Startup tldraw has released "computer," an experimental application that lets users create AI workflows using natural language commands and visual blocks on an infinite canvas.
According to tldraw founder Steve Ruiz, the system leverages the company's Infinite Canvas SDK to build a dynamic environment for working with AI models. The application uses components representing different elements like text fields, images, and audio clips on a canvas. Users connect these components with arrows to show how data flows and transforms between them.
Each component has associated procedures—instructions that execute based on input from connected components. Components can receive data from multiple sources and send their output to many others, including back to themselves.
When users run "computer," information flows from component to component, with each generation's output becoming input for the next step. This creates processes that can branch, loop, and iterate through multiple generations.
Video: tldraw
For example, when users type "Write a short commercial" into an instruction component, the system generates a reusable JSON script within seconds. This script can transform any combination of inputs into an advertising script, which then generates a second request to the language model for the final output. Users can display this output directly on their canvas or send it to other components to generate speech or images.
AI-powered workflow generation
One of the system's most impressive features is its ability to build workflows automatically. When users make high-level requests like "Create a marketing campaign based on this product description," the system figures out all the necessary steps and components on its own, turning a simple prompt into a complete workflow.
Under the hood, "computer" runs on Google's experimental Gemini 2.0 Flash model. tldraw says they picked Gemini because of its speed and versatility - it can handle everything from text and images to structured data, and its standardized output format makes it easy to work with.
A new approach to programming
According to Ruiz, "computer" represents the first step toward a future where people can program visually using everyday language. Instead of learning complex coding languages, users can create sophisticated workflows just by describing what they want to accomplish.
The platform works for both experienced developers and newcomers, with a library of tutorials and example projects covering everything from basic calculations to recipe generation and image analysis. Once users create something useful, they can save it as a template and share it with others.
While "computer" is still in its experimental phase, tldraw plans to refine it based on community feedback. The project follows the success of the company's previous AI experiment "Make Real," which let developers design software through drawings rather than code.
The timing seems right for this kind of tool. Major tech companies are already working on similar natural language computing projects, with Anthropic developing Claude Computer Use, Google building Project Mariner, and Apple creating Apple Intelligence. These efforts suggest that 2025 could be a pivotal year for natural language computing.
