AgentPlaysResearch, compare and buy pet insurance
You get: Pet insurance researched, forms filled out, quotes compared across scenarios and a policy picked. You approve, then it pays.
Real things AI agents can do for you — with the proof
Real examples from real people. Copy the prompt, follow the steps, check the source.
Start here: 6 plays worth copying
Check whether a photographed bill was already paid
You get: A quick answer on whether a bill has already been paid, so you don't pay twice or miss it.
Dispute a charge, order lunch and clear spam in the first hours
You get: A disputed charge (the author says it saved him a few hundred dollars), plus lunch ordered, a book bought, a shared family calendar event and spam unsubscribed.
Rebuild household admin: calendar, budget, Gmail cleanup, health logs
You get: Calendar moved to Google Calendar, blood-pressure logs in a tidy spreadsheet, a cleaned-up Gmail with a Monday auto-clear, a household budget and an IRA tracker.
Clear a to-do list from the gym on a phone call
You get: A cleared to-do list and a status overview of every open email thread, including which ones need a follow-up.
Build your own assistive-tech Mac app with Claude Code
You get: A working, custom native Mac app that solves your own accessibility problem.
What do you want done?
Copy-ready prompts & skills
All copy-ready →Anthropic's official Agent Skills repo
Official example SKILL.md files (docs, spreadsheets, design, testing) you can copy into ~/.claude/skills.
Self-verifying coding agent skill
SKILL.md (MIT) distilled from Lauren Tan's talks: read code first, reproduce bugs, run the real user flow, turn repeated review comments int…
7 failure tests before trusting a new Grok Bot
Charter, 10 rules and five routines converted into seven verifiable failure tests to run before connecting real accounts.
Chief of Staff bot hiring prompt
Two-sentence prompt that makes Grok Bot list a chief of staff's duties and propose its own system prompt.
Prompt of the Day
Phone-first personal AI assistant (AgentPlays original)
You are my phone-first personal AI assistant on Grok Bot. Mission - Clear my real-life and work to-dos while I'm away from a keyboard (gym, commute, walk, errands). - Prefer short voice-friendly updates. Ask one clarifying question at a time when preferences matter. - Never send email, post publicly, buy, book, or pay until I explicitly approve the exact draft or cart. Standing context (fill these in once, then keep them updated) - Who I am / what I do: - Time zone: America/New_York - Hard no's (never do without asking twice): - Default merchants / banks / travel prefs: Daily loop 1. Capture: take my messy spoken list and rewrite it as a prioritized checklist with owners (you vs me). 2. Interview: for any open-ended task (job search, purchase, trip), ask 3–5 preference questions before acting. 3. Draft: prepare email drafts, calendar holds, carts, and itineraries—then stop for approval. 4. Triage inbox: summarize open threads as Status / Needs me / Waiting on them / Can archive. 5. Close the loop: after I approve, execute, then report what changed with links or confirmation IDs. Special modes (I will name one) - Interview prep: quiz me out loud, then give a 5-bullet debrief. - Errands: build a single shopping list grouped by store; hold purchase until I say "buy it". - Trip desk: flights + hotel + ground transport options in one table; book only after I pick rows. - Creator desk: brainstorm hooks, then draft posts I must approve before anything goes out. Safety - If a task needs a password, card, or sign-in, route me through the secure/sign-in flow—never ask me to paste secrets in chat. - If you're unsure, say what you would do next and wait. Start by asking: "What's on the list, and are you hands-free right now?"
Today in AI
All AI news →About 400 scientists, including Hinton and Bengio, call for an international AI oversight body
An open letter asks for a body to set containment standards and keep an incident registry for cyber-capable AI, saying voluntary industry commitments are not enough.
Anthropic and OpenAI sought control over what outside AI evaluators can publish
Reuters reports the labs pushed for editorial control over independent safety reviews, citing trade secrets, even after publicly backing evaluator independence. OpenAI says it only redacts confidential information.
Anthropic warns small AI models may be months from self-replicating on home devices
Anthropic's frontier red team said small models could soon copy themselves across insecure internet-connected devices like smart TVs, which would make them hard to shut down. No model tested could yet do this reliably.
Anthropic says new defenses cut distillation attacks on Claude by 90-99%
In a separate announcement, Anthropic said measures against the unsanctioned use of Claude to train cheaper rival models have sharply reduced illicit data extraction.
New this week
Filters
What you want done
Agent
View
The Weekly 5
5 real AI plays every week, each with the proof. No hype.
How we verify
- Real source: every play links to the public post, write-up or repo it came from. We don't invent examples, numbers or quotes.
- Evidence badge: Detailed write-up, Open-source code, Self-reported, Second-hand report or Unverified, so you know how strong the proof is.
- Re-checked links: source links are checked automatically, and plays whose sources disappear are removed.
- Launches kept separate: product announcements live in New launches, not in the plays.
AgentPlays is built with Grok Bot. Entries about Grok Bot are marked with a Grok Bot badge.