Chris Tse

Generative software is poised to make app development 100x faster and 100x cheaper, by letting non-developers create functional apps using AI with just a few prompts.

GPT-5 has good taste. You can steer it away from vibe coding design cliches and make it just as polished as Sonnet 4 / Opus 4.1 for high-end designs.

@sawyerhood A week in AI is equivalent to a quarter in traditional tech. Especially this past week. Whew!

The AI revolution isn't slowing down—it's accelerating. Today's timeline is dominated by Claude 3.7 Sonnet, which developers are calling "the best model in the world for code" with exceptional UI generation capabilities. What's fascinating is how quickly the landscape is shifting. In just three days, we've seen Claude 3.7 Sonnet launch alongside Claude Code (an agentic coding tool), Cursor AI's 0.46 update with unified interfaces, and Perplexity unveiling Comet, their agentic search browser. We're witnessing the rise of agentic AI tools that don't just respond but actively assist in complex tasks. The developer experience is being reimagined at breakneck speed. Meanwhile, in the background: US implementing new tariffs on Canada/Mexico, and Balatro joining Xbox Game Pass. The pace of innovation makes me wonder: are we prepared for how quickly these tools are evolving?

All future software will be built like this. (Spiral is amazing, btw.)

AI: Digital Abundance Crypto: Digital Scarcity Both “artificial”

A generative AI app that generates a drawing app that capture a sketch of an app and then generate the actual runnable app

I made a design tool. 95% of it was coded by Claude AI. svg.designcode.io You can design, animate SVG patterns and export to vectors, png, video or GIF. Use presets, save designs and share urls. Works great on mobile and web. 10k lines of code. The future is insanely fun.

@tarngerine @bnj @sawyerhood Making an app from scratch with a no-code tool like @bubble is still a ton of work and super intimidating to most people. But every company wants custom fields, custom skins, custom forms, custom whatever. So may be real target for AI is "democratized software customization".

@bnj @sawyerhood same, this is my experience pre-llm, too, and my whole issue with the "no code/democratize software creation" pipedream is like... no one actually wants to make Apps (present company excluded)

@sawyerhood @websim_ai Partial re-generation will improve iteration speed. If you just want to re-style the CSS, no need to generate the whole JS shell app. We maintain an AST with Babel continuously within @cardstack. Working to expose more code editing as tool use, so it's more like PATCH than POST.

Things like @websim_ai / Make real / Windows9X are so close but there is still enough prompt engineering / it is expensive af to run / the iteration speed is slow enough that we haven’t hit on demand software yet, but we are getting closer and closer everyday

I’m hoping llama 3.1 is both so fast and good that we will get to the point that software is really just written on demand. Instead of tinkering with notion for task management for your project just describe your workflow and get the custom solution that works for you.

@sawyerhood I have high hopes that open source LLMs like llama will end up creating much, much more open source software than sum of all OSS code in its training set. Most of them will be personal-use software that is never published, but they will exactly what you need, no more, no less.

@Morshiel @cardstack Yes. Built-in AI Assistant integration in the latest internal build. Backed by GPT-4o for now. Will be switchable to support Sonnet 3.5 and other LLMs. This screenshot shows how we generate example data for a newly generated app.

@christse @cardstack Is ai incorporated with cardstack? I’ve held since the ico any exciting news?

Generative software is poised to make app development 100x faster and 100x cheaper, by letting non-developers create functional apps using AI with just a few prompts.

I've written an article that goes into what an AI-native software framework looks like, so creators can extract the deep domain knowledge that is inside the LLM and turn them into working software without coding. You can read more about our approach here: x.com/cardstack/stat…

We have been working on the @cardstack framework, which allows all these full-stack concerns to be expressed in a single Typescript file that AI can easily generate. Data is represented as JSON files and can be composed together via relationships and queries.

Cardstack is the 'Open World Software' platform, inspired by open-world games. It continuously evolves with community-contributed items, functions, and flows in these key areas: 🧩 Open Source 💾 Open State 🧠 Open Intelligence 💸 Open Economy ✨ Open Environment

The best engineers I’ve ever worked with create art with code. They can abstract complex code into simplistic frameworks allowing anyone to read and understand and contribute more on top. AI will replace the mediocre engineers but it will never replace the artists.