Cosmos & Inherent asks: How can AI address a specific coordination problem that has historically dogged human organizations? What would adopting this solution mean for the organization’s structure and the day-to-day experience of individuals within the organization?
- AI can radically change the conception of “who is in the org”
- historically, sharp delineation between “fulltime staff on inside” vs “users on the outside”
- especially true for consumer tech startups & companies. (eg Manifold, Google, Reddit, now OpenAI & Anthropic)
- What might be possible now:
- larger scope for what it means to be “in an org”. eg Manifold staff, but also moderators, volunteers
- and compensated accordingly
- weaker boundaries — more volunteers, work trials, part-time members
- orgs become more like collectives, communes
- why is it the case that most tech orgs are a just few fulltime people?
- small teams reduce communication costs
- specialization within a human expert
- more knowledge & skills accrues inside a single person over time
- but: maybe this will change — the AIs/agent transcript can specialize, while the human remains generalist, going around learning stuff
- humans remain:
- better at few-shot learning
- better at modeling other humans
- draw an economic boundary between a few extremely great people
- equity compensation
- but: leads to a more us vs them mentality?
- tech startups vs governments
- can AI lead to more new governments?
- trust, legitimacy, coherence
- (not sure that this can be passed on to AI actually)
- [ln] bifurcating humans? ai accelerating education vs getting left behind
- not sure if Austin means just 10% of population, or 100% (if we do more to carry them together)
- if 8b: conditioned on getting some things right. might be out of scope for the piece
- [ac] lydia’s approach vs austin’s approach, r vs k
- [ln] ai lets us do things faster — 20y vs 2y
- people move faster
- 1-day org
- minifest — Lydia first exposure to Austin universe
- you can join a lot of “1-day org”
- dunbar number for orgs
- [ln] not sure, if in this piece, do we expect more people to be like Austin?
- also fine (and exciting) to just do a small case study
- [ln] does Austin’s approach self-replicate better than other approaches?
- memes
- org culture
- journalist: not self replicating fast enough — why might that
- are my projects self replicating
- 90/30 club — i’m from CMU & NYU
- rate of self replication determines whether we see it
- CG/EA/constellation modes of doing things
- [ln] the way the world could be, vs how it is
- but psyched to put down ways the world could be
- (or, other people)
- not yet a reproducible recipe cookbook for “how to do what Austin does”
- what peter thiel did with 0 to 1
- can use the thiel cookbook from 0 to 1
- felt like PG’s essay
- nuno: vassals, grantmakers are orgs
- meta, google: food, healthcare, they provide for their people
- more normal disposition from people is to want a provider to take care of everything
- historically: family unit, to employer, more hierarchical and want one provider
- self employed entreprenuerial thing
- [ln]: accelerando, for institution building, compute rich world
- might be like the main character: venture altruist, make other rich, give ideas away for free
- why is it hard for an org to hire a lot of humans?
- trust:
- don’t know that a person is a good fit, worth the money. want to vet. adverse selection. resumes are low-signal. interviews are expensive, also select for “good at interviewing”.
- best signal of “can you do the job” is “doing the job” ⇒ work trials, or watching people do the job in other contexts
- sourcing
- not sure where to find the right humans?
- but AI should make matching markets much more efficient
- have your Claude apply for jobs, have their Claude interview your Claude to see if the fit is good. x1,000,000
- communication overheads
- dunbar’s number, one person can only model so many others
- already, internet & prosaic tech enables much larger/more profitable orgs, eg Google and Amazon
- internal tech like Slack, Google Calendar, Notion/GDocs allow for much broader scale of communication
- AI will probably go much farther — can simulate individual human
- costly
- humans cost a bunch of money!
- which serve as a major constraint on how large an org can grow
- maybe there will be much more kinds of work & participation which are less like “fulltime work”
- question to explore: why does facebook pay a facebook SWE (40h/week) a bunch of money, but not a facebook user (10h/week)?
- something about the facebook user’s contribution is much less valuable, from a supply/demand perspective
- facebook users love sitting around and chatting with each other. lot of supply of that in the market
- but: not that many people historically have the ability to maintain systems like facebook
- orgs are paying for:
- skillset, expertise
- reliability
- loyalty
- secrets
- why is it solved with AI? some guesses:
- historically, onboarding & context sharing is super expensive, but now AI can get past a lot of it
- example: Manifest. Winter & Carolanne just running Claude over the copious transcripts
- lower cost of comms between people?
- what is the equilibrium?
- maybe: instead of being associated with a single org, much more flexible membership
- ie people work 5 “jobs” each at 1-20h/week
- ie people associate with orgs the way they associate with social media sites today
- maybe: vast superorgs nation-level (think Terra Ignota’s hives) glued together with weird AI support & governance
- imagine X except the money feature was super well implemented, many people draw significant paychecks from their posts, there’s a taxation scheme & election for council & leader rather than just fully founder-controlled
appendix
- code: justwrite site self-modification
- organizations becoming more liquid, permissionless
- feedback
- users vs staff — now users can just make the change themselves
- historical problem: trust with power (budget, hiring authority)
- btw suggested pairings:
- charlie & jacob
- malaika & fizza
- austin & lydia