Oh it's Jacob - I used to work w/ him on his acquired startup, Pluto. He was very AI-pilled; I'm not surprised at the results. I remember wading through the AI-generated Python he had written. I think Pluto had something like 500k or 700k lines after 18mo. Everything was duplicated; there was no reuse of anything. Dozens of features, which mostly worked, not all did.
As a product that I would regularly use, Pluto missed the mark for me. As a SWE in 2023, I couldn't meet Jacob's mentality. Today, it's how a lot of us ship software.
I didn't love the product but I have to respect the process: he was able to sell his company. For some, what he did is success.
To me the main difference is that there is no staging (so no stashing needed), and conflicts are first class citizens. In practice this means that you can usually halt your work halfway, switch a commit, rebase everything onto head and continue working even if one of the branches off your commit has conflicts. In git you would have to resolve those before the rebase can finish.
Since conflict resolution these days mostly boils down to “hey LLM, resolve conflicts” this is much nicer.
Also jj tracks every change to the repo so undo works much better and is more readable than reflog
My favorite feature is the built-in distinction between what's mutable and what's immutable in my history. I can freely modify any commits I make locally without issue, since no one else would be able to depend on them, but if I push something or fetch it from a coworker, I have the safety of not needing to be careful when doing the equivalent of an amend or rebase because it will fail with an error telling me to use a flag to bypass the immutability protection if that's what I actually want.
Not having to stage changes in lieu of them automatically getting picked up was a bit of a learning curve for me, but now that I've gotten used to it I like it a lot better and I don't think I could go back. The equivalent of the reflog pretty much always has my changes in it if I accidentally lose them without manually saving due to the snapshotting, and I have no need for a stash because commits and branches are essentially just recording a checkpoint rather than being things that feel "heavyweight".
As an added bonus, dealing with conflicts is a lot less annoying because each commit records its conflicts separately, so fixing them doesn't require doing a stateful pipeline of changes in a big rebase; I can just do them piecemeal, in any order I want, and even just check out other stuff and work on them before coming back later if I feel like it.
I was "that guy" who would sometimes have coworkers come for help with git problems (https://xkcd.com/1597/ alt text). I've forgotten most of that knowledge (and all of that muscle memory) and didn't have to learn nearly as much to replace it with the ability to do a lot more with a lot fewer mistakes.
Obvious work that is behind state of the art. Not field defining. A reasonable paper to publish at NeurIPS or ICML but very middle of the pack.
If your paper is accepted to Nature it should be among the top results in your field for the year. This is just fine.
Edit to clarify: Nature has not been, historically, a venue for pure machine learning papers. It's been a venue for field changing work in the physical sciences. They already have a Machine Intelligence subjournal.
What this paper shows me is how desperate they are for ML papers, and how poorly their staff understand the field.
I think nature sort of has a reputation for sensationalism / probably overhyped (or outright wrong) stuff in the sciences now anyway... would not surprise me at all that they're desperate for ml stuff :-I
One thing about these tools - and, as discussed, there are many - that they mostly seem to share is that they all show coding agents work as a transcript of a conversation.
Why?
The underlying conversation has nothing to do with the work. That there is one context window is an implementation detail.
I've built one of these for myself, and instead it shares questions, plans, findings, assumptions as distinct UI elements. The conversation is hidden: your focus is on "what's next".
Are there other meta-harnesses doing the same thing? I'd love to stop building my own and use a product someone else has already built!
Why not keep building yours? It will quite literally be the same or better than 99% of these tools.
I also don't care much about individual conversations with agents as the main way of managing work, my personal dashboard shows projects (and subdivided by work trees underneath) but I have an orchestrator agent who is the one I talk to 90% of the time and who is responsible for summarizing information from various project manager agents and only update tiles in the dashboard when I actually need to do something, in the form of questions I can answer directly.
Over time as I add more preferences for how to work, the orchestrator sends fewer and fewer requests to me that it can handle itself.
Well, I do love building it... but it's a bit of a time-sink, y'know?
I used Orca for ages and that was great; I'm just tired of having to scroll back through a convo to find that "one last blocking question" that Claude or Codex spit out 15 tool calls and subagent messages ago.
I can't find the article first using the term, but it's like a hidden sponsored article - start by describing a general problem that people might have or something in the technology space, end by advocating for a solution that your company (or the company who paid you to write the post) sells.
Unfortunate to see an otherwise-good article padded by AI-isms ("But here’s the twist: that search I described earlier was never sent to botd. All the history was in SQLite, so I pointed another agent at it and got the analysis anyway. The tool died; the data didn’t.", amongst other examples).
Maisem Ali writes about how they have not stopped engineering despite handing production of code over to AI. I think that's an interesting perspective for a few reasons, but most of all, I'm not sure I agree.
First, we see a prime example of how that can blow up. "botd died this month. It crumbled under its own weight." It's hard to imagine how an agent harness (or is it an overlay on top of other harnesses?) could crumble under its own weight from sheer technical complexity. What makes a tool like this "crumble"? If properly architected, it seems that any outside change could be adaptable. Failing tests could be fixed.
This AI coding evolution has given us the ability to spin up prototypes we don't understand the inner workings of, but I believe for anything of value (and, I would argue, botd appeared to have value to Maisem) we should at the very least understand and influence the architecture and engineering of what we're building. If Maisem had done that for botd, it would not and _could not_ have "crumbled under its own weight". Whatever outside influence required change within botd would have been manageable.
Second, the off-loading of the production of this article indicates to me a lack of respect (with no offence to Maisem, because this is what much of the industry is doing today) for the value in thinking through the writing process. There's a lot that I've learned personally about software I worked on that only came to me through trying to write explainers. Taking a step back from the software we build and looking at it from a birds-eye view, to share how it is constructed or how to use it, helps develop our understanding.
Maisem's article is already very good, and I hope that is demonstrative of their own writing! I would read more! But I want to read words written by a person for other people. AI can help edit, it can help us build ideas, but in code and in our written word, AI should not be _the author_.
Having built a tool in this space, I enjoy seeing all the Calibre alternatives that have been coming out in the last few years! It's fascinating to see the specific wants that people have.
Bookshelf, for example -- what possesses someone to make an ebook library that _only works on object storage_? Is this a weekend project, answering the question of "what would this look like"? Or was it something the author wanted for a long time?
Author here. It mostly came from a personal need rather than trying to answer "what would this look like on object storage?"
I wanted something with as little infrastructure as possible, ideally no database or additional services to maintain, while still having my library available anywhere. Object storage fit that pretty well, especially with things like R2 where the free tier is already more than enough for a personal ebook library.
That said, I don’t want Bookshelf to be tied to one particular setup forever. The plan is to keep the core idea of minimal infrastructure, while adding more storage/deployment options over time.
> Having built a tool in this space, I enjoy seeing all the Calibre alternatives that have been coming out in the last few years!
If there would be just anything actually good. But for some reason they all have just very basic organisation-levels, and barely any support for handling custom data from calibre.
For some reason, streaming ebooks seems to be also not existing. One can either download the whole book, or use some lousy in-browser-reader. So strange how the ebook-world has no decent software for selfhosting.
Kavita has a custom epub library that actually streams the epub pages. The reason why it's non-existent is due to epub being a zip, with a manifest of pages and images. Each page is an html document, so it may have images requested from that manifest.
Most applications just use epub.js which needs the whole book loaded at once.
Edit: Also Kavita does support some calibre tags for metadata and def is much more customizable with a powerful filtering system which you can then bind to your dashboard/side nav.
www.kavitareader.com for those that might be interested (or check the wiki to get a bit more about it).
> what is it that you want from your library management tool?
Custom fields for metadata and accessibility from different devices. Usually there is only hardcoded metadata, tags or categories and one or more systems for collections. And moving data between different software is also quite limited of course.
At the moment I manage everything with calibre on desktop, and use calibre-sync(?) on mobile which has full support for calibres metadata, including the custom fields. But it has no builtin reader and a very unoptimized interface. So the whole handling is very cumbersome, I have to first filter for what I seek, then download it, then change to the reader-app, and maybe organize the file there too, and afterwards I have to delete the local file and manage stuff on desktop like read-state, rating, etc.
Agree so much! It's so interesting all the different use cases and interpretations of those use cases people have.
Like this self-hosted project aims for "simple deployment"/less infra, but I can't imagine how you end up at "use Cloudflare proprietary offerings" and reconcile that with "self hosted".
But if you host the node app yourself and point it at your own seaweedfs/minio/garage that's still more infra than node process+SQLite file.
It's not _theft_ because nothing is _stolen_. The tokens are being used in a way that breaches the contract agreed to by whomever set up the account with OpenAI, Anthropic, Kilo, Antigravity etc. but it's not theft.
Are we reading the same article? That's mentioned once, as a potential alternative, nested in a bulleted list of alternatives. This is not an article about credit card fraud.
This is about people circumventing the model company's attempts to protect their intellectual "property" (which, if you insist on that incoherent usage of the word "property", they themselves stole from the rest of us).
It's equivalent to buying a DVD in the US which is region-locked to Asia. Grey market, not black market. If you use a stolen credit card to buy that DVD, well tat's a totally separate matter.
It's mostly refunding and card fraud because "bad money drives out good" in any marketplace. It's equivalent to getting a refund on an Amazon package you received, which is why a frontier model is being sold at around 2% of the face value.
Right, theft requires that there be somebody who no longer has access. This is about too many people having access. Piracy might be a more fitting term.
Fun idea and a great easter egg, but as a game, the tutorial needs some work. I was so lost for so long and I'm still not sure I understand half of the game!
Thank you for your feedback! I appologize for the inconvenience. But honestly that's not my fault; the spreadsheet app doesn't well support for tutorial :(
As a product that I would regularly use, Pluto missed the mark for me. As a SWE in 2023, I couldn't meet Jacob's mentality. Today, it's how a lot of us ship software.
I didn't love the product but I have to respect the process: he was able to sell his company. For some, what he did is success.
reply