I think an interesting related question is what the best way to build the domain logic/workflow that an agent works in is. Nowadays I can reach for pi durable to handle the low level loops. Now I’m thinking about if it’s helpful or restrictive for any particular problem for an agent to have a state machine built into the harness or just stick to exposing tools etc, or something in between
The biggest one was column types like no uuid, timestamp, enum, etc but it looks like there were other schema features that would’ve prevented my drizzle generated migrations from running
PGlite also ships with a number of common pg extensions like pg_trgm which I wouldn’t necessarily expect from a pg compliant db but would block my testing use case
@jamespearce@tinybasejs Ran into a bunch of issues unfortunately. "supporting PostgreSQL-shaped SQL" is a bit misleading IMO, thought the "SQL compatibility" doc page is clear enough
@artman Most cutting edge real world factories have tons of brilliant engineers working at them who seem to like their jobs, not sure I’d equate working at a factory to stamping widgets
I think @jamwt’s builder vs engineer distinction is a useful perspective for software to adopt
I’m as excited about customizable software as anyone, but it’s going to be a maintenance disaster right? Like teams of professionals can’t indefinitely maintain software under an onslaught on feature requests, but the average person and a frontier model will?