← Back to context

Comment by ineedasername

19 days ago

coding agents are an order of magnitude better than they were less than a year ago: Claude code, arguable the best, didn't exist a year ago and nonetheless, is an order better than when it did in late February.

It's really not a great sense of things for someone working in tech to make such a strong statement as the above given these circumstances. There is no reason to expect improvement will not continue, that agents' ability to review code-- drastically improved even since May/June, won't improve and, to the extend your concerns are valid about current agents, be able to manage the review and surfacing of issues for review and remediation.

This also ignores the growing awareness within the community of professional developers that, yes, blind trust-- as naively but minimally understandable given lack of knowledge on capabilities-- that blind trust cannot be given, and reasonable review of generated code at the end of a session is good practice. Developers are quite often already changing their habits with agents, if they were even avoiding such review before.

As for companies that may fire developers, rely near completely on agentic coding? I agree with the sentiment, but I really have not seen that happen to any great degree.

I won't assume this is the case for you, you may very well have dug in depth and worked with agents like claude code, understanding how best to scaffold by creating skills and agents, and still come to your conclusions. I would hope so. For the most part though, a large number of the people making comments of this sort seem to betray directly or indirectly that they-- nearly by definition of distrusting these agents and therefore avoiding them-- have little idea of what they are dismissing precisely because they have not taken the time to do more than dismiss them after little more than cursory review.

At the same time, many notable developers of significant accomplished near daily comment in pages here sharing experiences of true capability and assistance by these agents.

> the growing awareness within the community of professional developers that ... blind trust cannot be given, and reasonable review of generated code at the end of a session is good practice.

This saddens me. It saddens me that this is a trend among professional developers. It's an old lesson. We've known it for decades. Running arbitrary code is probably not the best idea.