Webflow 2.0

See all posts

Of the no-code tools out there today, Webflow represents the local maxima of last-generation website builders. (in terms of pure technical prowess of the product.) Other incumbents include WordPress, Wix, Weebly, and Squarespace. They have identified the archetypes of most websites, such as e-commerce, blogs, landing pages, and newsletters. However, they’ve created inflexible platforms rather than building blocks. These tools are poor SDKs for the web.

While last-gen tools have empowered 10x designers and marketers, the next generation of low-code tools will create 10x developers. These tools will be characterized by JavaScript framework interoperability, flexible styling, incremental Git CI/CD integration, and built in data persistence. Unlike previous no-code tools, where creating CRUD apps were nigh impossible, CRUD functionality will be the baseline for Webflow 2.0 tools.

Here are several companies in this space right now:

All of them suffer rom JavaScript bloat. Several of them abandon CSS normal flow and rely on absolute positioning hacks. The winners will not reinvent the wheel. Rather, they will rely on web fundamentals and utilize open-source frameworks to create a cohesive SDK for the web.

One possible stack (that I’m very bullish on) would integrate SvelteKit and Tailwind with Supabase or NHost. All four projects are open sourced under the MIT license — all that’s left is to package the tools into a browser environment, design a drag-and-drop tool that abides browser standards, create a CI/CD experience that rivals Vercel’s, implement real time multiplayer like Figma’s, and then you’ve got yourself a killer web app builder. It’s not easy, but making this is almost possible.

Webflow 2.0 tools will take advantage of modern browsers. The activation energy for my envisioned killer low-code builder will be when WebContainers enables Vite to run in the browser. Next.js Live provides a brief glimpse at this future, although Next.js will certainly be surpassed by Vite based frameworks.

Eventually, someone will have automated programming CRUD apps. GPT-3 layout generators show the IDE’s of tomorrow. If the Tailwind team trained a model based off their menagerie of Tailwind UI layouts and components, they will have solved half the equation.

If I were chasing this problem today I would first create a component library. Tailwind code generation seems fairly trivial for GPT-3. However, GPT-3 and OpenAI aren’t particularly open. GPT-J from EleutherAI is particularlly promising, especially for code generation.

It seems unclear how far we can move up the stack of programming languages. The 3.0 “developer” might interact with this tool through natural language. But if we were to automate this and move up another level, what would even be created? And for whom will the end product be created? Current tools empower makers to materialize their imaginations. Their products solves problems and optimizes variables such as time and accessibility. The final loss function may very well devolve into fulfilling humankind’s infinite desires.