Post

We peaked at Pentium 4, then spent 25 years scaling out. Same with AI now: stop pushing the model, start surrounding it with concurrency, harnesses, and data validation. #AI #Engineering #LLM

Titus Kellog

@tituskellog.bsky.social

Yes - the useful harness is not just eval prompts. It is test data, adversarial cases, cost/latency budgets and rollback when a model change breaks behavior. Boring distributed-systems work, now pointed at LLMs.

DaShaun Carter

@dashaun.com

I don't think of distributed systems work as boring. But, I do think I understand what you are saying.

21/10/26, 6:18 pm0 reposts0 likes

Titus Kellog

@tituskellog.bsky.social

Fair correction 😄 I meant the familiar engineering work that tends to disappear from AI demos. There's plenty to enjoy in getting those failure paths right.