We’ve been working on exactly this, deriving documentation from external systems (like GitHub, etc) into a giant set of docs that agents can reference and edit. Works way better than I would’ve originally expected. Suddenly these things are able to reference granola notes, a slack discussion, and an RFC when making coding decisions. Super helpful in a lot of unexpected ways!
Every company now has thousands of agents trying to get work done, and almost none of them know what the organization actually knows. To tackle this problem, we built a context engine.
I still think there's some junction between software and the physical world that's the next exciting frontier. There's a whole new world opening up that we've never been able to tackle before. Imagine a world where we can shift all these bright folks from working on ads and SaaS to working on new materials discovery, fundamental sciences, healthcare access, climate change, etc. Working to make the world around us a better place.
We can now build apps with a TAM of 1 or 2 viable at scale - I still have hope there's a lot of good we can do for the world with that new reality. Purpose-built tools for super niche, high-impact problems, where software isn't the end goal, it's just a means to an end.
After a lot of iteration, we found thinking about prototypes built on Lovable, Replit, V0, etc as a communication tool helped clarify where it belonged in our overall process. Curious how others are finding it and where it fits into their flows.
I still feel like we're on track for more and more products to become undifferentiated and degrade in quality over time, if we keep this up. I think we'll rebound the other way at some point and the folks who've always been super valuable, the people who can work up and down the product from customer calls to design to engineering, will become even more valuable over time. Aka, these are the people who can differentiate your company from generic AI
Wow that was a fascinating experience - honestly the biggest revelation for me is how much I skim content without realizing it. Reading and remembering every single word, one at a time, and comprehending what was being said was more effort than I expected! I also took away how much I need the inter-word formatting to keep track of what's being said
Yeah man, I think this exactly captures what I've been feeling - every time I need to get something done, I consistently hit some experience breaking bug. And you're totally right, I've actually just come to expect it and work around it...
Do you think in the future apps that are stable and "just work" will be a moat? Feels like the direction we're trending in.
We've been finding that as engineering becomes less about the code itself, it becomes more about deciding what to build. Which in our experience is limited by creativity more than anything else. What has everyone else's experience of these shifts been?
Man the dichotomy of you have autonomy to now you're a data labeler in a short span of time must be incredibly rough to deal with. How does culture recover after something like that...Anyone have thoughts when this bubble is going to pop? What a bananas time
But will look into this now, see if we can swap some stuff out. We’ve really liked the idea of an offline mirror, makes a lot of collaboration use cases simpler
The surprising part was that “AI-generated prototype” is not a single artifact type. Some tools emitted clean standalone HTML, some produced Vite/React apps that needed bundling and asset inlining, and some depended on hosted runtimes or CDNs enough that they weren’t truly portable. The result was 6 fully offline files, 1 HTTP-served case, and several tools that were effectively not self-contained.
I wish, but maybe it’s more fair to say how people think has changed. It’s like for these complex questions, they can break the problem down quickly into smaller functions, but implementing the functions is slow? Sometimes a non-starter? Idk, it’s different. Our old ways of interviewing aren’t working the same
We’ve added collaboration and communication as big facets in our hiring loop rubrics, and reduced the complexity of our questions. We also tell candidates this ahead of time - been honestly working pretty well so far, it indexes more on those soft skills and gives us just enough insight into the hard skills to make a call.
We’ve also been increasingly finding very few candidates know how to solve a complex question these days without LLM support lol. How are other folks changing their loops these days?
Disclosure: I wrote this as a founder of ProductNow. The last couple of sections highlight what we're building, so I wanted to be upfront about that. Though the broader thesis that software is becoming a utility and the physical world is the next frontier is something I think about all the time, independent of the company. I deeply feel that software is just a means to an end, so can we get back to solving problems that matter, here in the physical world? Curious what this community thinks, especially where the argument breaks down.
The single file is interesting, we’ve been observing something similar too. Do you have a specific prompt that you load in by default to make this work? Are these react files, or just pure HTML/JS/CSS? Aka do you compile it via Esbuild or webpack or something, or are you asking the model to generate something that works out of the box?
We’ve been seeing Claude artifacts sometimes come out as JSX or TSX
What are people doing with prototypes afterward? Do you end up shipping it as is to production? What about at work? Are the prototypes useful in that context?
I think we need to look past software and spend more time and energy on the physical world. Let the digital world just be a means to an end, and we spend more time and resources on problems that matter. Reallocate societal attention to the next frontier
Btw these posts are embedded documents from ProductNow. Been experimenting with more 2-way communication between writers and readers. What do you think? Useful?
Curious how other's have been driving alignment on their teams. We've been finding as AI sped things up, keeping people in sync has been increasingly challenging. We personally have found a doc-driven culture has been super helpful - what are other's trying?
Didn't quite get around to open sourcing the code, but I have some code snippets in there in case someone else is inspired enough to try. Curious what other's think? Seem like a fun genre?
Yeah I like that, there’s probably something to this full stack builder persona that could keep people motivated long term. So long as they feel ownership, seems like that’s a pretty good bet for long term engagement
Curious what other teams are doing to keep encouraging people to think critically about their code? I’ve been finding it harder to keep people motivated, keep them engaged with all the changes coming in. And I can’t blame them, it’s been overwhelming. Is everyone else just using more AI..?
I really think as code becomes cheap, misalignment between people, teams, and organizations is going to hurt a lot more, especially when everyone is trying to move at break neck speeds.
I also think a big piece of this is human attention and inertia. Aka, why bother doing the hard work to coordinate with others when you can just ship whatever you’re thinking. I think whichever organizations can figure out the human and cultural aspects to this will do phenomenally