12 September 2026
How do we think about generative UI at Desert Labs?
How we tackle creation and improve agent capabilities in making generative UI for complex automations while keeping it custom to every customer.
The learning curve problem
How people view software is changing day by day. They want custom, simple, and easy-to-understand UI instead of having to learn a new piece of software. However, many SaaS companies bloat their apps with indecisive flows, customer-specific feature flags, and a failure to generalize the product properly, which creates a huge learning curve to get started. We believe generative UI, along with the right primitives, can solve this problem.
What generative UI can mean
At one end of the spectrum, generative UI could mean making a new page on every request; at the other end, it could mean making a unique page once per customer and iterating on it based on the user's request.
How we simplify UI while being able to make complex automations on our platform
At Desert, people describe their task, and we build it on the fly; this includes backend code and UI for the automation. We give the agent core primitives such as a design palette, a component library, page templates, and templates for interacting with external services like emails, messages, etc. So when a user explains their automation to an agent, we understand how they want to interact with it and how they want to see the results. Based on the automation and these two factors, the agent generates a UI that lives in an external service, and we present it via an iframe within the Desert app.
Serving the right UX
This sounds simple, but there's still a problem: how do we serve the right UX to the customer? I don't think any company has solved this yet, including us. These are some of the ways we are deciding to tackle this. Track each request where a user prompts the agent to fix UX issues; these could include flow changes, content display, layout, and more. We audit each one every week using agents and update either the primitives or the skills/prompts.
How much freedom to give the agents
Another thing I worry about is how restrictive the primitives are, since it's inversely proportional to the freedom we give the agents. If we believe we can achieve AGI, we want these agents to have as much freedom as possible to do what they think is right based on the situation they're in.
Our goal at Desert is to remove the learning curve so you can build the best possible software to automate your work! If this sounds interesting, book a meeting with us and get started on Desert!