yeah I really love the 'intentionally temporary' framing, that's actually a way better way to articulate what was I was getting at than how I wrote it. The trauma of throwing things away is real, I see it constantly in DD. People treat migration like an admission of failure when it's usually the opposite, you outgrew something, which is a good problem to have.
yeah you're totally right, good catch. S3 has been strongly consistent since 2020. That weakens the specific framing but the broader point about Iceberg's design assumptions still holds: the architecture was built for a world of eventual consistency and massive concurrent writes, and a lot of that complexity doesn't disappear just because S3 improved. But I should have been more precise there, appreciate the correction.
Curious what you mean by "agent harness" here... are you distinguishing between true autonomous agents (model decides next step) vs workflows that use LLMs at specific nodes? I've found the latter dramatically more reliable for anything beyond prototyping, which makes me wonder if the "model improvement" is partly better prompting and scaffolding.