Forget No‐Code. The Future is All‐Code (Thanks to LLMs)
As the general population becomes more familiar with LLMs, we hypothesize that code will become more accessible and software functionally ‘no-code’ for any point on the technical spectrum. Open-source software disproportionately benefits from this if one designs the library for LLM accessibility.
No-code tools have exploded in popularity over the last decade (e.g. Gartner report). Shopify, Zapier, and Webflow democratized access to technology, letting non-developers perform web operations (set up an eCommerce shop, perform integrative workflows, and design websites) that were previously restricted to developers. By letting more users access these tools, workplace productivity dramatically increased and so we saw corresponding adoption via B2B SaaS (Zapier, for example).
But no-code came at a cost: limited customization. At Roundtable we would often start using a no-code tool and then switch to a full-code solution when we needed anything other than a boilerplate solution.
A similar trend can be found even more recently on the other end of the spectrum, with many examples of LLMs increasing developer productivity (GitHub copilot, Cursor, Claude/OpenAI, see spend data on Ramp). These gains are ultimately about easier accessibility. Programmers don’t need to load as much into their brain context to code anymore and can query using partial code and/or natural language to automatically generate code chunks.
These two trends snowball on top of each other: there is demand for more flexible no-code tools, and LLMs provide an incredibly accessible API that enables large customization. Open-source software may have a competitive advantage in this space because an LLM can store the open-source library in context (or learn the library during training) and let users code with a natural language API.
We are testing this hypothesis at RoundtableJS, an open-source JavaScript library for survey programming. Our cloud offering includes a code editor preloaded with the library, letting users program and design surveys using natural language (link to video). This approach bridges the gap between the flexibility of custom coding and the accessibility of 'no-code' platforms. A non-technical user could request, "Create a multiple-choice question about customer satisfaction with a light blue color scheme," and the LLM will generate the RoundtableJS code.
A fun question to wrestle with is how do you design an ‘LLM-friendly’ library? We have spent quite some time doing so, but it is admittedly a work in progress. Making the library as customizable as possible – in terms of styling, logic, and features – maximizes the chance that these users have a good experience and get the survey they ask for. We also designed it so that the survey is completely defined in a single script, reducing the complexity and increasing the speed of AI updates. As we collect more data on how people use the AI editor, we can start to optimize the library to fit users’ query semantics. In the long run, we believe this approach will make RoundtableJS ultimately accessible to users across the technical spectrum.
As LLMs become more prevalent and powerful, the distinction between 'code' and 'no-code' will blur. Users will be able to harness the full customization capabilities of programming languages without needing to fully master coding syntax. These trends may provide an especially strong tailwind for open-source projects. To build software for the future, you may want to build open source and for LLMs.