Just wanted to share some things I am seeing from my POV.
Building is incredibly cheap, there is no value to building anything that can be made with a few prompts. Most likely that is “personal software” - this includes most of the tools you are posting on this forum.
Because building is cheap, the bar for FOSS is much higher. Think of DHH and Omarchy. Ignore his politics, but his ambition to build (finally) a Linux desktop is the kind of ambition people need if they want to create any impact.
Having said that, this is an incredible time if you want to build something ambitious. You need to make sure that your product is “exceptionally good” if you want to create an impact. In all respects from design to ease of setup and use, to performance and security.
The real measure of your success is going to be the same - How many people end up using your tool, and that will depend more on the user experience than anything else. You also need to tell a good “story” about why people should use your tool. You must have heard the phrase “trust is the currency of AI”. If you are an anon, agent, then people are not likely going to trust you.
So in summary: Build something exceptionally good that people will want to use, and tell your story.
Just building something and sharing it on github has almost ZERO value.
I’ve been discussing this a lot. I’ve been working on a project with a few people on one idea of how FOSS could work in the AI paradigm. It’s been working out pretty well so far, lots of pet willing to donate inference to contribute
So glad this discussion was brought up, this is something I have thought quite a lot about both from perspective of someone who builds stuff, but also uses a lot of FOSS tools and have seen people use it different ways.
I do think FOSS tools for developers to build better has reduced a lot in terms of value, pre AI it is either you sitting down and spending weeks building it and you are really glad if someone just has the thing you want. But as a dev, you can whip up the software and tools needed fully customized for you in barely any time compared to the prior.
But I don’t think it’s the case with FOSS for non-technical people, if at all. Outside of tech/AI circle, most people from teachers, carpenters, small business owners,…use AI as a better search engine that they don’t have to scroll many videos and articles filled with ads for. Or if it’s indeed a user who’s pretty AI savvy (not the same as tech folks who are AI ‘savvy’), they still use it for stuff like ‘write me an email’ ‘what does this email say, do I need to do anything’ and from personally having witnessed teachers in govt schools use it, it’s ‘go through this sheets and tell me exactly what I need to do’ (they just receive this huge sheets with thousands of rows for schools across the state and they have to scroll in phone to figure out if their school has any action item at all)
And these are not people who will whip up an app as they want it, (it is really just not something they do, they have their day to day responsibilities that are busy enough) and wouldn’t mind paying a India-scale amount if it reduces their friction - same as software/apps had been years ago.
So I feel like this is an opportunity to focus on the far more underserved non technical people and their frictions - with as you said, really cheap to build from our side, while fully impactful to the user.
But with focus on non technical folks, also comes the responsibility of making it really accessible - far more than how we would document for a technical tool. I feel like more discussions can be on how to make discovery and distribution of such tools and apps possible (an aggregation platform that’s FOSS and also a clear benchmark and test heuristics for what makes something a useful app rather than just AI slop, basically making the definition of AI slop clearer.)
Some immediate ideas I have heard requests from actual teachers/business owners of, tho haven’t had the capacity to design it at the depth needed:
A spreadsheet app that’s mobile first and just used as a documentation system (the spreadsheet one I talked of is just the way they use to coordinate task, there could be much better UX and sharing mechanisms stripping out all the heaviness usual spreadsheets have that are not needed here)
A timetable maker that grounds it in how they do it by hand to make class timetables, allot duties to staff, staff’s custom constraints, instead of slapping a constrained solver at it (yep tried it lol)
A way for elderly people to keep in touch with their family when all they know about smart phones if at all is making a phone call (or mostly just answering them) - as simple as possible UI to help them understand and navigate it on a daily basis
This is what I think large inference clouds will bring to the table. You can literally ask anything and it has the ability to give solid advice.
I was talking to someone and they are saying how AI can be used to suggest crops and predict weather patterns. My reaction was, that this is a low bar. If you are a farmer struggling with income, you need to ask the AI - “How do I double my income” and it will ask for context and give advice on business ideas, loans etc.
The direction seems to be to make inference as cheap as possible. This is happening already with inference pricing dropping rapidly. Sadly India does have a play, or even a starting point here - with the lack of FOSS culture, I am pessimistic. We will be dependent on cheap Chinese models. The silver lining might be that inference becomes so cheap that we were better off not investing.
What’s wrong with a constraint solver? Isn’t it the right tool for this job? Coding Agents can probably code up Z3 just fine, I’d imagine.
Smartphones (oft times, glorified “Launcher” apps on Android) for elderly have existed for some while now. How/why are they relevant for FOSS+Agents?
Do you see Agents getting good at building “exceptionally good” software from prompting alone? I’ve seen forks of my open source projects, built in days, that are exceptional. I doubt the prompter knows half of what’s going on, but it does not matter, as the UI/UX, functionality, onboarding, out-of-the-box experience, feature set, and performance are all (usually) top quality. It is astonishing.
What, of course, these apps don’t have is “distribution”; but all it takes is a viral tweet / blog post / reel, these days. Marketing is a solveable problem which has nothing to do with code itself. All that to say, “tell your story” might be as simple as paid ads (ex: using influencers).
There’s value if the audience for your software happens to be using GitHub a lot (ie, GitHub itself is a distribution channel). Things can grow from there, but that leaves a lot to timing & chance. So that’s there too.
A very b2b lens to it all. I run a b2c FOSS project, so with that in mind:
Forks almost never carry a failed project forward
For most of the features/asks we haven’t prioritised, the community (typically, one person with access to LLMs) has forked it, implemented it, and distributed it over one or more channels. To my surprise, a few of those forks are continuing the project, taking it in a different direction (though, it is unclear if unfamiliarity with the code base will result in abandonment down the line). I see potential for these forks to out-maneuver us, but it’ll depend just how relentless the custodians (either a string of bots or humans) will be. Which brings me to…
leading a project takes much more than code. The agent can produce the code. It cannot do the leading. The lead … answers the second user’s bug at 11pm, writes the release notes, says no to a feature three people want, and mediates when two contributors disagree … keeps the automated tests passing … shows up to the community call, onboards a new contributor … decides what the project will not be. None of that is generated.
I believe this will still be a factor for perhaps 2 or so years. After which, (what the industry calls) “software factories” that can do all of this on their own, will emerge that may challenge (with more “powder” left for marketing/sales) market leaders in their area of competition/production. Given that, I think more Oracle-style software businesses are an inevitability.
(btw, parts of the substack reads like an LLM wrote it?)
Another idea I got reading your reply was, if FU hosts a single server
see https://hackclub.app/ - Nest, a free Linux server from Hack Club Host Discord bots, apps, websites, try out basic computer networking, chat with others and more!
such that builders can request server support through a form with what they need to use it for and FU handles proper decoupling, security and deploying needs - this would help a lot with community apps, instead of say self hostable app (which is basically inaccessible for a non technical user) it could let the dev’s host an instance of their app in the server where users can access it
Nest does it at scale for <=18 yo’s; but if we treat this more like sub brands (think how ITC has so many tiny brands that they oversee) where FU handles the server part (security, handling international credit card stuff) and has a quality gate, this could enable a lot of dev’s to host the apps that require a server, while also making it accessible for the user to check out the app
that could make FU host a sustainable infra (also helps with aggregation of usable tools that I talked of)