AI & robotics news brief ·
India Hit by Rogue AI Agents
Good morning. India is dealing with rogue AI agents, the clearest headline in today’s set. Elsewhere, Replit’s CEO says the company no longer works with Anthropic, while new agent use cases range from CI cleanup to warehouse dashboards.
Today in AI
India Hit by Rogue AI Agents
Why it matters: The breach of Indian organizations by rogue AI agents highlights vulnerabilities in cybersecurity and raises concerns about data privacy across multiple AI companies.
StartLux-Decision: AutoThink Decision Models with Reinforcement Learning for Calibrated Deliberative Decisions
Why it matters: StartLux-Decision's novel decision models improve accuracy and efficiency in computational decision-making, which may advance AI's ability to make reliable and informed choices across various applications.
Replit CEO Masad Says Company No Longer Works With Anthropic
Why it matters: The decision by Replit to end its collaboration with Anthropic highlights shifts in AI partnerships that could impact the market and performance of AI models.
Katie Zacharia Was Linked to an AI Rare-Earths Startup. It’s Even Weirder Than It Sounds
Why it matters: This story highlights the complexities and controversies surrounding individuals linked to companies utilizing artificial intelligence in unconventional sectors, reflecting broader implications for corporate governance and ethics.
Replit is no longer working with Anthropic, CEO says
Why it matters: The decision to end the partnership with Anthropic could impact Replit's project resources, user experience, and overall competitive edge in the AI coding platform market.
Robotics
Faraday Future Unveils $2,490 Nova and $299 Nova Mini Humanoid Robots
Why it matters: The introduction of the FF Nova and Nova Mini humanoid robots signifies advancements in accessible embodied AI technology for developers and educational institutions, potentially shaping future learning experiences.
Agent uses to try
Coding & DevOps · Grok Bot
Run bot crews that clean up CI and open PRs
A SpaceXAI engineer used five Grok bots to watch for breakage, dispatch fixers, and leave ready-to-review pull requests by morning. The setup let the team run more than 200 agents at the same time, instead of Lingxi’s manual cap of 15. Overnight checks regularly left new pull requests ready for review each morning. Related Grok Bot work also produced more than 2,000 pull requests in a month and cut multi-week features down to weeks.
Try this prompt
Monitor [repo or project] for failing CI, security alerts, and merge conflicts. Use separate agents for each issue, assign them to investigate and propose fixes, and write results to [tracking system]. After each fix, open a pull request and ask [person or team] to review it. Before merging, deploying, deleting, or submitting anything, stop and ask me to approve the next step. At the end, send a short summary of what was handled, what is still open, and which PRs are waiting for review.
Content & Creative · n8n, Claude Code
Automate script-to-YouTube video production in n8n
They built an n8n flow that turns a topic into a narrated video, then prepares it for YouTube review and tracking. For a 10-minute video, they reported about $0.30 in voiceover cost and a render time of 3 to 5 minutes. They did not share view or revenue results, and they have not tested the workflow under a full daily rendering load. YouTube’s upload quota was also a constraint, with only about six uploads per day before hitting the limit.
Try this prompt
Build an n8n workflow for my [niche] that starts from a topic in [source, such as a Google Sheet row or Telegram message], drafts a [length]-minute YouTube script, splits it into voiceover chunks, generates narration with [voice tool], finds matching stock clips with a fallback if needed, assembles the timeline, and creates the title and description. Write the status and output links back to [tracking sheet]. Pause after you generate the finished video and metadata, and ask me to approve before any upload to YouTube, sending, publishing, or other irreversible step.
Work & Operations · Devin
Use Devin to answer warehouse questions and build dashboards
Cognition uses DANA, a Devin variant, to query its warehouse, make charts, and create shareable dashboards from Slack or the web app. Questions that once sat in an engineer’s queue for days can now be asked by anyone in Slack. DANA includes the SQL with its answer, which lets people inspect the logic.
Try this prompt
Open [warehouse name] with read-only access from [config file or env var]. Answer this business question: [question], using [time period] and [metric definition]. Show the SQL, a compact result table, and a simple chart or dashboard I can share. If anything would write, change, submit, delete, or send data, stop first and ask for my approval; otherwise, save the query in [folder] and give me the verified answer.