There's a trend in software (well I might be conflating a few trends) that I think is not going to bode well for most: - With how insane coding agents and the models are getting, it's very easy to just build "everything" but we're seeing just insane scope creep that don't really serve the purpose of the "thing" the product wants to solve for customers - Everyone is going to collectively relearn the painful lesson as to why super apps have never worked in the west (spoiler alert it was never that "software was hard") - A lot of companies are calling themselves "labs" as a way to excuse themselves from building *one product* that people actually use (repeatedly) but rather multiple mediocre products (non-stop launches on the timeline), few are focused on actual "research" but rather are doing sub-par R&D or product development - Because new software/products/features are so easy to build, it's easy to find new things to latch onto (technological innovation) that aren't the thing you should care about (like retention). I agree with Nikita that technological moats are mostly over - When everyone is gunning to build *everything*, there's massive alpha in just building simple things that are focused and well-executed

A perpetual problem: the more visible activity there is, the more impressive the work looks. Ten years ago people were building unnecessary distributed systems, with >1:1 microservice:dev ratio, that could've been simple monoliths. The same thing is happening with "factories".

The introduction of the infinite optimization machine won’t create new supply of scarce assets (physical resources, attention, etc.) but will substantially/infinitely create new demand for them.

"Sorry for the delayed response! Middling generative AI has fundamentally altered the strategic balance between offense & defense in the war for our attention, so I have chosen to ignore this messaging channel. Out beyond kino & slop, there is a field. I'll meet you there."

We apologize for our mcp blog post being too aligned with the mainstream narrative this week. @panth_bhavsar originally drafted this when people hated mcp but now it’s not spicy. As penance, Panth will be consuming these for lunch.

a great post by @Theoryvc! They’ve spent a year building an mcp-powered company brain and share what it takes to make it dependable enough for daily work. plenty of useful detail on where agents get stuck and how better tool design helps, including a look at fastmcp's background tasks for long-running work.