AI Agent Workflow Complexity Is Drawing Real Skepticism
A post titled 'Backpressure is all you need' about reducing human labor in AI agent workflows is getting pushback on HN. The core skepticism: the whole industry seems to be over-engineering AI creative and coding workflows, when the most effective creative execution looks like tight micro-iterations, not elaborate agent pipelines. One commenter specifically flags the PR review problem, where AI agents generate low-quality pull requests that waste human reviewer time.
This is appearing alongside a broader pattern in both HN and Reddit founder communities where the initial excitement about autonomous AI agents is giving way to frustration with reliability and oversight costs. The promise was fewer humans in the loop; the reality for many teams is more humans doing different, often more tedious, work.
The key tension is between what agent demos look like and what agent deployments actually cost in human attention and rework. The gap between the two is where a lot of founder frustration lives right now.
So what?
Before you build or buy an AI agent workflow tool, measure the actual human time cost of oversight and rework, not just the time saved on the automated steps. A lot of agent ROI math looks good on paper but falls apart when you count the engineering hours spent reviewing bad outputs. The skeptics on this thread are worth listening to.