Repository navigation
Replies: 3 comments 2 replies
|
Hi! This is interesting. There is a tension - yes, of course I am trying to learn what to do to achieve my goals. Like I have read, technical debt accumulates and things suddenly get difficult. I am not a software engineer by training and I am not sure what bar you think I clear (though I'm curios!) but I am a computer scientist (by training and profession) so I have a good foundation to build on. So yes, I'm trying to go about this as responsibly and cleverly as I can, for what it is worth. But I want people to be able to judge for themselves "what it is worth" -- and to a degree yes, I even want to turn them away. People may lose data. This is scary. I don't actually read the code. I am so, so thankful for the friendly community that has grown around Tine. I would rather somebody be turned away "unnecessarily" than to have angry comments from people who felt deceived that this is a hand-built product that I can vouch for more than I actually can. Your phrasings are of course correct, but I worry that they would not as forcefully stop someone in their tracks to ask themselves "do I really want to try this?". But maybe I am overly cautious. I think I will reconsider when I start feeling more positively about Tine again - right now we are on the long detour (which started around mid July) to get to a happy place again -- when that happens, I will update the website and post about Tine in a few places and maybe even consider changing the phrasing ;) Thanks again for your thoughts and encouragement. |
@martinkoutecky I was thinking the same all along. But I do see the merit of communicating this as risky. Speaking for myself, generally, if I know people don't assume stability, I am less afraid to break things and can move more aggressively. Perhaps you feel the same. I also like to see this whole thing as 'an experiment'. I am interested in more of your blog posts and I like your reflections in Discussion section. What I would like to see in addition is more of the inner working, the SKILLs and actual prompting you give to the harnesses. The workflows and etc. But of course this is your daily personal operating cycle, so its up to you how much you open it up. |
|
@Brikas I am not sharing those not because there is anything private going on, but because as I stated before I can't imagine what novel thing I'm doing :D and how to share what it is that I am doing. But I guess the answer is the same as always - I will ask my agents to look into the evidence and write up a post about it. I think one thing that helps is that I have a foundation in complexity theory, and that I have been using open source software on linux for 20+ years, so I have a strong sense of how things should and shouldn't work, and when something doesn't smell right, I ask the agents to investigate. (That is why #543 is taking so long - it turned out to be that "getting sqlite to answer everything" was maybe the right design but was executed poorly, and doing it well is a struggle, etc etc. - but anyway, now I know what is going on.) Keep reminding me, I'll write up something eventually :) |
most of us in software don't know what complexity theory is. This could perhaps be a very interesting blog post "how my background in complexity theory helps me in building better software [using AI]" ? |
Actually, I don't think there is a bar to clear at all. The degree to which a person can build high quality software has very little to do with their degree of formal training. I know graduated software professionals who build shitty products, and I know many people who are very successful at building good software, despite having no formal education whatsoever. I have a friend who was an electrician but learned programming by himself and got hired at Google as an engineer. There's many famous examples as well. Here's a few. The field of software engineering is very different from others such as health, architecture etc where you may only call yourself a doctor or an architect if you meet various formal requirements. In software, you can call yourself whatever, and build whatever, even stuff that handles personal and financial data is built by people without degrees (e.g. facebook, early on anyway) I've been in the software industry for 20 years and I know that there's some successful and useful software out there, despite the code being of bad - sometimes abhorrent - quality. I've seen it and I've contributed to it, despite my degree. Many of today's tools that require data consistency in distributed systems deal with all kinds of problems, even ones that are "production-ready", "battle tested", or written over years by some of the best in the business. For example, Kyle Kingsbury has uncovered many such cases in some of the best databases in the industry. Ironically, it seems Kyle also never had a formal software engineering education I would also add that, just like whether or not someone holds a degree or not has no bearing on their ability to deliver high quality software, whether someone uses AI or not, has no bearing on it either. There's plenty of examples of this. I understand what you're saying about wanting to warm people about possible issues with Tine, however. |
Uh oh!
There was an error while loading. Please reload this page.
https://tine.page/ says this:
Vibe coding is often understood to mean the worst way to develop software using AI: prompting an agent to produce anything that seems to be working well enough without any regard for rigor, quality, reliability, maintainability, security, etc.
Meanwhile, when you look at how Tine is actually developed, even though it uses AI, it is done so rigorously in accordance to many engineering best practices. This is quite clear, looking at the various artifacts such as the ADR's, technical discussions, release notes, etc. I found this post particularly compelling.
So I think by describing it as "vibe coded" you are needlessly inviting negative presumptions, and I suggest to frame it a bit differently. Like maybe you can say you use agentic engineering and use various processes and safeguards to meet a certain level of quality, without saying how high that level is (e.g. probably don't say "better than manually written code", even though I think that would be true for certain components compared to other projects)
Also, I saw you mention somewhere you're not a software engineer, I think you have definitely met the bar for software engineer :-)
All reactions