Last two weeks in AI: OpenAI, Google, Airtable, Cursor, CoreWeave, Stripe

Hey there,

Two weeks at once! No newsletter last week (I moved), so covering all of August so far. Let's get into it:

1. ​Astra release delayed as OpenAI hardens cyber concerns. OpenAI says it cannot rule out Astra reaching the "Critical" cybersecurity threshold under its Preparedness Framework. Previous models were rated High. Sam Altman chimed in adding, ".... we do not think it is a good strategy to keep powerful models to a chosen few." Fingers crossed that the US gov lets OpenAI ship Astra soon, I've got code to write.
2. ​Google announced broad changes to DeepMind. Four important technical leaders left to start their own company. Demis Hassabis is stepping over to chair/chief scientist but is giving up day-to-day operations of DeepMind. Jeff Dean and team really do have quite the resume (see a slide from their pitch deck at the end).
3. ​Airtable sells to Bending Spoons and spins out AI-focused Hyperagent business first. Everyone was shocked at the price, but the spin out makes the math square. The fast growing part of the business ejected before the purchase happened, so you've got a much slower growth remainder with no AI story.
4. ​Cursor acquisition finalized as the team launches new models + Grok Bot. A seriously busy 14 days for the Cursor SpaceXAI team. Lots of great ships. We might look back years from now and say 60b was underpriced.​
5. ​CoreWeave reported revenue up 112% from a year ago. More interestingly, they're still renting out NVIDIA A100s through 2029 at the same or higher rates. That's a 6 year old chip 😅
6. ​Stripe acquires OpenRouter for a reported 7 billion. Stripe handled 1.9 trillion in payment volume in 2025. I wonder how much of that was tokens? Given this purchase, my gut says the token spend growth trend isn't slowing down anytime soon.

A few updates from me:

• ​Human in the loop is a local maximum. Lots of systems add human in the loop by default, but it will never allow you to fully transform the task at hand. Think through how you could gain confidence without a human reviewing every LLM output.
• ​RuneScape created more economists than The Economist. A bot-only RuneScape server produced its own economy. Cheap labor made basic goods plentiful, scarce resources gained value, and inflation pushed agents toward barter.
• ​How to interview candidates in the age of AI. Leetcode is dead. Manual coding with no LLM tooling doesn't show you how good someone is at leveraging these new tools. Design the interview to be as close as possible to a day in the life.

As for this week, it seems quiet — almost too quiet. Would be a great time for someone to drop a new model.

LFG,

Drew