I agree that we need to understand what we're building, but I disagree with sticking with code as the abstraction of choice.
I also disagree with Claude being a sloppy intern. It's more of a single minded mid level or even senior. The code it writes is great, but mainly in the direction you have guided it. Which is fine if you know everything (how a lot of engineers started 2026), but now that we haven't written code in 9 months, our guidance is no longer necessarily the correct direction.
We need to get back to understanding what we're building so we can design/collaborate/push forwards, but without losing the efficiency of agents writing code for us. How do we do that? We need a new abstraction.
(full disclosure I'm building something in this space, mapping codebase -> architecture and keeping it updated on each commit: https://www.archme.io)
I agree that we need to understand what we're building, but I disagree with sticking with code as the abstraction of choice.
I also disagree with Claude being a sloppy intern. It's more of a single minded mid level or even senior. The code it writes is great, but mainly in the direction you have guided it. Which is fine if you know everything (how a lot of engineers started 2026), but now that we haven't written code in 9 months, our guidance is no longer necessarily the correct direction.
We need to get back to understanding what we're building so we can design/collaborate/push forwards, but without losing the efficiency of agents writing code for us. How do we do that? We need a new abstraction.
(full disclosure I'm building something in this space, mapping codebase -> architecture and keeping it updated on each commit: https://www.archme.io)