There are also obvious things like better compute and power. People should try and work in areas where people won't want to utilise LM agents.

(or maybe just work very closely with LMs to "feel the AGI more" and expect to pivot a bunch - which might be what the YC startups end up doing I guess)

* Maybe some kind of agent workforce management software (think Slack for the huge LM agent worker army) though it's plausible labs will have a strong edge on this one.
* Things that help you build data centres much faster than they can currently be built

* Security infrastructure that's expensive to build and can be sold to multiple labs (e.g. FPGAs to encrypt traffic between data centres).
* Physical Infrastructure Management (e.g. cooling systems for AI hardware)
* Products that help AI labs deal with regulation (especially from the EU).

A bunch of YC startups are wrappers around LLMs which, if successful, will probably end up being copied by an AI lab (e.g. an AI code editor).

If I were starting a company I'd try to build products that AI labs would rather buy when they have much more money than ~everyone else.

Vague idea 🧵