FirstClassTree

Author here, around for questions.

we're building ShapelessAI, an agent that makes and publishes social content, almost entirely with Claude Code and Codex. One failure worth separating from code quality: agents claiming they verified something they didn't. A test result and an agent's summary of a test result are different evidence; delegating the code doesn't make that distinction go away.

i'm building ShapelessAI, an agent that makes and publishes content, and long-running jobs dying is a real pain. The failure case I'd want demonstrated is a publish succeeding remotely, then the worker dying before committing its continuation. How does an explicit durable operation recover when the external API offers neither idempotency keys nor a way to reconcile the result?

i've had long-running coding-agent jobs die. I'd like to see the 105-session benchmark include an unclean exit halfway through a task: what does the next agent know was completed versus merely planned? Recovering architectural decisions and recovering unfinished work seem like different tests.

You have many errors and I personally find the writing style pretentious.

For example: Software engineer role is about automating people -> often not.

That just indicates lack of rigor. Also, if so. Who will make the ai automate people? GOD? People think poorly understood theory and gradient descent will produce God.

How is software verifiable whatsoever for correctness? Ironically hardware is

I mean pure functions maybe verifiable for correctness, but I don't know how you verify facebooks codebase is 'correct'

Maybe try to run a maximizing function on the stock value, very unfeisable ofcourse