Fast feedback loops are a growth flywheel. In Claude Code's early days, I'd fix every piece of user feedback within minutes — sometimes seconds. That speed makes people feel heard. Feeling heard makes people give more feedback. More feedback = better product.
Boris Cherny
@boris-cherny
This is the world now.\" What's the next big shift to how software is written that either your team is already operating in or you think will head towards, covering engineering tradeoffs, product desi
Three layers of AI safety: (1) alignment/mechanistic interpretability — watching what neurons are doing, (2) evals — a lab Petri dish with synthetic scenarios, and (3) behavior in the wild. As models get more capable, layer 3 becomes the most important.
I have never enjoyed coding as much as I do today, because I don't have to deal with all the minutia.
Latent demand is your best product signal: if people are hacking or misusing your product to do something it wasn't designed for, build a product for that. Facebook Marketplace came from 40% of group posts being buy/sell listings.
Since introducing Claude Code, Anthropic has roughly 4X'd the engineering team — but productivity per engineer increased 200% in terms of pull requests.
Coding is virtually solved. At least for the kinds of programming I do, it's just a solved problem — Claude can do it. Now we're thinking about what's next.
At Meta working on code quality across Facebook, Instagram, and WhatsApp, a gain of a few percentage points of productivity in a year with hundreds of engineers was considered great. Now we're seeing gains of hundreds of percentage points.
We saw people using Claude Code to grow tomato plants, analyze their genome, recover wedding photos from corrupted drives, read MRIs. Non-technical people jumping through hoops to use a terminal. That's latent demand — so we built Cowork.