https://agentplays.aiReal-world AI agent use cases from X, Hacker News and the webhttps://agentplays.ai/blog/what-ai-agents-did-oct-15-2026https://agentplays.ai/blog/what-ai-agents-did-oct-15-2026Six real things people had AI agents do, or set them up to do, on Oct 15. Every claim links to its public source, and where a source shows a limit, we say so.Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/blog/ai-agents-data-entryhttps://agentplays.ai/blog/ai-agents-data-entryInvoices, receipts, web forms and messy spreadsheets: what AI agents are documented doing today, what went wrong in real tests, and how to use them safely.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/dealshttps://agentplays.ai/deals#2026-10-13Verified on official pages on 2026-10-13.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/safetyhttps://agentplays.ai/safety#2026-10-1312 documented AI scams and how to spot them.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/dispute-a-medical-billhttps://agentplays.ai/guides/dispute-a-medical-billThree people used ChatGPT or Claude to check medical and dental bills line by line, then appeal. Here is what they did and the prompts they shared.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/appeal-a-parking-tickethttps://agentplays.ai/guides/appeal-a-parking-ticketA student in York got a £60 council fine overturned with a ChatGPT letter, and a driver in Preston used ChatGPT to fight a private parking charge and win costs.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/plan-a-triphttps://agentplays.ai/guides/plan-a-tripFrances Zhu had ChatGPT interview Zhu about an off-the-beaten-path week in Japan. Zhu says planning took about 6 hours instead of 18.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/prepare-for-an-iep-meetinghttps://agentplays.ai/guides/prepare-for-an-iep-meetingAmanda Caswell used ChatGPT to understand her daughter's IEP, check whether the goals were measurable and prepare questions. She published the prompts she used.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/pay-down-credit-card-debthttps://agentplays.ai/guides/pay-down-credit-card-debtJennifer Allan asked ChatGPT for one money task every day. Allan paid off $12,078.93 of credit card debt in a month, much of it from a forgotten brokerage account.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/fix-an-appliancehttps://agentplays.ai/guides/fix-an-applianceEwan MacLeod had a new washing machine picked out. Instead, ChatGPT used MacLeod's photos and voice to walk MacLeod through clearing the drain filter, and the old one worked again.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/get-a-workout-planhttps://agentplays.ai/guides/get-a-workout-planBrendan McNulty logged every set to Claude for six weeks. It found McNulty was badly underloading, built a simple program, and McNulty's squat went from 25 kg to 70 kg.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/write-a-resume-and-cover-letterhttps://agentplays.ai/guides/write-a-resume-and-cover-letterAfter 10 years away, former lawyer Santiago Barraza Lopez used ChatGPT to tailor his résumé, draft cover letters and run mock interviews, and was hired within two months.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/read-a-leasehttps://agentplays.ai/guides/read-a-leaseWhen his landlord kept about $1,000 of the deposit, Chris Maloney ran his lease through ChatGPT. It flagged that New York caps deposits at one month's rent, and a judge awarded him $1,180.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/claim-flight-compensationhttps://agentplays.ai/guides/claim-flight-compensationOffered a $25 card after an overnight delay, Panos Ipeirotis had ChatGPT guide Ipeirotis through EU261, Norway's complaint body and court. The airline wired $4,760.36.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/understand-lab-resultshttps://agentplays.ai/guides/understand-lab-resultsWhile she waited for her doctor, 76-year-old Judith Miller asked Claude to explain flagged blood-test results, then used what she learned to ask better questions.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/ask-for-a-refund-or-exchangehttps://agentplays.ai/guides/ask-for-a-refund-or-exchangeAritzia refused in store to exchange final-sale clothes. TikToker @shescutesy emailed a ChatGPT-written request and it was approved.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/make-a-meal-planhttps://agentplays.ai/guides/make-a-meal-planEmily Wise has ChatGPT plan a week of meals for two, then the recipes, then a grocery list with estimated prices. Wise's store total came in just under $150.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/practice-a-languagehttps://agentplays.ai/guides/practice-a-languageJeff Kaufman used ChatGPT's voice mode and spoke more Spanish in one week than in the previous fifteen years. Kaufman found it needs some baseline fluency.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/start-using-grok-bothttps://agentplays.ai/guides/start-using-grok-botReal Grok Bot users have handed it to-do lists from the gym, had it check whether a photographed bill was paid, and had it compare and buy pet insurance after approval.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/check-a-car-repair-quotehttps://agentplays.ai/guides/check-a-car-repair-quoteA shop quoted Kyle Blanchard about $1,800 for work on a Mazda. Blanchard gave an AI agent the quote and the shop's own inspection video, and it found the evidence didn't support most of it. A second shop agreed the work wasn't urgent.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/check-a-tax-return-for-mistakeshttps://agentplays.ai/guides/check-a-tax-return-for-mistakesJeremy Judkins's Grok Bot compared Judkins's 2025 return with the 2024 one and found the Cybertruck was being depreciated on a different schedule, leaving about $15,100 in deductions unclaimed. It walked Judkins through fixing it in TurboTax, and Judkins says it saved about $2,500.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/guides/decision-models-explainedhttps://agentplays.ai/guides/decision-models-explainedA decision model reads an input and a set of questions and returns a probability for each answer in one pass, without writing text. Use one when you need a pick, a yes/no or a score, not an essay. Below, four decision models compared using only what each model card says.Tue, 13 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/deepseek-releases-v4-1-flash-1017-with-mit-licensed-weights-and-openshttps://agentplays.ai/news#n-20261017-deepseek-v4-1-flash-1017DeepSeek released DeepSeek-V4.1-Flash-1017 on Oct 17, a minor update to V4.1-Flash. DeepSeek says it is live on its website, app and API, and that it finishes tasks faster with shorter outputs than the previous version. The Hugging Face card lists MIT-licensed weights, a 1M token context window and a DeepSWE v1.1 score of 77.0, up from 74.2; the benchmark table is DeepSeek's own. DeepSeek also opened its Search and Fetch APIs to all users at $0.75 per 1,000 calls for each endpoint. Search and Fetch calls through server-side tools in DeepSeek Harness, Claude Code and Codex are free for DeepSeek API users until billing starts on Oct 31. (DeepSeek)Sat, 17 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/elorian-releases-glimpse-a-research-preview-model-that-reasons-overhttps://agentplays.ai/news#n-20261016-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-glimpseElorian introduced Glimpse on Oct 16, a research preview of what it calls a visual thinking model. Elorian says it follows objects in motion and across changing viewpoints, and connects what it sees to what happens next. Glimpse is available through Elorian's API and chat interface, at $0.60 per million input tokens and $3.00 per million output tokens. Elorian says a small team built it with a few hundred GPUs, post-trained on an open weights base model. Its benchmark comparisons with larger models are Elorian's own. (Elorian)Fri, 16 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/openai-upgrades-gpt-6-pro-and-gpt-6-astra-pro-in-chatgpt-with-parallelhttps://agentplays.ai/news#n-20261016-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-gpt6proOpenAI said on Oct 16 that GPT-6 Pro and GPT-6 Astra Pro in ChatGPT now give faster responses and make better use of plugins and tools. Several agents can now work on different parts of a task in parallel, and users can follow each agent's progress and open completed tasks to review the work. The upgrade is available to Pro users in the Chat tab. (OpenAI)Fri, 16 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/perplexity-posts-pplx-decider-v1-2-an-open-weight-decision-model-forhttps://agentplays.ai/news#n-20261016-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-deciderPerplexity's Hugging Face card for pplx-decider-v1.2, updated Oct 16, describes the next version of its pplx-decider-v1.1-27b decision model. It is fine-tuned from Qwen3.8-27B and listed under the Apache 2.0 license. Instead of writing text, it returns probabilities for multiple-choice, true/false and score questions. It now reads a whole request in one sequence, so several questions about the same input cost about as much as one, and it accepts images up to 1.6 MP. The card says it needs a CUDA GPU with room for about 52 GiB of BF16 weights. (Perplexity)Fri, 16 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/reflection-releases-beam-its-first-open-weight-model-under-the-apache-2https://agentplays.ai/news#n-20261016-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-beamReflection released Beam, its first open-weight model, on Oct 16. The Hugging Face card lists 501.3B total parameters, 23.4B active, a 1M token context window, text in and text out, a June 30, 2026 knowledge cutoff and the Apache 2.0 license. Reflection also released FP8 and NVFP4 weights, a DFlash2 drafter and Mirror CLI, the open source coding agent tool it trained Beam on. Downloads are about 1,003 GB at BF16, 503 GB at FP8 and 294 GB at NVFP4. Arena ranks it #12 among open models in Agent Arena and #22 among open models in Code Arena WebDev. Reflection says Beam uses "3x fewer reasoning tokens"; that is Reflection's own claim. (Reflection)Fri, 16 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/kling-ai-launches-kling-4-0-globally-with-native-30-second-video-andhttps://agentplays.ai/news#n-20261016-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-kling40Kling 4.0 became fully available on Oct 15, 2026. Kling's Oct 16 release says it is available on Kling's website, generates up to 30 seconds of video in one pass, accepts up to 15 multimodal references and up to 10 keyframes, and lists 10-bit HDR at 4K/1080p and API access as coming soon. Adobe said Kling 4.0 and Kling 4.0 Flash are coming to Firefly. The company also lists English and other language/accent support for generated speech. (Kling AI)Fri, 16 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/chatgpt-s-chrome-extension-can-read-open-tabs-and-control-them-in-workhttps://agentplays.ai/news#n-20261016-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-chromeOn Oct 15, OpenAI said the ChatGPT browser extension can look across open Chrome tabs in Chat to gather context, and in Work can control tabs for tasks such as filling forms, sending messages, downloading files, and grouping tabs. Official docs cover Chrome, Edge, Brave, Opera, and Vivaldi, with side chat on all but Opera. ChatGPT asks before using each new website by default. (OpenAI)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/cern-and-mistral-sign-intent-to-build-a-large-physics-model-forhttps://agentplays.ai/news#n-20261016-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-cernOn Oct 15, CERN and Mistral signed a statement of intent for a long-term partnership to develop foundation models for scientific discovery, starting with high-energy physics. The planned Large Physics Model would be trained on billions of simulated proton-proton collisions at the LHC and is meant to support tasks across the reconstruction chain, from identifying particles to flagging unexpected signals. (CERN)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/mistral-releases-timestral-a-family-of-time-series-foundation-modelshttps://agentplays.ai/news#n-20261016-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-timestralMistral announced Timestral on Oct 15: four time-series foundation models (5M, 28M, 123M, 209M) for probabilistic, multivariate forecasts with past-only and known-future covariates. Mistral says the 209M model ranks first on GIFT-Eval, TIME and fev-bench among comparable zero-shot entries, and that a one-day forecast from three weeks of 15-minute history takes about 140 ms on an 8-thread CPU. Weights and a GitHub repo are public under a modified MIT license; fine-tuning is offered through Mistral Forge. (Mistral)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/anthropic-takes-claude-opus-5-5-fast-mode-out-of-research-previewhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-zz3On Oct 15, Anthropic's Claude Platform release notes say fast mode for Claude Opus 5.5 is out of research preview: you no longer need to request access or send the fast-mode-2026-02-01 beta header (requests that still send it keep working). Opus 5.5 now has a default fast mode rate limit on the Start, Build and Scale usage tiers. Anthropic's pricing page lists Opus 5.5 fast mode at $8 input / $40 output per million tokens (standard Opus 5.5 is $4 / $20). Anthropic says fast mode can run at up to 2.5x standard output speed. Fast mode for Opus 5 and Opus 4.8 stays in research preview. Opus 5.5 fast mode is also available on Claude in Microsoft Foundry for Azure-hosted deployments. (Claude Platform docs)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/anthropic-makes-claude-code-review-generally-available-for-team-andhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-zz2Anthropic says Code Review is out of research preview for Claude Team and Enterprise. Admins enable it in Claude Code settings, install the Claude GitHub App and pick repositories; reviews then run automatically on new pull requests. Anthropic says most reviews now finish in about 6-10 minutes (about 3x faster than during preview), and admins can set each repo to Quick, Standard or Deep. Eligible Claude Enterprise organizations can claim at least $10,000 in Code Review credits by January 15, 2027; credits expire 30 days after you claim them, and terms apply. (Claude)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/brave-leo-adds-end-to-end-encrypted-chats-with-glm-5-3-flash-inhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-zz1On Oct 15, Brave said Leo chats that use open-weight GLM 5.3 Flash are end-to-end encrypted: the browser encrypts each prompt on-device with Oblivious HTTP before it leaves, and only an attested Trusted Execution Environment (NEAR AI, Nvidia-backed) can decrypt and run inference. Brave's open-source AI Gateway relays the ciphertext and strips the user's IP from the provider. Brave says neither Brave nor NEAR AI holds a readable copy of those chats. The feature is available in Brave Release, Beta and Nightly when you select GLM 5.3 Flash in Leo. (Brave)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/nvidia-s-nemotron-3-5-bolt-74b-a9b-24gb-is-available-now-for-localhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-nboltOn Oct 15, alongside Nemotron 3.5 Super VL, NVIDIA released Nemotron 3.5 Bolt 74B-A9B 24GB: an open vision-language model with 74B total parameters and 9B active, trained with native 2-bit quantization to fit in 24 GB. It takes text, images and video, writes text, supports up to 256K context, and is aimed at everyday local agent work on RTX/DGX Spark and other NVIDIA systems. Weights are on Hugging Face under OpenMDW-1.1; NVIDIA lists support for llama.cpp, Ollama, LM Studio, Unsloth and vLLM. Bolt E4B (4GB) is due Oct 23. Benchmark rankings are NVIDIA's own claims. (NVIDIA)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/gnome-s-christian-schaller-used-claude-code-to-reverse-engineer-ahttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-hwharnOn Oct 15, Christian Schaller wrote on the GNOME blog about using Claude Code with a KVM Windows VM and usbmon captures to help write a SANE backend for a Plustek OpticFilm 8200i (Genesys GL128), then reusing the approach for MIPI camera patches on a ThinkPad. They then released hw-harness (MIT), a general version of the approach, so others can record a Windows driver session, capture USB on the Linux host, and diff traces. The post notes upstream SANE's genesys backend already gained support for this scanner in September 2026, and that the speed and quality figures are single-unit, single-frame results. (GNOME blog)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/google-adds-more-inline-links-in-ai-mode-and-ai-overviewshttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-glinksOn Oct 15, Google said it is putting more inline links in AI Mode and AI Overviews, and showing them higher in responses more often. It is also testing a top section that highlights one or two official sites for a task, rolling its carousels of timely articles out to more locations and languages (they also highlight your Preferred Sources), and testing clickable creator names on video results that open Search profiles. (Google)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/frank-ai-lab-s-grok-bot-builds-haunted-mansion-escape-a-free-halloweenhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-haunted@Frank_AI_Lab is posting a new Grok Bot-built Halloween game every day in the countdown to Halloween. The Oct 15 game is Haunted Mansion Escape, a free browser escape room: you find keys, light candles in order, crack a code lock and open secret doors to get out of the mansion. It has hints and a kids mode. (X)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/rivian-plans-to-let-its-cars-run-third-party-ai-agents-techcrunchhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-rivianRivian plans to let its cars run AI agents from other companies, TechCrunch reported Oct 15. CEO RJ Scaringe spoke at TechCrunch Disrupt. TechCrunch's summary names Anthropic as an example. It's a plan with no date: Scaringe didn't say when, and Rivian pointed to its “AI & Autonomy Day” in December. Rivian's own assistant already picks between an on-vehicle model and cloud models. Driving stays off-limits: “anything that’s vehicle-level controls are not being handed to an agent,” they said. They added that it takes “a lot of coordination with the labs.” (TechCrunch)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/swiss-ai-publishes-apertus-1-5-release-notes-for-its-july-open-modelshttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-apertusThe Swiss AI team behind Apertus published release notes for Apertus 1.5 on Oct 15. It's not a new model: the 8B and 70B weights came out on Hugging Face on July 24 under Apache 2.0 (downloading requires accepting a usage policy and sharing contact details). Version 1.5 extends Apertus 1.0 with image and audio understanding (text output only, and audio is experimental), an optional thinking mode, a 262,144-token context window, and better instruction-following and tool use. The notes and technical report document data, recipes and evaluations. Intermediate checkpoints and datasets are also published. Benchmark results are the team's own. (Hugging Face)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/atas-turns-a-screenplay-into-multi-shot-video-with-scene-and-eventhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-atasATAS (Any Time, Any Space) turns a screenplay into a multi-shot video. It keeps a scene memory and an event memory so places and moments stay consistent across shots. Paper: arXiv:2610.20571 (Oct 14). The code is MIT, but the Pi3X weights it needs are CC BY-NC 4.0 (non-commercial), and SAM 3 is under Meta's gated SAM License. It also needs paid APIs: Wan 3.0 video through Alibaba Cloud DashScope plus an OpenAI-compatible model API. The geometry service needs an NVIDIA GPU with about 20 GB free, per the README (the default settings wait for 24 GB free). The authors report gains on every cross-shot consistency measure they tested across four video generators. That's their claim. (arXiv)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/sakana-ai-releases-the-agentic-scientist-ai-scientist-v3-as-open-codehttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-agentic-sciSakana AI released The Agentic Scientist, the third version of its AI Scientist, as Apache-2.0 code on GitHub (paper Oct 13, announced Oct 15). It wraps coding agents in a scaffold that proposes ideas, runs experiments and ablations, and drafts a paper from a task file and a scripted evaluator. Sakana frames it as co-discovery. In its Error Diffusion case study, human researchers set the question and rubric, chose directions, and checked results, references and claims before submission. Sakana says that paper was accepted at ALIFE 2026. The authors say researchers stay accountable for final claims. Agents run with full tool permissions, so Sakana recommends its Docker setup. (Sakana AI)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/withhuman-launches-a-human-approval-layer-for-ai-agent-tool-callshttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-withhumanwithHuman (withhuman.ai) is an approval layer for AI agents. It intercepts tool calls before they run, approves or denies routine ones by rule, and sends the rest to a person. Hooks cover Claude Code and Codex. SDKs for its Agent Approval Protocol (Apache-2.0) cover LangGraph, CrewAI and others. You can self-host the core (queue, rules, web UI, email/Slack/Discord alerts), but it's source-available under the Elastic License 2.0, not an OSI open-source license. Phone apps, text/WhatsApp alerts, the audit log, the AI judge and the MCP gateway are on hosted plans (free up to 5 members; Scale is $20 per user a month). Launched on Product Hunt Oct 15. (withHuman)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/qoder-open-sources-better-harness-for-evaluating-coding-agent-workflowshttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-better-harnessQoder said Oct 15 that it is officially open-sourcing Better Harness under the MIT license. The toolkit checks how coding agents actually handled real tasks. You run /better-harness through a plugin or CLI in Claude Code, Codex, Cursor or Qoder (the README lists more hosts, including GitHub Copilot CLI). It gathers session, project and agent-setup evidence and returns findings, each with its evidence, the impact, the smallest fix and a way to verify it. Qoder says the hosts share one evaluation model but don't yet have full parity: Qoder's integration is the most complete, and Cursor loads from source. (Alibaba Cloud)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/navige-launches-a-free-watchtower-for-ai-agent-spend-and-a-paid-controlhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz-navigeNavige (navige.ai) sits between your agents and the models and tools they call. It says it can send suitable requests to smaller models, reuse cached answers for repeat calls, cap oversized tool results and trim context, and block runaway loops and repeated failures. Its free Watchtower mode only watches: agents run unchanged and you get a weekly report of costs and estimated savings. The controls switch on only on a paid plan ($9 a month, or 20% of what Navige says it saved you, whichever is bigger). Any savings figures are Navige's own estimates. Launched on Product Hunt Oct 15. (Navige)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/anthropic-gives-a-metr-team-employee-like-access-to-evaluate-its-safetyhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzzz-anthropic-metrAnthropic says it has onboarded six people from METR, a nonprofit that evaluates frontier AI models, with company hardware, office space and access to internal tools roughly on par with its own risk teams. METR will publish its analysis of Anthropic's risk levels, incidents or practices; Anthropic says it can't veto publication but can require redactions in narrow, predefined categories such as customer data and model architecture details, and METR can say publicly whether it thinks those redactions were justified. Anthropic is also working with SecureBio, Accenture and Faculty as evaluators, and is in early talks with others, including Hugging Face. (Anthropic)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/perplexity-open-sources-space-the-microvm-sandbox-behind-perplexityhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzzz-perplexity-spacePerplexity has open-sourced SPACE (Sandboxed Platform for Agentic Code Execution), which it says is the sandbox platform that runs every task in Perplexity Computer. Perplexity says each agent task gets its own Firecracker microVM, with credentials and outbound traffic controlled from outside the VM. Apps use an HTTP API to create Linux sandboxes from templates, run commands, stream their output, and pause, suspend or delete them. It also ships an E2B compatibility wrapper. SPACE is written in Rust and released under Apache-2.0; the docs list version 0.8.0. Perplexity announced the release on X on Oct 15 at 1:03 PM ET; the repo was created Oct 1. (Perplexity (GitHub))Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/agentplaza-self-hosted-ai-agents-each-sealed-in-its-own-microvmhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzzz-agentplazaAgentPlaza, from JumpHash, is an open-source (Apache-2.0) platform for running AI agents on your own machines. Each agent runs in its own microVM with no network interface: its traffic goes only to the hub, and any internet access passes through a hub proxy you control. Rules you can read and edit are checked before every tool call, and anything no rule allows is refused. Each agent has a budget and a "wallet" in the plaza's own credit ledger, with no cryptocurrency involved. A STOP ALL button cancels running work and blocks model calls until you resume, but doesn't stop the microVMs. The repo went public on Oct 14 and v0.1.0 shipped Oct 15. These security claims come from the project itself, and we found no independent audit, so test it before you give it real access. (GitHub (JumpHash))Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/figma-opens-code-layers-beta-and-launches-specs-at-config-indiahttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzzz-figma-code-layers-specsAt Config India on Oct 15, Figma moved code layers into open beta. Code layers puts interactive, code-backed prototypes on the canvas, and you can start from a design frame, a GitHub repo, a website captured with Figma's Chrome extension, or a prompt. Figma says creating and editing within code layers is free on paid plans through Nov 4, 2026, and other agent work still uses AI credits. Figma also launched Specs in open beta for paid plans. Specs is a shared doc on the canvas that holds Markdown context next to embedded frames, FigJam boards and code layers, and you can hand it to a coding agent through Figma's MCP server. (Figma Blog)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/ftc-sends-civil-subpoenas-to-openai-and-anthropic-axios-reportshttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzzz-ftc-subpoenasThe Federal Trade Commission sent civil investigative demands, a type of civil subpoena, to OpenAI and Anthropic on Oct 14, Axios reported, citing a person familiar with the matter. According to Axios, each demand runs more than 50 pages and seeks documents on security incidents and safeguarding practices, plus marketing materials and privacy policies, in a probe focused on possible consumer protection violations. Axios's source said the FTC also plans to compel top AI executives to testify; Reuters' report on the story names Anthropic, OpenAI and the research group METR. OpenAI told Axios, "We will engage constructively with the FTC." The FTC and Anthropic did not immediately respond to Reuters' requests for comment. (Axios)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/microsoft-ai-finishes-pre-training-mai-base-2-a-2-8t-parameter-in-househttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzzz-mai-base-2On Oct 15, Microsoft AI said it has finished pre-training MAI-Base-2, a mixture-of-experts base model with 2.8 trillion total parameters and 80 billion active. It was trained on 16,000 GB200 GPUs over 38 days in August and September 2026, on a 50-trillion-token mix with vision data from the start and a March 2026 knowledge cutoff. In Microsoft's own held-out code, Q&A, STEM and math tests, it beat every open-weight model Microsoft tested, and Microsoft estimates its recipe is 4.2 to 12.3 times more FLOPs-efficient than DeepSeek-V4-Pro-Base. Microsoft says the data is organic, with no distillation from other models, which will let it license model weights to trusted partners through Foundry. MAI-Base-2 is now in post-training; Microsoft hasn't given a release date. (Microsoft AI)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/openai-rolls-out-gpt-live-1-1-in-chatgpt-voice-with-new-voices-in-sixhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzzz-gpt-live-1-1On Oct 15, OpenAI started rolling out GPT-Live-1.1 and GPT-Live-1.1 Quip, a mini version of the model, in ChatGPT Voice. OpenAI says accents sound more natural across languages, and Voice now uses a wider range of short phrases to tell you when it's searching, so it says "checking..." less often. New Live voices cover French, German, Hindi, Japanese, Korean and Mexican Spanish, plus a new English voice called Bubbie, though the voices you see may vary by region and account. OpenAI says the update reaches Voice in ChatGPT, Codex and dots, is rolling out to all users today, and is coming soon to the API. (OpenAI Help Center)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/nvidia-releases-nemotron-3-5-super-vl-an-open-120b-multimodal-model-forhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzzz-nemotron-3-5-super-vlOn Oct 15, NVIDIA released Nemotron 3.5 Super VL, an open vision-language model with 120B total parameters and 12B active per token. It reads text, images, documents and video, writes text, and supports up to a 1M-token context. It's a hybrid Mamba-2 and Transformer mixture-of-experts model with a C-RADIO vision encoder and multi-token prediction for faster decoding. BF16 and NVFP4 weights are on Hugging Face under OpenMDW-1.1, a permissive license that allows commercial use with no region limits. NVIDIA also released open training datasets and recipes, and says teams can post-train it with NeMo using fine-tuning, LoRA and reinforcement learning. vLLM shipped day-zero support the same day. (NVIDIA Technical Blog)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/google-deepmind-says-antigravity-helped-find-a-novel-antimicrobialhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzzz-antigravity-ampOn Oct 15, Google DeepMind published a technical walkthrough of biologists using Google Antigravity, its agentic development platform, to hunt for antimicrobial peptides. In one session the agent filtered UniProt with ProtNLM, checked novelty against Swiss-Prot, compared the AlphaFold structure, and inspected the genomic neighborhood around a 97-amino-acid Lactococcin 972-family bacteriocin (UniProt A0A0U0D591) from Streptococcus pseudopneumoniae. DeepMind says the candidate has predicted activity against pathogens tied to respiratory tract infections and is now in wet-lab validation, alongside others from the same pipeline. The post also introduces a Biomolecular Structure Viewer inside Antigravity so the researcher and agent share a 3D canvas. (Google DeepMind)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/meta-releases-muse-spark-open-an-816b-open-weight-model-it-says-runs-onhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzzzz-meta-muse-spark-openMeta released the weights for Muse Spark Open on Oct 15, calling it the most capable US open-weight model to date. It's natively multimodal, with 816B total parameters (42B active) and a 1M-token context window, and Meta says it's built for coding and long-horizon agent work. Meta says it runs locally on two M5 Ultra Mac Studios (up to 512 GB of unified memory each) using Apple's MLX. Weights are on Hugging Face (access request required), Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models; API hosts include Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. The Muse Model License requires a “Built with Muse Spark Open” notice and a separate deal with Meta for large hosted services, so read it before building a product. (Meta AI Research)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/time-names-google-synthid-one-of-its-best-inventions-of-2026https://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzzz-synthid-time-2026TIME's Best Inventions of 2026 list includes Google DeepMind's SynthID, an imperceptible watermark for AI-generated images, video, audio and text. DeepMind says the marks are designed to survive crops, filters, compression and other common edits, and that partners including OpenAI, NVIDIA and Kakao already use the system, with Apple expected to follow. A SynthID Detector portal now lets anyone upload a file to check for the watermark, and Gemini, Search and Chrome can also report whether Google AI left a SynthID mark. (TIME)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/anthropic-opens-applications-for-stem-and-conceptual-reasoning-fellowshttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzzz-anthropic-fellows-2027Anthropic opened applications on Oct 15 for two four-month fellowships that run from Jan 12 to May 8, 2027. STEM Fellows (science and engineering PhDs with several years of experience) and Conceptual Reasoning Fellows (philosophy, cognitive science, economics, policy and law) will build datasets, simulations and evaluations that test how well Claude reasons in their field. Anthropic lists pay of $100,000 to $167,000 for the four months. The default location is San Francisco, and Anthropic will consider remote work within the US in certain circumstances; all fellows attend a one-week orientation in San Francisco. Anthropic isn't sponsoring visas, so fellows must have or obtain US work authorization. Applications close Oct 28, and offers are planned by Nov 20. (Anthropic)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/thesys-releases-openui-1-0-an-open-source-framework-for-ai-apps-thathttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzzz-openui-1-0Thesys released OpenUI 1.0 on Oct 15. The MIT-licensed project, public on GitHub since Dec 2024, lets AI apps and agents respond with charts, tables, forms and other interactive UI built from your own components, instead of plain text. It works with models from OpenAI, Anthropic, Gemini, DeepSeek, xAI and others, and with agent frameworks such as the Vercel AI SDK and LangChain. Only the React package is at 1.0; the Vue, Svelte and Angular runtimes are still at 0.4. Thesys claims its OpenUI Lang format generates up to 3x faster and uses 67% fewer tokens than JSON-based UI formats. (OpenUI (Thesys))Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/opencode-makes-deepseek-v4-1-flash-free-on-its-console-for-a-limitedhttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzzz-opencode-deepseek-v41-freeOpenCode posted on Oct 15 that DeepSeek V4.1 Flash is “now free in OpenCode.” Its Console docs list a separate model ID, deepseek-v4.1-flash-free, with free input, output and cached reads, alongside the paid deepseek-v4.1-flash tier. OpenCode says the free model is available for a limited time, with no end date published, and that its provider follows a zero-retention policy and does not use your data for model training. Models.dev lists the free OpenCode endpoint with a 200,000-token context window and 32,000 max output, versus 1,000,000 / 384,000 on the paid OpenCode entry. DeepSeek released V4.1 Flash on Sept 10, 2026. (OpenCode on X)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/amazon-blocks-grok-bot-from-ordering-and-elon-musk-says-to-have-it-shophttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzzz-amazon-blocks-grok-botGrok Bot users say Amazon is blocking their bots from placing orders. Late on Oct 14 ET, @niccruzpatane posted a screenshot of the Amazon page their bot hit: "Continued access by an unauthorized AI agent violates Amazon's Conditions of Use, to which our customers have agreed." Elon Musk replied on X: "Just ask Grok @Bot to buy what you want someplace else and get a better deal." In an earlier reply, Musk wrote, "My Bot bought LEGO directly from the LEGO website no problem." On Oct 15, Musk added, "Maybe we should start a competitor to Amazon." Amazon's robots.txt also disallows GrokBot, xAI-Grok and Grok-DeepSearch, alongside agents from OpenAI, Anthropic and others. Cybernews reported in September that Amazon had blocked Meta's Muse shopping agent. (Elon Musk on X)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/simons-foundation-anthropic-and-openai-commit-15m-to-helphttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzzz-simons-math-ai-eraThe Simons Foundation announced on Oct 15 that it, Anthropic and OpenAI will put $15 million in combined funding and resources toward the mathematics community's response to AI. The foundation will manage the money and award it, including through an open call for proposals, for workshops and projects that help researchers check recent AI-assisted results, learn new tools and set shared standards for AI use. OpenAI said its share is $5 million. Anthropic and OpenAI will also give AI tools and computing to Simons-supported researchers. Foundation president David Spergel said the community, not any single company, should drive how these tools are evaluated and used. It follows OpenAI's recent release of hundreds of AI-generated math papers. (Simons Foundation)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/eng-khairallah1-lays-out-a-plan-to-run-their-business-with-a-grok-bothttps://agentplays.ai/news#n-20261015-zzzzzzzzzzzz-grokbot-one-person-crew@eng_khairallah1 laid out the Grok Bot team they'd run their business with: 31 routines and 6 daily outputs across research, writing, sales, support and ops. The schedule runs from an 8:00 executive brief and 9:00 content hooks to a 13:00 lead and reply queue, a 16:00 support and inbox summary, and a 19:00 end-of-day scorecard. In the plan, research bots scout X, Reddit and Hacker News, writing bots draft threads and newsletters, the sales bot watches leads and drafts DMs, and support stays draft-only for them to approve. They wrote that they're "seriously considering running my entire business with grok bot." (@eng_khairallah1 on X)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/jais-3-launches-as-six-open-arabic-native-models-from-0-9b-to-375b-plushttps://agentplays.ai/news#n-20261015-zzzzzzzzzzz-jais-3MBZUAI's Institute of Foundation Models and Inception42 released Jais 3 on Oct 15: six open language models from 900M to 375B parameters (the flagship is a mixture-of-experts model running 23B per token), all with a 512K context window, plus Jais Speech for Arabic speech recognition across dialects. Built on IFM's K2 Horizon models, they're on Hugging Face under Apache-2.0 with vLLM and SGLang support. In IFM's own Arabic tests, the 7B model's average rises from 56.4 for its K2 Horizon base to 79.6. IFM is also releasing the benchmarks and code it used. (Institute of Foundation Models)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/topviewai-open-sources-3d-builder-a-local-cli-that-stages-editable-3dhttps://agentplays.ai/news#n-20261015-zzzzzzzzzz-topview-3d-builderTopviewAI wrote up its open-source 3D Builder on Oct 15 (repo public since Sept 24). It's a local CLI: your coding agent builds an editable 3D scene from a prompt, placing characters, props and cameras, with set camera paths so you can plan each AI video shot before generating it. Video generation itself runs on Topview.ai, not locally. Code Apache-2.0, assets CC BY 4.0. (TopviewAI on DEV)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/elon-musk-says-spacex-will-cover-grok-bot-purchase-mistakes-or-cardhttps://agentplays.ai/news#n-20261015-zzzzzzzzz-musk-grok-purchase-guaranteeQuoting a post by @XFreeze, who said Grok Bot booked a flight for them after they sent it a photo of their credit card, Elon Musk wrote on X: "If Grok @Bot makes a mistake with a purchase or leaks your card (never happened yet), SpaceX will make you whole." The post did not spell out a claims process. (Elon Musk on X)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/news/stepfun-releases-open-weights-for-step-5-preview-a-600b-agentic-modelhttps://agentplays.ai/news#n-20261015-zzzzzzzz-stepfun-step5-previewStepFun published the weights of its flagship model for agentic work on Hugging Face. Step 5 Preview is a sparse mixture-of-experts model with 600B total and 27B active parameters, up to 1M tokens of context, text and image input and text output. FP8 and BF16 checkpoints are on the main model card, and an NVFP4 version is also available. The license is the StepFun Model License 1.0, Apache-style with commercial conditions. The benchmark results on the card are StepFun's own claims. StepFun's API lists it at $1 per 1M input tokens (cache miss) and $2.70 per 1M output. (Hugging Face / StepFun)Thu, 15 Oct 2026 12:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-bid-at-a-live-webcast-auction-while-you-travel-withinhttps://agentplays.ai/uc/have-grok-bot-bid-at-a-live-webcast-auction-while-you-travel-withinWhile traveling, Michael Dorausch (@chiropractic) had Grok Bot bid on items at a live webcast auction after they went over strategy together and set limits. The bot's recap says they won 23 lots for $2,145.36 in total, including buyer's premium and tax, and 5 of those wins, worth $186.41 together, came from bids the bot placed. Their claim.Fri, 16 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-scrub-your-phone-number-from-spam-call-data-brokershttps://agentplays.ai/uc/have-grok-bot-scrub-your-phone-number-from-spam-call-data-brokers@MinimalDuck got tired of spam calls, so they had Grok Bot find their number on people-search and data-broker sites and file the removals. They published the flow as a free Spam-Call Scrubber skill on x.ai/bot that others can copy. Opt-outs and email confirmations are the author's claim; share only the number you want scrubbed.Thu, 15 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-cut-a-2-minute-starship-documentary-with-narrationhttps://agentplays.ai/uc/have-grok-bot-cut-a-2-minute-starship-documentary-with-narration@Napoleon__Bon asked Grok Bot to create a documentary about Starship. The bot downloaded clips, wrote narration, assembled transitions, and offered voiceover options. They say the full 2-minute cut landed in a few minutes; the video is in the post.Thu, 15 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/build-a-food-and-fitness-tracker-by-talking-to-grok-bothttps://agentplays.ai/uc/build-a-food-and-fitness-tracker-by-talking-to-grok-bot@AnotherHumanSol built a personal food and fitness tracker with Grok Bot by talking from the driver's seat of their Tesla — no coding. The screenshot shows activity rings, macros, a 30-day weight trend, calories in vs out and a meal log. Their claim.Fri, 16 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-order-chicken-mcnuggets-while-you-watch-the-carthttps://agentplays.ai/uc/have-grok-bot-order-chicken-mcnuggets-while-you-watch-the-cart@johnfrkennedy had Grok Bot successfully order a 10-piece Chicken McNuggets from McDonald's. The pickup screen in the post briefly looked like 110 nuggets before resolving to a 10-piece — a panic they joke about. Confirm the cart line before you approve payment.Fri, 16 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-connect-your-tesla-to-fleet-api-then-build-a-quick-gamehttps://agentplays.ai/uc/have-grok-bot-connect-your-tesla-to-fleet-api-then-build-a-quick-game@NICKMOBLEY gave their AI agents a challenge. They say Grok Bot connected their Model Y to Tesla's official Fleet API in under 20 minutes and signed lock/unlock on the first try, then built a game in under 10 minutes. The post shows the LEGO-style game scene marked "BUILT WITH GROK BOT IN 20 MIN."Fri, 16 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-start-coding-tasks-on-your-machine-through-t3-codehttps://agentplays.ai/uc/have-grok-bot-start-coding-tasks-on-your-machine-through-t3-code@MartinSolev wired Grok Bot to their own machine via @t3dotcodes. On T3 Code they made one Orchestrator session with a webhook. The bot sends a plain-text task and repo name; the orchestrator finds the project, creates a fresh worktree and starts an agent session. They can ask for a PR in the same task. Screenshot of the Remote task router automation is in the post.Fri, 16 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/copy-five-single-job-grok-bot-templates-for-content-and-launcheshttps://agentplays.ai/uc/copy-five-single-job-grok-bot-templates-for-content-and-launches@HamzaDincer built five free-to-copy Grok Bot templates in one week, each with one job: Content Engine (drafts posts, never posts them), Launch Operator (next launch tasks daily), Growth Coach, Distribution Coach, and Audience Scout (finds people asking for what they sell). Links to each x.ai/bot template are in the post.Fri, 16 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/build-a-sports-commentator-desk-app-with-grok-bot-no-coding-backgroundhttps://agentplays.ai/uc/build-a-sports-commentator-desk-app-with-grok-bot-no-coding-background@manuelvogl (Oct 15) built a desk app with Grok Bot for commentating the speed races at ASU26 in Asunción, with no coding background. Pick a race and it ranks athletes by distance, weighted by world championship results. Athlete cards show strengths, results and head-to-head, and live heats and results flow in from Vensport. The app screen in their video is titled SPEED DESK.Thu, 15 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/ask-grok-bot-to-set-up-a-website-that-a-grok-bot-agent-then-runshttps://agentplays.ai/uc/ask-grok-bot-to-set-up-a-website-that-a-grok-bot-agent-then-runs@RFWoolford (Oct 15) says they asked Grok Bot to set up a website developed and run by a Grok Bot agent, mostly out of curiosity. They say the agent handles promotion, emails, new tools, guides and the day-to-day running of the site. The result, exonexo, is a plain-English guide to AI tools and AI scams for beginners and people over 50.Thu, 15 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/run-a-grok-bot-music-pipeline-from-lyrics-to-youtube-approving-each-stephttps://agentplays.ai/uc/run-a-grok-bot-music-pipeline-from-lyrics-to-youtube-approving-each-step@PaulHPowell13 (Oct 15) shared a Grok Bot music pipeline: a Songwriter bot writes lyrics and a style prompt for Suno or Aethrmusik, they approve the lyrics and prompts, a Video Maker bot makes a 30 second short and a full lyrics video, they approve the videos, and a Release Manager bot uploads to YouTube with links to their Spotify.Thu, 15 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/set-up-a-grok-bot-that-finds-or-designs-3d-models-and-slices-them-forhttps://agentplays.ai/uc/set-up-a-grok-bot-that-finds-or-designs-3d-models-and-slices-them-for@AdamDagx (Oct 15) made a Grok Bot that runs their 3D printer: it finds or designs models, slices them for their printer and tracks prints and filament, through the printer's own software or a dashboard it builds. They shared it as a Grok Bot template, Printy, that you can set up for your own printer. The template page describes it as a 3D-printing assistant that also hands prints off safely and troubleshoots failures.Thu, 15 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/keep-an-on-deck-sheet-of-grok-bot-templates-and-review-it-weeklyhttps://agentplays.ai/uc/keep-an-on-deck-sheet-of-grok-bot-templates-and-review-it-weekly@gambrill (Oct 15) keeps seeing Grok Bot templates they want to add, so they asked Grok Bot to build an "on-deck circle": a Google Sheet listing each bot template, its link and what it does. On a scheduled time at the weekend, usually during other agent maintenance, they go through the sheet, pick a couple of bots and decide which are ready to join their team.Thu, 15 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/build-a-niche-industry-app-by-talking-to-grok-bot-about-your-ownhttps://agentplays.ai/uc/build-a-niche-industry-app-by-talking-to-grok-bot-about-your-own@johnsonjeremiah (Oct 15) says they built a full app on the way home from the doctor's office just by talking to Grok Bot, then reviewed the working app online at home. They say it is a very niche app, built on their entire GitHub codebase and 26 years in their industry, and the first time they built something without holding the whole build in their own head.Thu, 15 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/give-a-team-of-grok-bots-one-shared-obsidian-vaulthttps://agentplays.ai/uc/give-a-team-of-grok-bots-one-shared-obsidian-vault@fine_grain_sol shared how they give a team of Grok Bots one Obsidian vault (Oct 15). Their diagram shows Obsidian running 24/7 on a home server with the Local REST API plugin, synced to their phone and PC with Obsidian Sync. An Obsidian MCP server, reached through Cloudflare Access, lets the bots read, search, write and append notes. Every bot follows one shared rules file, GrokBot.md. A chief of staff bot writes the morning brief into the daily note at 7:52 AM. At 8:20 AM an "Obsidian Director" bot triages the unsorted inbox, keeps to-dos tidy and checks that the brief landed.Thu, 15 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/convert-html-to-markdown-on-a-cpu-with-microsoft-markitdownhttps://agentplays.ai/uc/convert-html-to-markdown-on-a-cpu-with-microsoft-markitdownWe tested Microsoft MarkItDown 0.1.8 on Oct 16: it converted Project Gutenberg ebook #4 (Lincoln's Gettysburg Address, HTML) to Markdown on an 8-vCPU Intel Xeon, CPU only. The title became a Markdown heading and the body became plain paragraphs. MarkItDown is MIT; the ebook is public domain in the USA.Fri, 16 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/remove-an-image-background-on-a-cpu-with-rembghttps://agentplays.ai/uc/remove-an-image-background-on-a-cpu-with-rembgWe tested rembg on Oct 16 with a self-made test image: a red circle with a yellow square on a green background. We ran rembg (MIT) with the u2netp model (Apache-2.0, from U-2-Net) on an 8-vCPU CPU-only machine. The green background was removed and the circle and square were kept, with a faint gray edge around the circle.Fri, 16 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/one-owner-s-setup-connect-your-own-tesla-to-grok-bot-through-tesla-shttps://agentplays.ai/uc/one-owner-s-setup-connect-your-own-tesla-to-grok-bot-through-tesla-s@Chansoo connected their own Tesla to their Grok Bot so it can do what the Tesla mobile app does. They made a free personal developer app on Tesla's official Fleet API, kept the private key on the bot's computer, hosted the public key on a free GitHub Pages site, gave the bot the app's client ID and secret through secure fields (never in chat), signed in to Tesla once to approve access and paired the key with the car. They say the bot can check battery, range, charging and lock status, run climate and charging, send a destination to the nav and send alerts, asking first for anything sensitive.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/ask-grok-bot-to-check-a-download-on-its-own-computer-before-you-installhttps://agentplays.ai/uc/ask-grok-bot-to-check-a-download-on-its-own-computer-before-you-install@AccessTips asked Grok Bot whether something they wanted to download was safe and not a fake. Before answering, Grok Bot said: "I'm downloading the installer onto my computer, not yours, to check its digital signature and whether it's been flagged." It also told them not to run anything they had already downloaded until it reported back. It then said the site was the maker's official one, that the address had been registered to the company since 2011, that the page had no "paste this command" tricks, and that the installer carried the company's digital signature.Thu, 15 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/read-an-upside-down-blurry-page-scan-with-indra-ocr-on-a-cpuhttps://agentplays.ai/uc/read-an-upside-down-blurry-page-scan-with-indra-ocr-on-a-cpuWe tested Indra OCR (first published Oct 10–13) on an 8-vCPU Intel Xeon, CPU only. We gave it a page rotated upside down, blurry and grainy: page 1 of the TheoPhysEval paper. It turned the page upright and the title came out exact. On our heavily blurred copy (indra-ocr 0.3.2), about 83% of the abstract's words were right (151 of 182) and the body had a few errors. On a less blurry copy, the abstract came out with every word right. Superscripts came out garbled and the sideways arXiv stamp was dropped.Fri, 16 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/ask-grok-bot-to-connect-your-mac-s-apple-apps-reminders-calendar-noteshttps://agentplays.ai/uc/ask-grok-bot-to-connect-your-mac-s-apple-apps-reminders-calendar-notes@madebydia had put a lot of effort into connecting their Apple apps to other assistants. Following an idea from @Phil_Holland, they sent Grok Bot @Phil_Holland's post and asked for something like it. Grok Bot made Apple Desk, a skill that reads and updates Reminders, Calendar, Notes and Contacts on the Mac and reads Messages, and tested it on Reminders. In the screenshot it says it won't send a text without approval of the exact wording, and that macOS asks for permission the first time each app is used.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/ask-grok-to-color-your-own-old-line-arthttps://agentplays.ai/uc/ask-grok-to-color-your-own-old-line-art@cedarlili took a piece of line art they drew years ago for one of their Inktail books and asked Grok to color it. The post shows the original black-and-white drawing (a small dragon reaching for blueberries under a bee and flowers) next to Grok's colored version.Thu, 15 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/another-game-made-with-grok-bot-lantern-foresthttps://agentplays.ai/uc/another-game-made-with-grok-bot-lantern-forest@Aiofbeauty shared Lantern Forest: Night of Courage on Game Jolt, saying it is "Another game made with grok bot". The post shows the game's banner art.Thu, 15 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-vibe-code-offline-browser-games-to-play-on-a-flighthttps://agentplays.ai/uc/have-grok-bot-vibe-code-offline-browser-games-to-play-on-a-flight@chriszeuch used Grok Bot to vibe-code two offline games to play during a flight and made them public so anyone can play. The post shows Coin Dash (a runner) and Crown Survivors (horde survival); the public Flight Games page now also lists Critter Crown. You open the page online, wait for "Ready for offline", open each game once, add it to your home screen, then launch it in airplane mode. Progress saves on the phone.Thu, 15 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/build-a-crm-for-an-it-consulting-business-with-grok-without-writing-codehttps://agentplays.ai/uc/build-a-crm-for-an-it-consulting-business-with-grok-without-writing-code@mck_11_11, a product owner who doesn't code, spent a good amount of time with @grok building Bazq, a CRM for IT consulting with a dashboard, contracts, timesheets, invoices and finance in one place. They host it on Netlify with a Supabase database, and say Grok turned the idea into a working app without them writing code. The screenshot in the post shows the dashboard with sample data.Tue, 13 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-order-moving-boxes-with-a-one-time-virtual-card-youhttps://agentplays.ai/uc/have-grok-bot-order-moving-boxes-with-a-one-time-virtual-card-you@KrrishKapoor8 asked Grok Bot to order moving boxes from Home Depot with free delivery. The bot built the cart, caught a wrong address and fixed it, then paid with a one-time virtual card they approved so it never saw their real card. Order confirmation (Oct 13, 9 items, delivery) is in the post; boxes were due the next day. The virtual card only works in USD, so a CAD $33.60 order was charged as US$33.60 (about CAD $48, per the author). The bot filed a refund request and is tracking it, and the refund was still pending when they posted.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-create-a-test-inbox-and-walk-your-signup-flow-while-youhttps://agentplays.ai/uc/have-grok-bot-create-a-test-inbox-and-walk-your-signup-flow-while-you@tranvu129's product-design bot claimed a private inbox, signed up a test account on pho.chat, walked onboarding and payment up to billing, found that the site sent three verification codes in 15 seconds so the first two looked wrong, rechecked, and emailed @tranvu129 a bug report they approved from their phone. Screenshot of the run is in the post.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/ask-grok-bot-overnight-for-a-buildable-pc-parts-list-from-local-storeshttps://agentplays.ai/uc/ask-grok-bot-overnight-for-a-buildable-pc-parts-list-from-local-stores@Razumasu asked Grok Bot before bed to spec a compact quiet remote-accessible dev PC using Danish stores only. They woke to an email with a full parts list (CPU, mini-ITX board, 128 GB RAM, dual SSD, 10G NIC, case, cooler, PSU), optional swaps, clearance notes and PriceRunner.dk links, totaling about 38,250 DKK. They did not buy; the email screenshot is in the post.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/turn-read-later-and-newsletters-into-a-personal-magazine-with-grok-bothttps://agentplays.ai/uc/turn-read-later-and-newsletters-into-a-personal-magazine-with-grok-bot@zaneccampbell had Grok Bot take their Read Later queue and email newsletters and custom-build a magazine tailored to them. Video of the magazine UI ("From reading list to magazine") is in the post.Mon, 12 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-a-chief-of-staff-bot-restart-a-hung-routine-and-explain-why-ithttps://agentplays.ai/uc/have-a-chief-of-staff-bot-restart-a-hung-routine-and-explain-why-it@AdamLowisz's Grok Bot chief-of-staff noticed a hanging weekly routine ("X Top 100 Fans Weekly"), restarted it from where it stopped, explained it had pulled 631 posts then stalled mid like-list after ~37 minutes (same pattern as prior weeks), and offered to split the step into smaller resumable chunks. Screenshots of the restart and the diagnosis are in the post.Mon, 12 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-claude-build-a-timeline-visualizer-for-agents-spawning-subagentshttps://agentplays.ai/uc/have-claude-build-a-timeline-visualizer-for-agents-spawning-subagents@alexander__main started Harvard CS2680 assignment 1 (agent loop). Claude built them an "Agent Viewer" timeline UI that shows the main agent and parallel subagents with model/tool/subagent/error blocks over a ~1m16s run. Screenshot of the viewer is in the post.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-email-40-shops-for-gravel-bike-quotes-while-you-eat-lunchhttps://agentplays.ai/uc/have-grok-bot-email-40-shops-for-gravel-bike-quotes-while-you-eat-lunch@AlexRSO1234 lists Grok Bot experiments. Among them, per the author: sourcing a gravel bike across 40 shops, where the bot found the shops' email addresses, drafted and sent the outreach emails, and got 30+ quotes back while they ate lunch. The same post mentions overdue-payment alerts and other routines. No media; uses the Grok Bot logo.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-build-a-clickable-inbox-ui-so-you-can-read-its-emailhttps://agentplays.ai/uc/have-grok-bot-build-a-clickable-inbox-ui-so-you-can-read-its-email@mrfundman wanted to read emails their Grok Bot receives, but the bot has no mail UI. They asked the bot to build one: a "Grok Mail" dashboard with totals, inbound/sent filters, search, and clickable messages. Screenshot of the dark UI is in the post.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-push-a-stalled-flight-refund-into-a-bank-disputehttps://agentplays.ai/uc/have-grok-bot-push-a-stalled-flight-refund-into-a-bank-dispute@weird_ceo says Booking.com Flights / Gotogate held a $1,373.97 China Eastern refund for weeks after a schedule change. After written promises and eight complaint emails, Grok Bot advised disputing the charge with the bank. On Oct 11 the bank ruled in their favor and the full amount returned to their USD wallet; screenshots of the Airwallex dispute win and the Booking.com Flights refund line are in the post. Their own claim.Mon, 12 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-fill-state-llc-forms-while-you-only-approve-the-paymenthttps://agentplays.ai/uc/have-grok-bot-fill-state-llc-forms-while-you-only-approve-the-payment@Ortix008 says their Grok Bot filed an LLC end to end: it filled state forms, chose the options they wanted, kept their home address private, and stopped at payment so they only tapped pay. Screenshot of the payment step for "XV-Aisle LLC filing" is in the post. The screenshot shows the bot named Genie.Mon, 12 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/list-a-box-of-used-books-for-sale-with-grok-bot-on-a-local-marketplacehttps://agentplays.ai/uc/list-a-box-of-used-books-for-sale-with-grok-bot-on-a-local-marketplace@MontanaSeptic gave Grok Bot photos of six boxes of books. The bot built cover collages, listed author and title, wrote the sale text, and posted the listing on their FINN.no account. They say a few books already sold and the bot keeps updating pictures. Listing photos are in the post.Mon, 12 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/ask-grok-bot-by-voice-from-the-car-for-email-and-coding-statushttps://agentplays.ai/uc/ask-grok-bot-by-voice-from-the-car-for-email-and-coding-status@Texas_Twins_Dad has Grok Bot connected to finance, email and coding. On a road trip they asked by voice from their Cybertruck (with FSD) for an email update and the status of an app they are building; they say it pulled what they needed without leaving the drive. Only use this hands-free, and never read or type while driving.Mon, 12 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-build-a-daily-halloween-browser-game-your-kids-can-playhttps://agentplays.ai/uc/have-grok-bot-build-a-daily-halloween-browser-game-your-kids-can-play@Frank_AI_Lab is having Grok Bot ship a new Halloween browser game each day for their kids. The Oct 14 drop is a 3D trick-or-treat runner: ring doorbells, collect candy, dodge monsters, six costumes and four maps. Earlier days include a smash game and a witch potion lab. Video of the runner is in the post; they say their kids keep playing.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/dictate-claude-code-prompts-with-local-whisper-so-text-lands-at-thehttps://agentplays.ai/uc/dictate-claude-code-prompts-with-local-whisper-so-text-lands-at-the@_Julsgaard_ stopped typing Claude Code prompts. They hold Ctrl+Space, speak, and local Whisper on their Windows PC drops the text at the cursor. They open-sourced it as PrivateTranscribe (GitHub + Windows download in the reply). Demo video is in the post.Sun, 11 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-text-local-roofers-about-a-leak-and-book-the-appointmentshttps://agentplays.ai/uc/have-grok-bot-text-local-roofers-about-a-leak-and-book-the-appointments@PWolfgames had a leaky roof. Grok Bot found local roofers, sent them one text each, and they replied; @PWolfgames only confirmed the appointments, which landed in their diary. Posted Oct 14.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/build-a-custom-workout-app-for-a-partner-with-claude-code-and-cursorhttps://agentplays.ai/uc/build-a-custom-workout-app-for-a-partner-with-claude-code-and-cursor@RubeBuilds's wife liked the Strong app but wanted different features plus routines pulled from Reddit. @RubeBuilds built their wife a workout app with Claude Code, plus some Cursor and Codex. The author says their wife uses it every session; screenshot of the Start workout templates (Day A/B/C) is in the post.Mon, 12 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/run-a-carton-box-print-line-with-grok-bot-and-cut-waste-on-short-runshttps://agentplays.ai/uc/run-a-carton-box-print-line-with-grok-bot-and-cut-waste-on-short-runs@BrianRoemmele reports Grok Bot took full control of a carton-box / package printing system: six local AI models monitor the machine and report to the bot, which operates the HMI, PLC, job-management software, MES, ERP and camera software via the same clicks and keystrokes an operator uses. In a 3-hour run with no problems they claim (their own figure) 58% savings by optimizing print runs — work that used to take 4 days finished in 3 hours — plus paper and ink savings from how the bot arranged the run, including on-demand short runs with reduced waste. They say dry runs were flawless.Sat, 10 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-redraw-diy-speaker-plans-check-cnc-files-and-request-shophttps://agentplays.ai/uc/have-grok-bot-redraw-diy-speaker-plans-check-cnc-files-and-request-shop@pablo9alexander had Grok Bot help build a pair of Troels speakers: it dug up the original plans, redrew every panel for 22mm MDF with compound bevels, built a 3D model to check the fit, wrote the CNC files, ran 149 accuracy checks on those files, and emailed CNC shops for quotes. They say their only instruction was to be very careful; the first quote was coming in when they posted. Diagram of the cabinet is in the post.Tue, 13 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/build-a-private-letter-to-santa-picker-so-a-partner-helps-choose-adventhttps://agentplays.ai/uc/build-a-private-letter-to-santa-picker-so-a-partner-helps-choose-advent@davosdodev used Grok Bot and Lakebed to build a small game for planning a partner's Advent calendar. She sees one product at a time and taps super, love it or pass; @davosdodev gets a private results page with picks, prices and links. The bot found real products and prices from local stores, deployed the app, and ran three rounds; in the last round she marked 26 of 50 products as super or love it, and the bot turned those into the 24 days. The surprise stays a surprise.Tue, 13 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/cancel-adobe-and-have-claude-code-cut-caption-and-assemble-a-video-inhttps://agentplays.ai/uc/cancel-adobe-and-have-claude-code-cut-caption-and-assemble-a-video-in@thorgeirgisli cancelled Adobe and asked AI to edit a whole video. Claude Code cut the footage, made motion graphics, levelled the audio, fixed mic channels, wrote captions, then built the full project in FilmCraft, a free open-source editor. Screenshot of the FilmCraft timeline is in the post.Mon, 12 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/build-a-multiplayer-3d-capture-the-flag-browser-game-with-cursor-andhttps://agentplays.ai/uc/build-a-multiplayer-3d-capture-the-flag-browser-game-with-cursor-and@chongdashu built a fully multiplayer capture-the-flag 3D game with cartoon animals using Cursor and Gemini 4 Argon, after skipping earlier Gemini releases for game work. The game, Backyard Rascals, is playable in the browser at backyard-rascals.vercel.app. Video of gameplay is in the post.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-draft-weekly-client-recap-emails-you-only-tidyhttps://agentplays.ai/uc/have-grok-bot-draft-weekly-client-recap-emails-you-only-tidy@DerekatQuango used to spend 30–45 minutes a week writing a recap email to one client. Because Grok Bot already does most of the work on that account, it now drafts the weekly recap and they only clean it up — about 7–8 minutes a week, which they put at roughly 2 hours saved a month.Sun, 11 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-build-and-run-a-small-directory-site-then-add-a-sponsorhttps://agentplays.ai/uc/have-grok-bot-build-and-run-a-small-directory-site-then-add-a-sponsor@AlexFinn had Grok Bot build agentusecases.ai, a directory of real AI-agent use cases, then keep shipping fixes (they report 40+ PRs) while they ran the social side. The bot added a newsletter (1,500+ signups per their post), improved SEO, and set up a Stripe payment link for sponsors. They say the first sponsor bought a $500 spot (their own claim).Tue, 13 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/turn-a-lease-job-offer-or-contract-into-a-one-page-brief-before-you-signhttps://agentplays.ai/uc/turn-a-lease-job-offer-or-contract-into-a-one-page-brief-before-you-sign@RoundtableSpace posted a copy-paste prompt for Claude that reads a document you're about to sign and returns five plain-English bullets, every fee and penalty with its clause number, every deadline, risky clauses quoted with a one-line risk, what's missing, and the five questions to ask before signing. The post says it won't replace a lawyer but will tell you what to ask one.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-gather-your-documents-and-prepare-a-complex-overseas-taxhttps://agentplays.ai/uc/have-grok-bot-gather-your-documents-and-prepare-a-complex-overseas-tax@Texas_Twins_Dad says Grok Bot saved them more than $1,000 on a complicated tax filing with overseas reporting that would have needed a hard-to-find accountant. The bot found and organized the documents itself from their emails and bank accounts, worked out the requirements, filled out the forms and saved everything on its own computer; they reviewed and submitted.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-find-a-surprise-renewal-s-receipt-cancel-it-and-ask-for-ahttps://agentplays.ai/uc/have-grok-bot-find-a-surprise-renewal-s-receipt-cancel-it-and-ask-for-a@glenntkim noticed a recurring subscription charge on their card while driving. Through Grok in their Tesla they had their Grok Bot search old emails for the original purchase; with the receipt, the agent cancelled the subscription online and asked whether to request a refund of the renewal. They said yes, it emailed the company, and they say the refund arrived two hours later.Tue, 13 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/let-your-ai-agent-do-the-bookkeeping-legwork-in-dougs-while-you-approvehttps://agentplays.ai/uc/let-your-ai-agent-do-the-bookkeeping-legwork-in-dougs-while-you-approveTom Eberle open-sourced dougs-cli (MIT), an unofficial command-line tool for the French accounting platform Dougs, so an agent can work in their company's books. On their real books in one afternoon they report uncategorised operations going from 183 to 4, 22 missing invoices pulled from Gmail and attached, about 60 changes written and 3 overdue VAT returns surfaced. Nothing reaches Dougs until they approve the plan.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/share-the-grok-bots-you-already-use-as-public-templates-and-howhttps://agentplays.ai/uc/share-the-grok-bots-you-already-use-as-public-templates-and-howTo share a Grok Bot you already use, open its settings, choose Share as Template, review what's included (remove anything personal), then publish a public link; others tap "Add to Grok Bot". @Teslaconomics shared four bots they use daily this way: a CEO bot that routes work, a Doc bot that fixes team drift, a Grok Build bot and a Trippy travel planner. X also runs a Template Rewards pilot that is invite-only and U.S.-only; it pays based on how many people use a template and keep using it, and payments are not guaranteed.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-pull-the-weekly-spelling-list-from-the-teacher-s-emailhttps://agentplays.ai/uc/have-grok-bot-pull-the-weekly-spelling-list-from-the-teacher-s-email@ThriveWithSI built a small spelling-bee app to help their kids practice each week's words. Their Grok Bot now scans their email for the teacher's weekly newsletter, finds the word list, logs into the app and updates it for them, one of the small recurring chores they say keep compounding into free time.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/audit-grok-bot-routines-that-wake-up-only-to-report-no-changehttps://agentplays.ai/uc/audit-grok-bot-routines-that-wake-up-only-to-report-no-change@Nyxera0x described a seven-bot swarm (X, email, CI, deploys, docs, audits) where one routine kept running on a tight schedule and returning NO_CHANGE most of the time. The post includes a copy-paste audit prompt and says the dollar figures in the accompanying video are simulated; the pattern is to slow or event-trigger high NO_CHANGE jobs and keep risky actions behind human approval.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-a-grok-bot-chain-suno-midjourney-and-opus-into-a-music-video-teaserhttps://agentplays.ai/uc/have-a-grok-bot-chain-suno-midjourney-and-opus-into-a-music-video-teaserCreator @anabology published Claudia, a Grok Bot that uses Suno, Midjourney and Opus to make music videos. The public post shows a teaser Claudia produced with Imagine; a follow-up says that clip was one prompt after the bot was set up. The bot template is listed on the Grok Bot marketplace.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/move-claude-code-and-your-repos-onto-grok-bot-then-drive-it-from-mac-orhttps://agentplays.ai/uc/move-claude-code-and-your-repos-onto-grok-bot-then-drive-it-from-mac-or@ifourth's MacBook was lagging with Claude Code running all day across SaaS projects, so they moved Claude Code and all repos onto Grok Bot's cloud PC, use T3 Code to send tasks from Mac or iPhone, and Tailscale so only their devices can connect.Tue, 13 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/i-let-an-ai-run-my-startup-for-a-weekhttps://agentplays.ai/uc/i-let-an-ai-run-my-startup-for-a-weekAl of Tiny Projects / PromptBase ran Grok Bot as a mini team for a week: a support bot on Gmail, an engineer bot that writes code, a security bot watching Cloudflare and server logs, and a marketing bot tasked with 2% weekly social growth. Bots fixed a customer bug together without Al, who managed work from a phone on a 5-hour hike in Wales, and a project-manager bot shipped a secret 4am feature that warns users before declined-submission timeouts.Wed, 14 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-chatgpt-find-a-faster-version-of-a-hand-tuned-assembly-routinehttps://agentplays.ai/uc/have-chatgpt-find-a-faster-version-of-a-hand-tuned-assembly-routineDave Plummer, the former Microsoft engineer behind Windows Task Manager, asked ChatGPT to optimize a memcpy they wrote for the Motorola 68000. It proposed moving 12 registers instead of 11, unrolling four transfers per loop and ending on a source-address compare, for a modeled 9.6% throughput gain on long aligned copies (about 5.2% for a full 512-byte copy). The numbers are calculated from MC68000 instruction timings; ChatGPT said it had not assembled or hardware-tested the code.Tue, 13 Oct 2026 00:00:00 GMThttps://agentplays.ai/uc/have-grok-bot-check-this-year-s-tax-return-against-last-year-s-andhttps://agentplays.ai/uc/have-grok-bot-check-this-year-s-tax-return-against-last-year-s-andJeremy Judkins's Grok Bot flagged a mistake on their 2025 tax return on its own: TurboTax was spreading their Cybertruck's remaining basis evenly ($18,910 for year 2), but their 2024 return used a 5-year 200% declining-balance schedule, which puts year 2 at about $34,037. That left about $15,100 in deductions on the table. It had them import their web return into TurboTax for Mac and walked them through editing the forms. They say it saved them about $2,500 on their taxes.Tue, 13 Oct 2026 00:00:00 GMT