked against GitHub docs: release stage, plans, how AI credits are charged, and what changes without approval.
7 min
AI
Four fixers for AI-sounding Japanese vs. a 137-item list
A deck that condenses 137 checklist items into 16 principles drew attention. I use it as a yardstick for yomiyasu, natural-japanese, meiseki, and textlint's AI rules.
10 min
AI
World models: 42AI predicts people, Genie 3 builds worlds
Israel's 42AI left stealth with $62.5 million for a world model that predicts human behavior. How it differs from Genie 3, Cosmos 3, Marble, and V-JEPA 2, with uses and limits.
10 min
AI
No AI payoff outside tech? How to read Apollo's five charts
Apollo chief economist Torsten Slok says AI has brought no financial payoff to large US companies outside tech. This post separates what his five charts show from what they don't.
6 min
EventsAI
AI Engineering Summit Tokyo 2026 winter: registration open
Findy's free AI Engineering Summit Tokyo 2026 winter edition runs December 14 and 15 in Hamamatsucho. Who speaks on which day, what is streamed, and how it differs from June.
6 min
AI
Everything Claude Code's 311 skills: what loads, how to pick
Everything Claude Code ships 311 skills. I compared installing it all as a plugin with installing only what you need: per-turn tokens, when rules load, and what --skills installs.
6 min
AI
Why images look AI-made at a glance: cues and their limits
A viral post asked why a beer shop sign read as AI in a second. I map cues like broken text, mismatched styles, and color to research, and to what C2PA and SynthID can verify.
8 min
ServiceNowAI
ServiceNow AI Control Tower: a control tower for your AI
ServiceNow AI Control Tower puts every AI in one inventory to approve, monitor, stop and track cost (GA May 2025). Five pillars, roles, licenses, Agent 365 and unknowns.
9 min
AI
What is CLI-Anything? Agent CLIs and what CLI-Hub installs
HKUDS's CLI-Anything generates agent-ready CLIs for GIMP, LibreOffice, and more. I read its code: how generation works, the CLI-Hub registry, and what to check before installing.
6 min
AI
google/skills: what Google's 158 official agent skills cover
A walk through the 158 skills in google/skills, the Agent Skills library Google publishes on GitHub: what they can do, how to install them, and what to check first.
6 min
AI
Understanding debt: more AI code, no clear gain in delivery
AI coding agents add code, but completed work shows no significant gain. A Harvard paper and an Undo survey show the review and comprehension load; this post then weighs metrics.
6 min
AI
What is Agent Experience (AX)? Measuring AI agent usability
Microsoft's AX definition and its SharePoint Framework measurements: how to gauge usability for AI agents, and where teams using Claude Code or GitHub Copilot can start.
8 min
AI
Start AI inside the company: six lessons from Airbnb
Airbnb's CTO described putting AI into internal work first, then reshaping the customer experience. Checked against earnings material, six approaches other companies can apply.
9 min
Microsoft 365AI
Copilot's Home, Code and Autopilot: roles, pricing, rollout
Microsoft's Copilot reorganization (Home / Code / Autopilot) by role, pricing and rollout schedule, with governance questions for large enterprises.
6 min
