I've been trying to formulate this thought: AI lowers the barrier to useful proofs of concepts which raises the requirements for what software has to provide to make a purchase.

Can they rebuild Calendly? No. Can they build the few pieces they need to get started without it? Probably.

Not specifically yet but no concerns about it. Wide variety of server and client tasks in the last week.

It does do that often, even unprompted in the middle of a task. Have been pleasantly surprised. The only drawback: it’s almost too human in that it will regularly expect *you* to implement things rather than using the tools it has.

Ah, Claude removing things and the human not noticing! Here's a former version and I think they fit closely to this part.

I've found this thread to be super interesting so I've continued to think about it. I agree it's difficult to map to 2 axes. I think how you defined it in v2 makes sense but I don't seem to naturally think about it like that. Attached my current mental model (also Claude artifact).

Few 💭:

Reliability !== Reliability: Constraints that make a predictable (reliable) output don't mean that output is reliably useful.

Consider: Autonomy as an axis instead of Reliability. Generalization and Reliability have quite an overlap here (imho).

Top right is missing examples. Holy grail?

That’s where my thought process went: how do you do *anything*? Where is the line?

Hot take: this is all the python packages. Ones typescript-first seem to be more focused on DX. Curious which you tried…