I found the essay yesterday very interesting & have been thinking since about how to apply its' insights to my own circumstance. As a former software entrepreneur now retired & concerned solely with managing the apartments I bought in my pre-retirement phase, i find fundamentally rethinking my tiny enterprise workflow a challenge, as y'all suggested. But perhaps not only for the reasons U identified: As a 'getting to know U' phase of adopting AI, I am using Claude to code up a replacement for QkBks, whose rentier behaviors have become increasingly irritating. But I have no illusions - this is scarcely even the lightbulb stage of improvement, since some of QkBks functionality I won't be able to duplicate & will have to replace (e.g. my remote CPA being able to easily log in to see my records at tax time). Be that as it may, simply rebuilding software without changing its design & function is a bit of a knaves game, is it not? A more thorough (tho still quite limited) approach would be to replace my bookkeeper, & ultimately an even better goal is to replace most of myself. So why not focus on these end-points instead? At this point we run directly into 2 sorts of limitations. On the one hand, the need to securely connect my new software daemons ('agents' we're now calling them) to a bank feed to capture $$ inputs (rents) & authorize CC outputs (eg utilities). Even tho troublesome, these security-driven concerns are the simpler issue. The gnarlier bits are communicating with tenants, negotiating Leases, & arranging maintenance. I can foresee agent-written Leases, & perhaps even agent-scheduling, certainly agent-payment, but these are not items that are going gentle into that good night. My picayune troubles are hardly the stuff of shaping economies, but I would be interested to hear what thoughts this mundane example excites in your well-educated head.
This is well thought through. You're right that the relational problems will be the hardest because there is so much tension around AI. There is not much room to fail (and anyone trying to figure out stage 3 will fail lots of times). Likely best to start with the options both buyers and sellers would enjoy (e.g. if AI gets the plumber out quicker, great!). I think there will be a 'co-movement' towards solving these problems with the model makers and enterprise (FDEs will be a two-way channel).
I found the essay yesterday very interesting & have been thinking since about how to apply its' insights to my own circumstance. As a former software entrepreneur now retired & concerned solely with managing the apartments I bought in my pre-retirement phase, i find fundamentally rethinking my tiny enterprise workflow a challenge, as y'all suggested. But perhaps not only for the reasons U identified: As a 'getting to know U' phase of adopting AI, I am using Claude to code up a replacement for QkBks, whose rentier behaviors have become increasingly irritating. But I have no illusions - this is scarcely even the lightbulb stage of improvement, since some of QkBks functionality I won't be able to duplicate & will have to replace (e.g. my remote CPA being able to easily log in to see my records at tax time). Be that as it may, simply rebuilding software without changing its design & function is a bit of a knaves game, is it not? A more thorough (tho still quite limited) approach would be to replace my bookkeeper, & ultimately an even better goal is to replace most of myself. So why not focus on these end-points instead? At this point we run directly into 2 sorts of limitations. On the one hand, the need to securely connect my new software daemons ('agents' we're now calling them) to a bank feed to capture $$ inputs (rents) & authorize CC outputs (eg utilities). Even tho troublesome, these security-driven concerns are the simpler issue. The gnarlier bits are communicating with tenants, negotiating Leases, & arranging maintenance. I can foresee agent-written Leases, & perhaps even agent-scheduling, certainly agent-payment, but these are not items that are going gentle into that good night. My picayune troubles are hardly the stuff of shaping economies, but I would be interested to hear what thoughts this mundane example excites in your well-educated head.
This is well thought through. You're right that the relational problems will be the hardest because there is so much tension around AI. There is not much room to fail (and anyone trying to figure out stage 3 will fail lots of times). Likely best to start with the options both buyers and sellers would enjoy (e.g. if AI gets the plumber out quicker, great!). I think there will be a 'co-movement' towards solving these problems with the model makers and enterprise (FDEs will be a two-way channel).