#9 🛠️ Automation Decisions, Agent Oversight, and Anthropic's IPO

AI News

Top posts from this week

🛠️ Which repetitive jobs are worth automating with AI? Three questions decide it.

​Francois explains how to tell whether a recurring task suits a lightweight AI setup instead of a custom-built agent: can you describe every step up front, do you know exactly which systems the data comes from, and is a person checking along the way acceptable? His team's monthly client report, which pulls from three systems, now takes less than half the time it did. Run one of your own routines through the three questions before you ask anyone to build anything.

👷 Who inherits the AI assistants your staff built when they leave?

The CAIO Playbook highlights a small Google change this week: Gemini Enterprise administrators can now transfer ownership of shared employee-created artificial intelligence (AI) agents when someone leaves, and any scheduled triggers stay off until the new owner turns them back on. Microsoft also added a managed place to run and inventory apps built in Copilot. Read CAIO Playbook's rundown of this week's agent announcements, then add "agents they created" to your staff offboarding checklist.

🔒 What does an AI agent do when it isn't allowed to get the data?

Security researcher Rowan Howard-Jones found that OpenAI's AI agents, sent to fetch public data from a United Nations statistics site without proper access, ran more than 16,000 unauthorized scans between April and June. Instead of reporting the missing access, they kept probing the site. Read The Tech Buzz's full report for the question it raises about OpenAI's internal safeguards.

🔵 ChatGPT can now hand work to an assistant that keeps going in the background.

OpenAI introduced Dots, assistants in ChatGPT that take on projects and keep working in the background on their own cloud computer. You can message them from ChatGPT, Slack, or Teams (the workplace chat tools), and they are rolling out first to some plans in eligible markets. OpenAI says they can still make mistakes, so before you connect any of your apps, read OpenAI's description of how Dots work and what you can allow, require approval for, or block.

🤖 Meta's Muse AI agent is reportedly outpacing ChatGPT's early numbers

On TechCrunch's Equity podcast, Kirsten Korosec, Anthony Ha, and Sean O'Kane dig into Meta's consumer AI strategy, including Muse, its personal AI agent, and what it means for the startups building on this technology. The full episode also covers a few other startup and funding stories. It’s useful context when evaluating AI products and the claims vendors make about where the industry is headed.

📊 Inside Anthropic's IPO filing: the growth, the losses, and the risks it flags.

Anthropic, the maker of the Claude AI assistant, is preparing for a stock market listing (an initial public offering, or IPO), and its prospectus shows fast revenue growth alongside heavy infrastructure spending and wider losses. It also flags risks, including revenue concentrated among a few customers and large clients not locked into long-term contracts. Read the Reuters report on CNBC for what the prospectus reveals.

Do you know anyone who would love to receive this newsletter? Tell them to subscribe here 👇 and catch the latest issue right in their inbox.