Introduction

The rapid evolution of AI tools has transformed how we build software. Model providers like OpenAI and Anthropic offer powerful large language models (LLMs); libraries such as LangChain help orchestrate complex LLM-driven workflows; cloud platforms from AWS to Vercel make deployment easier than ever. Yet one piece of the puzzle has lagged behind: the user interface. How do we present these AI capabilities to users in a seamless, interactive way? This is where Generative UI (or GenUI) comes in Thesys positions itself within the modern AI developer ecosystem as I note in What is Generative UI. Thesys is pioneering the concept of an AI-driven frontend, providing the UI runtime infrastructure for AI-native software. In this blog, we’ll explore how Thesys compares to and integrates with other leading tools (LangChain, Vercel, Retool, OpenAI, Anthropic, and more) and map out a mental model of the AI stack that clarifies Thesys’s role. By the end, you’ll understand how C1 by Thesys, the Generative UI API, serves as the missing link that turns LLM outputs into live, dynamic user experiences.

The AI Tooling Landscape: From LLMs to User Interfaces

To see where Thesys fits, it helps to visualize the AI development stack as a set of layers or components, each with distinct roles:

Foundation Models (Brains of the AI): At the base are the LLMs and other generative models themselves model providers, offering APIs that developers call to get AI results. However, these services typically return text (or other data) and don’t concern themselves with how results are presented to the end-user.
Orchestration & Agents (Logic and Reasoning): Building on the models are orchestration libraries and agent frameworks. Tools like LangChain help developers chain prompts, handle context, call external tools via function invocation, and build LLM agents with complex behavior. This layer is about backend logic
User Interface Layer (Presentation and Interaction): This is where Thesys comes in. Traditionally, developers would use frontend frameworks (React, Angular, etc.) or UI builders (like Retool for internal tools) to craft the user interface. But those approaches require manual design and coding of every screen and form. In an AI-native context, where the AI’s responses and capabilities can vary widely, static UIs become a bottleneck. Generative UI flips this dynamic by letting the interface generate itself in response to the AI’s outputs and the user’s needs. Thesys provides the runtime infrastructure for this generative UI layer .
Infrastructure & Deployment (Hosting and Scaling): Underneath or alongside all these is the deployment platform that host the application and deliver it to users. For example, a developer might host their AI-powered web app on Vercel. Thesys is designed to integrate within these environments: you can run the generative UI frontend as part of a web application and deploy it just as you would a traditional React app. In other words, Thesys doesn’t replace your cloud infrastructure; it works with it. You might still use Vercel (or another platform) for serving your app, while C1 by Thesys API powers the dynamic UI within that app.

This landscape can also be thought of in terms of a stack or pipeline. Data and user input flow into the model layer (LLMs), the AI reasoning layer handles logic (chains/agents), and then the output goes to the UI layer where it reaches the user. Thesys occupies that topmost layer AI UI layer

Generative UI: A Paradigm Shift for AI-Native Software

So, what exactly is Generative UI? In simple terms, it’s a new approach where the user interface isn’t fully predetermined by developers, but is instead created on-the-fly by an AI as I discuss in What is Generative UI. A Generative User Interface means the layout, components, and interaction elements can be generated or adjusted in real time based on context, user requests, or AI outputs, rather than being hard-coded upfront. This concept marks a shift from static designs to dynamic UI with LLM guidance.

Think of it this way: traditionally, if you build a dashboard or form, you decide ahead of time which charts, buttons, and fields are on the screen. In a generative UI approach, you might simply ask the AI for what you need, and the interface will materialize to deliver that. For example, imagine asking an AI assistant, “Show me the sales analysis for last quarter.” Instead of just getting back a paragraph of text, an AI powered by generative UI could build UI with AI in response builds itself for you, guided by the AI understanding of your intent. Every user could get a custom, real-time adaptive UI that’s tailored to their query or task, rather than everyone using the same one-size-fits-all screen.

This isn’t science fiction LLM UI components make it possible: developers define a set of allowable components (charts, text blocks, input forms, maps, etc.), and the LLM’s output can include instructions to instantiate those components. Under the hood, the AI might output a structured specification (like JSON or a function call) saying, in essence, “display a chart here with these data points” or “create a form asking for X, Y, Z.” A rendering engine (like Thesys’s runtime) takes that and creates actual UI elements in the application. In effect, the LLM-driven product interface is born from a combination of the model’s output and a library of UI building blocks provided by the developer. This approach has been compared to giving the AI “LEGO pieces” of interface that it can assemble as needed.

The implications of this paradigm are huge. It enables real-time adaptive UI that can personalize itself to each user’s needs and context. Interfaces become AI UX tools that optimize themselves: an AI agent can decide the best way to present information or gather input, rather than forcing the user to adapt to the software. Early experiments in the field show what’s possible. OpenAI’s ChatGPT, for instance, introduced function calling in 2023, which allows the model to return JSON data or call a tool instead of just text LangChain framework, known for chaining LLM calls, has added support for streaming outputs as React components, essentially letting a language model populate parts of a web UI dynamically. Open-source projects have demonstrated “AI agents that take control of your application” by rendering custom React components based on LLM outputs. Even traditional UI builders are incorporating AI assistance frontend automation driven by AI is becoming a reality. We are moving beyond chatbots that just talk LLM agent user interface systems that can draw charts, create buttons, and truly interact with users through a rich UI as I discuss in What is Generative UI.

Crucially, generative UIs aren’t about flashy graphics for their own sake; they solve real problems. They make software more user-centric. Instead of a rigid UI forcing every user through the same workflow, the interface can adjust to what the user is trying to do in that moment. This leads to significant benefits: personalization at scale (every user’s interface can be a bit different, tailored to them), context-awareness (the UI responds immediately as the situation evolves), and increased engagement (users get visual feedback and interactive elements that match their needs, rather than walls of text or complicated menus). From a developer’s perspective, it also means less time spent painstakingly coding every possible dialog or screen dramatically speed up development cycles. Early adopters of AI-driven frontends have reported cutting their UI development time substantially and delivering features faster (Louise, 2025). In fact, some enterprises are already using internal generative UI tools (with LLMs under the hood) to automatically generate form pages and dashboards, handling 70freeing human developers for more critical tasks (Louise, 2025). The bottom line is that generative UI isn’t just a cool demo

Thesys: The UI Runtime for AI-Native Software

Thesys was founded to be the infrastructure that makes generative UI possible for any development team. Its flagship product, C1 by Thesys, is a Generative UI API that provides the building blocks and runtime environment for AI-generated interfaces. In essence, C1 by Thesys is an AI frontend API UI runtime layer) that interprets an LLM’s output and turns it into actual UI elements in your application.

How does this work in practice? Developers integrate C1 by Thesys API into their app (for example, a web app or dashboard). They also give the LLM knowledge of a set of UI components that it can use Chart, Table, Button, Form, etc. Then, when the user interacts with the AI (say through a chat prompt or some command), the LLM’s response can include a special structured payload invoking those components. C1 by Thesys catches that and renders the corresponding UI in real time. Importantly, Thesys stays model-agnostic: you can use OpenAI, Anthropic, or open-source models muscle and bones of the interface. It integrates with the LLM via standard protocols (like function calling outputs, JSON parsing, etc.), so if your AI can produce the right structure, Thesys can turn it into a live interface.

One way to think about Thesys is as the runtime engine for AI-native frontends. Just like a game engine takes game logic and renders graphics on screen, Thesys takes AI logic and “renders” an interactive UI for the user. This is fundamentally different from a tool like Retool. Retool and similar GUI builders (Bubble, PowerApps, etc.) are about making it easier for humans to assemble interfaces (often via drag-and-drop). Thesys, by contrast, enables the AI to assemble the interface. This doesn’t mean developers lose all control “build UI with AI”, which is especially powerful for AI-driven applications where the ideal interface is not known ahead of time.

A key strength of Thesys’s approach is integration and flexibility. Since C1 by Thesys is delivered as an API and works with common front-end frameworks (Thesys provides a React-based renderer, for example), you don’t have to rewrite your whole stack. You can call C1 by Thesys from your backend or incorporate it in your frontend, and host your app wherever you normally would Vercel for a frictionless deployment, or on your own servers for full control. Thesys is not a hosting platform; it’s an infrastructure component (What is Generative UI). In fact, many teams using Thesys treat it as a layer within a larger application: the AI brain might run on OpenAI’s cloud, the logic orchestrator might be a Python server using LangChain, and the UI is powered by Thesys within a React app delivered via Vercel. This composability means you can integrate Thesys with existing tools rather than having to choose one over the other.

What about security and consistency in a UI that changes itself? Thesys addresses this by allowing developers to enforce constraints on what the generative UI can do. Because UIs are generated using predefined components, you’re not letting the AI arbitrarily code a new frontend from scratch (which could be risky). Instead, the AI’s choices are constrained to known-good elements. You can also apply design systems or style guides, so even though the content and layout may vary, the look and feel stays on brand. This approach yields a balance between dynamic UI and reliable user experience. It’s analogous to providing a sandbox for the AI: it can build anything within the sandbox using the toys available, but it can’t go outside the bounds. For developers, this means adopting generative UI via Thesys doesn’t mean chaos

Already, the traction for Thesys suggests this UI layer was badly needed. According to a recent InfoWorld piece, over 300 teams were already using Thesys tools by April 2025 to design and deploy adaptive AI interfaces (Krill, 2025). These range from startups looking to add intelligent UI features, to enterprises seeking to transform internal dashboards. By providing the “glue” between LLM outputs and on-screen components, Thesys has carved out a unique niche: it’s not a model provider, not just a dev tool library, but truly an AI UI platform. In the broader ecosystem, Thesys stands out as one of the first dedicated solutions for LLM-driven user interface creation. Just as we have databases for handling data, and servers for handling logic, Thesys wants to be the standard for handling the front-end for AI agents and applications.

Integrations and Comparisons: Thesys and the Modern AI Stack

Let’s directly address how Thesys compares or connects to the other major tools in this space:

OpenAI, Anthropic, and other Model APIs: Thesys is complementary to these. You will typically use an LLM from OpenAI, Anthropic, Cohere, or open-source in conjunction with Thesys. The model provides the intelligence and content; Thesys provides a way to display that content interactively. In fact, C1 by Thesys can leverage features like OpenAI’s function calling to better integrate UI generation. For example, you might define a function like create_ui(component_spec) that the model can invoke to produce a UI element show what that model can do in a richer way. Think of model providers as giving the brains, and Thesys giving the canvas on which the brain’s outputs appear.
LangChain and Orchestration Frameworks: Many teams build complex AI applications with orchestration frameworks like LangChain, LlamaIndex, or Haystack. These tools manage calling multiple models, handling retrieval of information (RAG), and implementing agent behaviors (tool use, planning). Thesys can work hand-in-hand with such frameworks. For instance, LangChain’s output could be an object that includes both a text answer and a suggestion for a UI element (LangChain even documented an example of building an LLM-generated UI (LangChain, 2023)). Instead of printing to a console or returning only text, you plug Thesys into the final step of your LangChain pipeline. In practice, this might mean writing a LangChain agent that, when it wants to present results, formats them for C1 by Thesys. Conversely, you could use Thesys to create a more engaging front-end for a LangChain-powered chatbot or agent: the agent does its reasoning in the backend, and whenever it produces an output, Thesys displays it nicely (be it a chart, form, or just a well-formatted response). The key point is that Thesys is not trying to replace your AI orchestration; it’s giving that orchestration a UI layer. If we compare to a traditional stack, LangChain is like your controller logic, and Thesys is the view. They serve different purposes and integrate via simple contracts (like passing JSON UI specs or using API calls).
Agent Frameworks and Tools: There are many emerging “agent” frameworks (AutoGPT, BabyAGI, etc.) and domain-specific AI tools. These often focus on autonomy “agentic UIs”
UI Builders and Frontend Frameworks (Retool, etc.): It’s also worth comparing Thesys to traditional UI development aids. Retool, for example, is a popular platform for rapidly creating internal dashboards by dragging and dropping pre-made components and hooking them up to data sources. Retool is fantastic for quickly making an admin panel or a form without much coding. However, it’s inherently a manual, human-driven design process UI is generated by AI based on user context and prompts. This means Thesys is suited for scenarios where the interface might need to change frequently or handle a wide variety of tasks that weren’t all known beforehand. For example, if you’re building an AI dashboard builder that lets end-users ask for any chart or report ad hoc, a tool like Retool wouldn’t easily support that level of on-demand flexibility dynamic and AI-driven. In many cases, it could even embed within a static interface: imagine a Retool dashboard that has one panel powered by Thesys, where an AI can conjure up sub-interfaces or assist the user within that panel. That kind of hybrid is possible because Thesys runs in standard web environments. In summary, compared to UI builders, Thesys is less about low-code and more about no-code (for the end-user)
Backend Infrastructure (e.g. AWS): Finally, consider how Thesys relates to platforms like Vercel, which are often mentioned in the same breath. Vercel is a cloud platform optimized for frontend developers on Vercel. In fact, a likely setup is: your frontend is a Next.js app using the C1 by Thesys library for rendering generative UI components, and you deploy that on Vercel for the global, fast CDN and serverless functions to handle the AI calls. Vercel itself has been exploring AI integrations (with their SDKs for AI and examples of deploying LLM apps), but those still leave UI rendering up to the developer. Thesys plugs into this gap, so you could use Vercel’s infrastructure and still have an AI-driven UI via Thesys. Another way to see it: Vercel covers where your app runs, Thesys covers how your app adapts its interface at runtime. They are complementary layers in the stack. The same goes for other cloud services

By integrating with all these layers, Thesys essentially completes the picture of an AI-native application stack. You have your AI brain (LLM), your reasoning engine (chains/agents), your runtime UI (Thesys), and your deployment platform. Each is specialized, and when combined, you get an application that can think, act, and interact with users in a truly intelligent way. Few other products squarely address that UI interaction piece. Some companies are nibbling at the edges UI runtime for AI is a distinguishing factor. It treats the UI layer as just as dynamic and important as the rest of the AI stack.

Conclusion

The emergence of generative AI has led to an explosion of tools for building with LLMs, but it’s also highlighted the need for a new kind of frontend. As we’ve discussed, the modern AI developer ecosystem spans everything from low-level model APIs (OpenAI, Anthropic) to high-level orchestration frameworks (LangChain, agent systems) and robust hosting solutions (Vercel, cloud platforms). Thesys enters this landscape as the missing UI layer that bridges AI and users. It doesn’t try to replace your model or your logic

By providing a Generative UI platform, Thesys allows developers to build AI-native software that feels truly intelligent end-to-end. Users of such applications get UIs that evolve in real time, providing the information or input controls they need when they need them. I talk about in What is Generative UI that developers, in turn, can iterate faster, offloading a chunk of UI creation to the AI and focusing on higher-level application logic. The mental model to keep in mind is a layered stack: models -> logic -> generative UI -> deployment. Each layer has its experts, and Thesys has firmly planted its flag in the UI layer for the AI era.

The broader implication is that software design is shifting. We no longer have to design every pixel for every scenario. Instead, we define the toolkit and let the AI generate interfaces on the fly. This can make software more adaptive, personalized, and scalable than ever before. But it also requires robust infrastructure to work reliably what to do, Thesys helps show what it did in a user-friendly way; OpenAI gives your app smarts, Thesys gives it a dynamic face; Retool speeds up humans building UIs, Thesys enables UIs that build themselves via AI. These are complementary pieces of the next-generation software puzzle.

Final Thoughts

Thesys is at the forefront of this new frontier of Generative UI. As the company behind the world’s first Generative UI API (C1 by Thesys), it provides developers with the tools to turn any LLM into a savvy UI creator. If you’re envisioning AI applications with LLM-driven interfaces or wondering how to generate UI from a prompt, Thesys offers a powerful solution. It’s an invitation to imagine interfaces that aren’t static, but rather dynamic UIs with LLMs working behind the scenes to serve each user. To learn more about Thesys and see C1 by Thesys in action, check out Thesys and the Thesys documentation. As AI continues to reshape software, Thesys is there to ensure that your frontend can keep up

References (APA/MLA)

Deshmukh, Parikshit. Generative UI: When Your Interface Builds Itself, Just for You. Thesys, 8 May 2025.
Deshmukh, Parikshit. Building Frontends with AI: From LLM UI Components to Dynamic Dashboards. Thesys, 4 June 2025.
Krill, Paul. “Thesys introduces generative UI API for building AI apps.” InfoWorld, 25 Apr. 2025.
Thesys Introduces C1 to Launch the Era of Generative UI (Press Release). Business Wire, 18 Apr. 2025.
Louise, Nickie. “Cutting Dev Time in Half: The Power of AI-Driven Frontend Automation.” TechStartups, 30 Apr. 2025.
Tarbert, Nathan. “Build Full-Stack AI Agents with Custom React Components (CopilotKit + CrewAI).” Dev.to, 28 Mar. 2025.
LangChain. “How to Build an LLM Generated UI.” LangChain Documentation, ver. 0.3, 2023.
Firestorm Consulting, "Stop Patching, Start Building: Tech’s Future Runs on LLMs." Firestorm Consulting, 14 June 2025.

FAQ

Q: What is Generative UI in simple terms?
A: Generative UI refers to user interfaces that are created or adjusted by AI in real time, rather than fully designed by humans in advance. In practice, it means the software’s UI can change dynamically based on context, user requests, or AI outputs. For example, an AI-driven app might generate a new chart or form on the fly to respond to a user’s question. This is different from traditional UIs, which are static and only change when developers update them.

Q: How does Thesys differ from using a normal front-end framework like React?
A: Thesys isn’t a replacement for React (in fact, it works with React)

Q: Can Thesys work with any AI model?
A: Yes. Thesys is model-agnostic. You can use it with mainstream models like OpenAI’s GPT series, Anthropic’s Claude, Google’s PaLM, or open-source models like LLaMA, as long as you can get the model to output the right format for UI instructions. Many models can be guided via prompting or fine-tuning to produce JSON or function-call outputs that Thesys’s runtime can understand. Essentially, as long as your model can describe the UI it wants (within a schema you define), Thesys can render it.

Q: Is Generative UI only useful for chatbots or does it have broader application?
A: It has broad applications anywhere AI and user interaction meet. While the concept gained attention through AI assistants (chatbots that could show rich responses), it’s equally useful for data analysis tools, business intelligence dashboards, form-driven applications, and more. Any scenario where different users might need different interface elements, or where you want the software to adapt to unpredictable user queries, can benefit from generative UI. For example, an internal tool at a company could use generative UI to build custom data entry forms on demand. A customer-facing analytics app could let users generate new visualizations via natural language queries. It’s not limited to chat

Q: How do we ensure the AI doesn’t create a bad or broken UI?
A: The developer remains in control of the building blocks. With Thesys, you define a set of components that the AI is allowed to use. The AI isn’t writing raw code (which could break things); it’s selecting from your predefined components and feeding them data. This means the AI can’t introduce something completely outside the rules you set. Additionally, you can validate the AI’s UI output

Q: How does Thesys compare to simply using something like ChatGPT’s function calling?
A: ChatGPT’s function calling is a great new feature that lets the model return structured data or call an external function. Thesys can actually leverage that

Q: Is Generative UI production-ready, or just an experimental idea?
A: It’s becoming production-ready very quickly. While the concept is new, we’re already seeing real products and enterprise tools built on generative UI principles. Thesys’s own technology is being used by hundreds of teams as of 2025 to deploy live applications, which indicates a level of maturity. Of course, best practices are still evolving, and developers are learning how to design good prompts and UX around AI-generated interfaces. But the building blocks (like C1 by Thesys, component libraries, etc.) are here today. Like any new technology, it should be adopted thoughtfully

Related articles

May 2026 Security Incident

Why We're Open Sourcing OpenUI

AI Copilots: A Complete Guide to Intelligent Assistance

What Are AI Agents: Types, Capabilities, and Real-World Use

ChatKit: What It Is, What It Isn’t, and Where C1 Fits In

PowerPoint Automation Reimagined with C1 Artifacts API

The Future of AI Interfaces is Generative

MCP UI: Making AI Apps Interactive

What Is AG-UI and Why It Matters for AI-Native User Interfaces

How to design AI-Native Conversational Interfaces : From Templates to Generative UI

GPT 5 vs. GPT 4.1

How to Build Generative UI Applications

Implementing Generative Analytics with Thesys and MCP

Evolution of Analytics: From Static Dashboards to Generative UI

Why Generating Code for Generative UI is a bad idea

Building the First Generative UI API: Technical Architecture and Design Decisions Behind C1

How we evaluate LLMs for Generative UI

Generative UI vs Prompt to UI vs Prompt to Design

Generative UI: What It Is and Why It Matters