Hey friends,
I hope everyone is having a great weekend. A lot of the discourse this week was around rogue AI and the related security risk. More on that in next week’s round-up.
If you’re new here, every Sunday I curate the top AI tweets, podcasts, blogs and tutorials (so that you don’t have to).
Let’s jump in!
If you only read 3 things
How Bridgewater Built an AI Analyst That Does Hours of Expert Research in Minutes — Very cool interactive demo on building a hedge-fund specific chat agent on your internal data (and with guardrails).
What nobody tells you about writing agent skills — Is it time we stop using “skill creator” and write our skills by hand?
Claude Design is here, can it make consulting slides? — I’m a bit late to Claude Design, but it looks promising.
Build your own agents in Claude Cowork (without any coding)
Imagine sitting down Monday morning and your meetings are already prepped, your inbox is triaged, and last week’s notes are organized — all before you open your laptop. That’s Claude Cowork running on a schedule you set once. This self-paced course teaches you to build that system. Finish it in a weekend.
Use code FABLE-5 to save $100 until August 21st.
From the desk of Khe Hy ⤵️
New from the labs
Using OpenAI’s new voice feature to manage everything in your life — I have been using this new “form factor” and it is changing how I interact with AI. (And I’m not being dramatic.)
Auto-mode is now the default permissioning for Claude Code — No more living on the edge with “dangerously skip permissions.”
The first OpenAI gadget is reportedly a hockey puck-sized smart speaker — Looks like we’ll be adding an Alexa look-alike.
AI and the future of work/investing
Why some junior employees work well with AI (and others don’t) — Are you an AI amplifier? Fluency doesn’t drive success — that comes from problem framing and challenging assumptions.
Chamath’s AI investing guide — He’s bullish on land/power/shells and applications. And bearish on neoclouds and silicon.
Why isn’t everyone using agents? — SF engineers don’t understand why normies don’t have agents. Here are some reasons.
Changing how we build (from the head of CoPilot) — AI adoption is more than just improving existing processes. It involves identifying the new bottlenecks and reimagining the “job itself.”
Bridgewater’s AI analyst pairs internal data with guardrails—what builds trust fastest in a high-stakes agent: source transparency, human review, or measurable decision quality?