There are some really powerful UIs that are extremely flexible, like spreadsheets, or ComfyUI, or certain issue trackers. On the whole however I would say they all still are mostly their own interaction silos and integrations with other programs have to be explicitly built for each app.
The great thing about a shell is that you can glue together programs that don't know about each other. This is not a silver bullet, with shell programs being text oriented etc., but it does give rise to some powerful possibilities that don't really work with UIs.
Article is AI generated, maybe lightly human-edited.
> TypeScript, using Node.js as the framework and V8 for the execution engine
How this passes review is beyond me - Node.js is not a framework (at least not how the term is used in the industry) and there is not much choice for the engine since it comes with V8.
> That’s a respectable choice for a TUI application
Meaningless AI filler text.
> A project that would have taken a whole team of developers a year or two before agents
It would have taken a single person less than an hour to write a shorter blog post with the same information, before agents.
The downvote made me reconsider the way I posted, which was unnecessarily breathless by saying "How this passes review is beyond me". I will try not to use this style of writing in the future, which, although very human, is very human in a not so good way.
I think I do make a good point however that the article was very likely AI generated since it is difficult to see a human to write like that, and these are not the only examples of AI writing, the article is full of it.
I think it's fair to be critical of AI writing, especially when put out by a big organisation like GitHub. There was a time where you could click on an article published by the tech company whose product you use every day and expect to learn something interesting. It's another sign of the demise of a once great company.
Maybe you're not familiar with Stephen Toub, the author? I'm not saying for sure he didn't use AI at all on writing this piece however he is very well known for the time and effort he puts into his technical writing. Especially the performance improvements posts he writes for every new .net release.
A bit of a tangent, but the linked rsnapshot tool is about 7k lines of perl code. The COPYING file is 20 years old and it's still maintained, the last change was 6 months ago. I'm feeling really old.
Yes, rsnapshot is excellent, I'm glad the article mentioned it. I use it for my weekly backups. So wonderfully simple. I wish the config file weren't quite so strict about syntax, but I only need to change it about twice a decade so no big deal.
This is my attempt to improve my agentic coding workflow by giving the agent an index into the source code through vector embeddings and LLM generated descriptions with local sqlite as the storage backend.
There are many similar tools, but I wanted to learn how such a tool can work by building it myself. What I like about my own version is that I kept things simple - no MCP, no server, no watches, just a CLI that uses git commit hashes to reindex only what has changed.
This reminds me of a video I saw from Alan Watts [1] where he talks about how from childhood it is hammered into our minds that we must work we must do this or do that or we will be punished. And he talks about how you can choose to change your perspective or mindset and see work as a form of play, where you live in the moment.
This is probably harder to do than to say, but this video resonated with me because your mindset is something you do have control over, and choosing not to get stressed about work, and choosing not to take things "seriously" and choosing not to do something for the outcome but just for performing the act, just like you would play music (an example he gave) or to see it as a game is something you can do.
I think the argument is not to not work, but to change the mindset that is so ingrained in our society that you must work. You don't want to starve, so you work, but you can do it with mindset that is fundamentally different, where you choose to focus on being in the moment, and don't let yourself get stressed by outcomes you have no control over, just as you would if you are playing a game or playing music.
Changing your mindset is hard, you are fighting against an entire society where your are expected to be "serious" as an adult and identify yourself with your work, but you do have a choice
At least in the USA, due to the lack of social safety nets I have to disagree. It's very much work or starve. Yeah you can try to change your perspective, but unless your in a privileged situation where you have a safety net (i.e. a wealthy family to support you if things go wrong) then it's difficult to stop thinking about it.
There are substances to let you debug/rewire your brain, but they are usually illegal because if enough people used them, it might threaten those who prefer things to stay they way they are.
> And he talks about how you can choose to change your perspective or mindset and see work as a form of play, where you live in the moment.
I think it's hilarious that they just let me play with a £150k Nexus 9000 stack at work, like it's fucking Roblox or something.
"Here you go EB, figure that one out, all it's doing is connecting our core network with our VM servers and SAN, no pressure! Documentation? Naaah, we never had any of that..."
It was fairly early in my career in IT that I realised that very few other people at the company really understood what it was that I did, or at least that, as long as the outcome was 'good enough' then how I got there didn't matter and no company has the time to have people monitoring _how_ anyone is doing their jobs (outside of where such things are heavily regulated or prevent fatal accidents kind of things).
It's liberating. There seems to be more scope for freedom in what you do than most people think there is.
And, as with many things, as soon as you stop forcing yourself past the point of ease and fluency, you somehow both enjoy it more and also are more effective.
You do have control over it, but the external pressure to be “normal” is enormous and insurmountable for a majority of people. There’d be no such thing as “normal” otherwise.
I look at my cohort who I schooled with, and they are all so terrifyingly sober and adult, and take themselves so very seriously. They toil, they spin, they die.
Me… I don’t really know what happened. As a child, the more I was told to be normal, the more I dug my heels in, not out of contrarianism but out of a still indefatigable conviction that all of these people were wrong.
And the proof lies in the pudding. I am in my 40s, still a child, still stop to explore incessantly, still enjoy poking things with a stick, still finding joy, still being told to just for the love of god be normal, when will you get a proper job, settle down and become a banker or a lawyer or a doctor, and all that jazz.
Maybe it’s just the (until recently undiagnosed) ADHD talking - maybe I have been graced with a surfeit of good luck - but I’m pretty content with how being a cast iron weirdo panned out.
On the remembrance of childhood, much of it is clear to me - and now that I have a toddler of my own, I so often find her doing X or Y and going “good god, I remember being that kid”. Her other doesn’t remember her childhood at all, and struggles to relate - whereas I’ll gleefully fill my pockets with sand and run around hooting Yankee Doodle at full blast. It’s fun. People don’t have enough fun.
> I look at my cohort who I schooled with, and they are all so terrifyingly sober and adult, and take themselves so very seriously. They toil, they spin, they die.
Ah, this is the problem, I work with people who take themselves so terribly seriously all the time.
But they do not take what they do seriously.
I don't take myself seriously at all, which really seems to irritate some people - but I take what I do very seriously indeed.
And then the people paying the wages don't seem to understand why my time is *so* *bloody* *expensive*. Yeah, it's not expensive, it's just a lot of money.
I noticed during my own automatic programming [1] that although code is free (or more free, at least compared to before coding agents) understanding is not. Instead of investing time to understand and master someone else's software tools, I now still need to understand and master what I build for myself, which can be a lot of fun if you build up your understanding from scratch while building out the tools.
I have to spend hours understanding how other humans have asked the computer to do something, and they are also bad at asking it to do so, so I could have gotten a better result if they just came to me with their desire...
> You need audit, security, maintenance and accountability.
I think this is where the role of humans will move towards in the future. Things like accountability can't really be outsourced to AI.
This relates to solving the alignment problem: if you give the AI a specific goal and it has to plan sub-goals, how can you be sure the sub-goals align with your interests.
For tasks in an isloated environment, this doesn't really become an issue. You can give a sandbox the least privileges it needs to do the job you want it to do.
It becomes a problem if you give it open-ended access to systems that connect to the real world and which can have real consequences. There have been stories about openclaw deleting someones email inbox because it though that was what the owner wanted. AI taking control of public wikis to coordinate, which was recently posted is another one.
For software concerns with potential real catastrophic consequences are security, durability, availability. I think for future software systems these are the concerns where you need to limit the AI in a way that it cannot circumvent guarantees that you give around these concerns.
Minor nit:
> AI doesn’t change the question: it creates new choices and moves the thresholds.
This cause an adverse reaction when I read the post. This was probably not written by an LLM since the rest of the article doesn't look like it, but maybe in the future we need to all be more concious about leaving LLM-tells out of our writing.
I think most developers and most dev work is now in the field of websites, apps, and other stuff that is by nature connected to the internet, and has some relatively monolithic backend that has not been built with "blast doors" or other internal safeguards. I guess it will take considerable time until most software is in a state where the impact of AI changes can be reliably estimated without looking at the changes themselves, and some patterns for how to achieve this probably still need to be discovered.
> maybe in the future we need to all be more concious about leaving LLM-tells out of our writing.
I expect the gruesome writing style of LLMs to be fixed relatively soon, seeing how it can already be prevented with just a few lines of stylistic instructions.
I just wonder: how long will it take us then, after the fix, to not be thrown off every time we see some unnecessary catchy or contrastive phrasing?
Being traumatized by poor writing was not on my bingo card for 21st century technological progress ...
> I expect the gruesome writing style of LLMs to be fixed relatively soon, seeing how it can already be prevented with just a few lines of stylistic instructions
What instructions do you use to prevent the gruesome writing style? I have… not found this to be the case.
Currently I append the snippet below to prompts where I care about the resulting text. Just putting it in the context up front isn't as effective, and I assume if the overall prompt is large, or the context window is full, it will also be less effective.
> Use plain, clear, everyday language with a linear deductive flow; avoid hyperbole, juxtapositions, metaphors, analogies, punchlines.
> I expect the gruesome writing style of LLMs to be fixed relatively soon, seeing how it can already be prevented with just a few lines of stylistic instructions.
I have not found this to be the case even with SOTA models as recently as last month.
It holds for a short while into the content but reverts very quickly to its default style.
The default style is so heavily embedded in the weights I don't think simply adding prompts will help.
False. Accountability is the result of difficulty of things. As things get easier you need fewer accountable people.
Like, you need a guy who understands Linux and be accountable for Linux working well in your company. But if Linux does a good job that doesn’t leak, you don’t need a Linux guy.
Who decides whether Linux is doing well enough? Who is responsible for fixing bugs or incorporating new requirements? Who is checking that it follows this month's new regulatory requirements?
As the tasks scales to become more important, any organization needs to have systematic answers to those kinds of questions. And yes, you can use software for that (we always did!), but you still need a human somewhere who is exercising judgment.
I mean with the METR report on the HF attack we saw spontaneous self organization in that model. I'm not saying that currently replaces human judgment but it does appear there are paths to complex AI agents systems that handle a lot of this.
Agreed, as things get easier you need fewer accountable people. Organisations will shrink, people will be more accountable by themselves for things they ask the agent to do instead of having an accountable person do it for them.
On the other hand, I think for complicated software there is such a long tail of things that can go wrong, and in your linux example with disastrous consequences, people will pay good money for a long time to have someone who they can call if there is a problem. In the business they have the term indemnification - that's what companies like redhat provide for linux for example and it's probably a big part of what makes them a billion dollar business.
Of those four, I think accountability is truly the only moat. The other three are varying amounts of both testable and iterated on via adversarial agents steel-manning the implementations.
Accountability will not come before AIs achieve legal personhood, and that probably will not happen in my lifetime. And if it does in some jurisdictions (which I am not betting money on, this is a full full AGI scenario after multiple more philosophical goalposts move first), I will completely bet it will not be a global recognition.
That's really where we're heading, though. We're mostly at the point that, for a lot of code, the humans are there for manual testing and not creating the code.
At some point it will become a strong competitive advantage to zag where others zig.
If you understand the codebase deeply and have trust in it can roll out changes quickly without manual testing you can run circles around the people scratching their heads wondering if this vibe coded 2,500 pull request is going to be the one that takes down the system for a day.
Ive written many significant changes without manual testing by leaning heavily on well crafted automated tests, carefully honed type systems and clean architecture. These foundations serve as the bedrock of high quality applications I can move fast on.
IME vibe coding isnt just not capable of building these foundations, it actively chips away at them where they already exist. shrug
Here is another technique to save tokens: allow the model to read a skeleton of the source code before reading the code, to give it an index into the code so it can read targeted chunks.
There is a tool that uses ripgrep and treesitter that does this [1], adapted from the maki coding agent.
[1]: https://github.com/mizorewww/laya-coreml
reply