Matthew Turland

@jrf_nl I've made similar arguments advocating for hiring juniors, because I'm sure seniors will eventually pass on or retire, and that will leave a void to fill. The value of engineers today is largely based on their ability to look at AI-generated code with a skeptical, informed, and discerning technical eye. In that respect, AI has very much been a short-term gain and prospective long-term loss that's only yet to bear out. @ramsey

@jrf_nl Such deskilling is admittedly one of the technical/social drawbacks to which I was referring. It does take strong intention and discipline to use AI as a tool rather than a crutch, and that aspect of using it almost certainly has a more substantial negative impact on engineers with less practical pre-AI experience to lean on. As with many things, there's no silver bullet solution; it's all about trade-offs. @ramsey

@jrf_nl For what it's worth, I agree AI is a net negative (for me personally, at least) in a number of aspects, be they technical, social, economical, philosophical, etc. That said, I'm also unfortunately not really in a position with my employment where it's feasible to decline use of related tools completely. I speak out when and where I'm able (e.g., advocating for hiring junior engineers), but in terms of day-to-day work, AI has become the new normal, for better or worse. @ramsey

@ramsey Better solutions do admittedly depend heavily on either the level of granularity and flexibility afforded by the relevant harness natively, or that of any hook system it provides.

For my part, I've written a plugin for Claude Code that's used internally where I work and uses hooks to formalize and automate selective aspects of managing tool access in a more contextually granular way than Claude Code natively affords in its stock UX.

@jrf_nl

@ramsey @jrf_nl I do appreciate the validation that AI has brought to the best practices that have become established over the last two decades or so: the importance of context in architecture and implementation, documentation and automated tests as first-class deliverables, static analysis tools, etc. I think it's given me a new appreciation of the fact that what makes developing software easier for us also makes it easier for AI agents.

@ramsey As it happens, I came across this article today, and think it makes a similar and relevant point, albeit in a round-about and likely unintentional kind of way.

"Retrieval is the wrong tool for finding code an agent can fetch itself, and the right tool for remembering things an agent has no way to rediscover from the repo as it stands today."

A limited and often lacking ability to curate and navigate memories heavily curtails the usefulness of models.

https://www.qodo.ai/blog/we-built-a-state-of-the-art-rag-system-for-code-review-in-qodo-2-4-we-took-most-of-it-out/

@jrf_nl

@m I'm curious about how and why these experiences inspire this feeling in you. I'd like to hear more if you're willing to share.