I am so annoyed… I am trying to verify the claims of this article with my friend Claude but daddy Dario won’t let me because “cyber”. I pointed it at this article and a PoC [1] I found on DuckDuckGo. We live in a stupid world. I can’t run it right now so I am trying to figure out if it’s serious.
I’m old enough to remember when Wikipedia was not allowed to be cited in high school papers because it wasn’t a sufficiently authoritative source. How the turn tables…
That time was also when people could just add useful information to articles without some self-declared gatekeeper/“curator” removing everything they cannot grasp or simply deem uninteresting again.
But my attempts to make him a page on English Wikipedia keep getting rejected -- failed to establish notability. The fact that he won some awards and kids read his stories on the other side of the planet is apparently not good enough for Wikipedia.
Looks like your drafts didn't have much in the way of sources. Someone else did a complete rewrite in September with additional sources to establish notability.
This is true. I think the problem has changed but it still there. Don’t get me wrong, I don’t think it’s a reliable academic source. I just think it’s ironic that an academic is saying it’s a moral duty to protect it.
“Do not be rash with your mouth,
And let not your heart utter anything hastily before God.
For God is in heaven, and you on earth;
Therefore let your words be few.
For a dream comes through much activity,
And a fool’s voice is known by his many words.” -Ecclesiastes 5:2
"Answer not a fool according to his folly, lest you be like him yourself.
Answer a fool according to his folly (some translations "as his folly deserves"), lest he be wise in his own eyes." - Proverbs 26:4,5
I think part of what's happening is that the bad ideas and bad reasoning get diluted among 9000 words and 100 references. It's hard to put your finger on one sentence and say "here's where you're wrong", because the conclusion does not (obviously) rest on any one sentence. It rests more on the volume of words.
But "proof by volume" is not actually a thing. (I mean, it's a technique that can be effective in practice, but logically, it's not valid whatsoever.)
How do you answer something like that? If you try a point-by-point rebuttal, you're going to be into it 9000 words yourself (at least), and it is rarely worth it. I think that's "don't answer a fool according to his folly".
"Answer him as his folly deserves" is less clear to me. Is that to just flat out say that his ideas are foolish? Is it to give a concise one-paragraph statement of why his conclusion is folly? I think the latter is preferable (if you can). I mean, the rationalist is just going to claim that you haven't engaged with his arguments, though you can reply "no I didn't; your conclusion is so wrong that I don't need to engage with how you got there". And the concise rebuttal may be helpful to bystanders.
I still haven’t found an IDE that I feel really helps me develop with AI better than the command line. I have built bespoke tools that work the way I want. But a really good integrated solution would blow my current workflow out of the water. No one seems to have built an IDE with deep support for worktrees, parallel session, walking me thru diffs, and optionally handing control back to me to write the occasional hand-written code myself. Plus most of my work is done on mobile, so Jetbrains has an amazing opportunity…but they seem to be chasing the competition instead of focusing on their moat.
Right now there's a huge friction with IDEs, that if I have it open while claude is working on something, then any caches it has get stale very quickly, and it churns processing trying to keep up with edits.
Any leverage that claude could benefit from LSPs or language/AST MCPs is undone by that tooling not being able to keep up or stay fresh, so the IDE just "runs hot" for no purpose.
Which means I'm incentivised to close my IDE while claude is working, and now I've got used to not looking at the code.
Like you, I really want a system that works better. I want claude to search and edit through better tooling, it would be faster, more accurate, and should in theory be able to leverage compilers to apply edits better and more accurately.
I don't want claude wasting an awful lot of time grepping for things that a compiler cache can answer, but if that compiler cache can't keep up then it's actively harmful.
I'm sure there must be a better way, there must be a way to have smarter more change-aware cache invalidation, and smarter more cache-aware claude tool use, but so far none has happened.
Long term I'm sure the pairing will emerge, but unfortunately I don't think that's going to come from the IDE vendor side, it'll have to come out of some harness research lab, where Anthropic or a competitor finds a way to reliably leverage this stuff for a speed and accuracy boost.
It's completely bonkers to me that an LLM can almost "natively" read profiler output, but doing so involves vast amounts of text and context ingestion, when there are real tools sitting there that it ignores in preference for either reading vast amounts or iffy python one-shot scripts to parse stuff.
This 100%. My current workflow is switching between CC and VS Code to review code, and it's extremely clunky. There is still not a single harness/ADE that can do everything correctly in one tool.
Yeah, one limit for me is I need to use the subscriptions, so I'm tied to the Claude Code and Codex harnesses. The best solution IMO is a "bring-your-own-harness" one, where you can drop in whatever harness works the best for you and your development environment can be driven by it.
Air itself as well as the new Air plugin for the IDEs and the AI assistant support ChatGPT subscriptions, though you may have to disable JetBrains as a provider so that it's not billed against that. (It's weirdly called logging ot of JetBrains, though you still remain logged in to JetBrains for all other purposes.)
Anthropic is a different story, but I think it may still be possible via the Claude ACP agent (JetBrains' support of ACP is generally pretty good; I use it with anothe harness for open weight models regularly).
Maybe something like delta? This is still pretty new from the folks that make Zed, but looks promising and would address some of the things you mention https://delta.dev/
Cursor has support for work trees, parallel chats, will walk you through diffs line by line. This is what I primarily use for the walking through diffs feature. If I need something to just throw over the wall, I fire up claude.
When you are paid to sit on your ass to review as much generated code as you can, or to create as much good code as possible given the constraints of a limited amount of tokens, sure. But when you're on your own clock, the trade-off between latency, token budget and quality shifts considerably. And when it actually is a side project (yeah, many meanings of "serious" will be out, I give you that), mobile is just awesome.
That slow trickle of back and forth that gets our intelligent machines called "toasters" in a way not predicted by Battlestar Galactica at all, that really starts shining on mobile. Hands off on terms of signing off every little detail, hands on in terms of never being too far out of the loop, keeping the unattended steps small.
I'm not sure what the problem is after rereading their post? All you need to support everything they said is a remote connection to a heirarchy of worktree-supporting sessions per project, with each having a proper scrolling feed of agent actions taken and your input box.
> I still haven’t found an IDE that I feel really helps me develop with AI better than the command line.
With Codex GUI app and Astra you can communicate through screenshosts and collages. WAY faster and more precise for many types of jobs and workflows. Literally just scribble crap into the screens and it knows... no need to write out 50 paragraphs of prose to describe a button and its position.
I will have to try it out. I haven't tried it yet mostly because it seemed like it costed extra or would put me on API pricing. I have tried a workflow with Android Studio with Claude Code, but whenever I opened a worktree it took 5 minutes for Gradle to sync.
Sublime was/is still the best, unfortunately had to finally abandon it due to lack of first-class agent support. Hard to find something that emphasizes core text editing first and that doesn't constantly have some new update to apply on start that may break or change something.
It feels like JB is trying that, but it hasn't materialized in a way I understand yet; they have a handful of AI tools, and I don't fully understand them at this point, nor how they integrate and what roles they fill. (June, Air, AI asssistant, AI autocomplete)
Orca is too maximalist IMO, and it isn't a lot of work to switch together something with Herdr and a few vibed up plugins to get much (if not all) of the benefits that matter to you without the complexity and weight.
I agree Orca is packed with functionality but I don’t know what “weight” or “complexity” is. It’s just an app on my machine, not some software I’m maintaining. It can be the most complex system for all I care behind the scenes, as long as it runs fast and satisfies my needs I’m pleased - which so far it has been and more. I want to spend as little time thinking about my dev environment and the most amount of time thinking about what I’m developing.
To each their own though, I’ve heard great things about Herdr setups.
I was 8 with an eMac, running OS X Tiger. I had whatever software came on disks or bundled with the OS, and was interested with playing around with whatever I could on the machine. I didn't really have anything specific, but I just wanted to get it to "work".
reply