I'm using Claude all day at work and deepseek privately for very similar tasks. The price/performance difference is extreme.

Indeed, the acronym ADR is extremely confusing for everyone knowing Architecture Decision Records. Obviously i (and many others) misunderstood and so, my comment is moot.

From your screenshot alone I can tell that those are not architecture decision records on the left side.

Ask yourself the question for each: is this truly an architecture decision or merely some design decision or requirement that now got some option a/b/c slapped onto it?

In the end you will just have a repo with hundreds of meaningless ADRs that your agent has to wade through.

Would you mind, if you could suddenly, over night, be a carpenter, a watch maker or glassblower?

as I spent a substantial amount of time in the console, even today, i sometimes need to think of some complex combination of git/find/ffmpeg/xargs and friends and so i've been using https://github.com/kagisearch/ask for some time.

Since it appears to be unmaintained i started my own version (yes, thanks to some LLMs work) and i am quite happy with it. It appears to have gained a lot more features than i initially thought but there it is.

My personal highlights are:

- can install shortcuts for bash/fish/zsh for pressing CTRL-W and having the current prompt replaced with the "answer"

  $ find all images in this directory

  $ find . -type f \( -iname '*.jpg' -o -iname '*.jpeg' -o -iname '*.png' -o -iname '*.gif' -o -iname '*.bmp' -o -iname '*.webp' -o -iname '*.svg' -o -iname '*.tiff' \)

- there is also a review TUI that explains the commands (which can become complex) or retries with another model

In any case, it's open source, maybe someone else will find it useful as well.

Comments URL: https://news.ycombinator.com/item?id=49656218

Points: 2

# Comments: 0

In the end, the new implementation added some new (maybe unique) features, like pressing CTRL-W in a shell prompt and having the text being replaced by some proper command to be executed with enter.

Here is a quick glimpse of what it looks like the terminal:

  $ find the top 5 largest files ever committed in this git repository (-> hit CTRL-W <-)
  $ git rev-list --objects --all | git cat-file --batch-check='%(objecttype) %(objectsize) %(rest)' | sed -n 's/^blob //p' | sort -rn | head -n 5

Comments URL: https://news.ycombinator.com/item?id=49518777

Points: 1

# Comments: 0

I think it's most fair to compare the plain token pricing that is used by everyone.

This is just an assumption and the whole article falls flat if this turns out to be wrong. In my limited (as everyone else's) experience, working with agentic AI needs good documentation, good specification (spec driven, you know it's all the hype nowadays). Those alone lead to much improvement. Now take into account that probably your senior dev also has more time to think about the big picture, to improve all those little things that were a nuisance in the past but now are a mere "Claude, fix that" in a worktree away.. I would not bet on the assumption of this article.

For example, this issue aside, I'd rather split such a workload into multiple smaller instances, naturally. Because the impact of a crash on this single node, heavy load, many cores, many clients scenario would be huge.

Since we will never know it might be a good idea to feature gate the change, change the default and let users decide to change it back. This may give some feedback on the lkml or else to decide if the change is worthwhile?

Comments URL: https://news.ycombinator.com/item?id=47851734

Points: 28

# Comments: 2