Have we entered the era of blindly trusting agents?
retained text
Ive been deep in reviewing PRs, architecture designs, even tickets and slack threads by engineers no less than "senior" level. I feel like im seeing increasing levels of quite obvious fundamental gaps in that I don't think would have been there maybe even a year ago. Don't get me wrong, I understand people make mistakes and we're all here to learn and grow from each other regardless of seniority. There's always room to learn something new, and that's not really what I'm talking about here. I'm more so talking about things that seem to contradict some pretty fundamental understanding about our systems and stack. When you've been working on a system for a while, building things together, making decisions, understanding why certain things work the way they do, I feel like you naturally develop some level of intuition about it, what we're trying to achieve, and that understanding kinda informs the development going forward. Now for sure along the way things diverge and new ideas take over but in all those instances there are fruitful discussions and brainstorms that always pave the way. But lately I've been seeing more and more things that just seem to come out of left field. Like I'd be reviewing a PR or an ADR and genuinely think why are we suddenly doing this? And doesn't this go against the whole reason we designed this thing the way we did in the first place? But what's actually worse is when I ask about why its that way, theres really no answer...like they cant defend it or provide their idea behind what theyve built and instead just "oh right, i'll change that". And it's not like these are necessarily new people who don't have the context. Even people who have been involved in building these systems, who were part of the discussions and decisions that made along the way. It can also be the opposite issue. I've come across changes where I can almost tell that a particular piece of code exists because there's probably some instruction in the AGENTS.md somewhere telling the agent to always do things a certain way. And sure, that convention might make sense for most cases. But these particular cases isn't one of them. If you actually think about what the code is doing and why, you'd realize that the convention doesn't really apply here, but again, the PR gets raised and when asked about it, there was again no thought or reasoning behind that. I feel like even with AI, things like that would've previously been caught by the engineer themselves (these very same people) before even putting up the PR let alone asking for a review. I thought the design, problem solving and creativity is the value that human engineers still provide in this crazy agentic era but is that just...going away too? Im also slightly sensing accountability and ownership of your work slipping away in place of blaming the agent... Idk if anyone else has ever experienced this lately but id love to hear other perspectives. This makes me kinda sad tbh :( hopefully these experiences are just outliers
document fetched 2026-10-26 00:23 · canonical https://reddit.com/r/cscareerquestions/comments/1xew4fw/have_we_entered_the_era_of_blindly_trusting_agents