Fabric has evolved into something closer to infrastructure, analogous in spirit to AWS, but running on everyday devices instead of centralized data centers. For decades, modern software has run on centralized infrastructure. Even with the rise of AI, that has not meaningfully changed. Large language models, training pipelines, and inference systems still rely on massive, centralized data centers. The tooling has evolved, but the underlying assumption has not: compute lives in a few places, owned by a few providers. That model works, but it is not the only future. As smaller, more specialized models emerge and jobs become more modular, we are starting to see a shift away from monolithic infrastructure toward systems that are distributed, composable, and flexible by default. That shift is where Fabric lives. ## How Fabric Evolved When we first introduced Fabric, we described it as a distributed compute network. Over the last few weeks, it has become clear that this description, while accurate, was incomplete. Fabric has evolved into something closer to infrastructure, analogous in spirit to AWS, but running on everyday devices instead of centralized data centers. Not a single product. Not a narrow tool. A foundational layer that supports many capabilities built on top of the same primitives. The abstraction that makes this possible is the job. ## Jobs Are the Atomic Units A job is the smallest meaningful unit of work on Fabric. Each job has: - Defined inputs - Deterministic execution - A concrete output - No persistent state It runs once, completes, and disappears. Examples include: - Fetching a URL - Rendering a page - Generating embeddings - Running LLM inference - Transcoding video - Simulating agents - Probing a website from a real consumer device Jobs are stateless, parallel, composable, and infrastructure-level primitives. Today, Fabric supports more than 20 job types. Each can run independently. They are used to offload specific pieces of work: training models, processing data, running simulations, or performing inference. But jobs alone are only half the story. ## From Atomic Units to Applications Atomic units do not exist in isolation. When you start combining jobs, sequencing them, storing their outputs, adding configuration and history, you do not just get a bigger task. You get an application. This raises a key question: _How do you preserve the power of jobs while making them usable by more people?_ The answer is Fabric Applications. ## Fabric Applications Fabric Applications are persistent, user-facing capabilities built on top of jobs. They: - Orchestrate multiple jobs - Maintain configuration and history - Expose functionality through a UI - Can be reused over time **Applications do not replace jobs. They consume them.** Jobs remain the atomic units of Fabric. Applications are how those units become accessible, especially to people who do not want to reason about distributed systems. This distinction matters because it allows Fabric to serve two audiences at once: technical users who want low-level primitives, and non-technical users who want outcomes. Without forcing either group to compromise. ## Why Distribution Matters Fabric Applications are possible because the infrastructure is distributed. Applications built on Fabric can: - Run anywhere - Be embedded directly into websites or apps - Avoid managing servers or clusters - Cost significantly less to operate Users do not need to understand nodes, scheduling, or compute routing. If you can sign in, configure a few settings, and click deploy, you can use Fabric. The complexity stays in the infrastructure, where it belongs. ## The First Fabric Application: AI Assistants The first Fabric Application is AI Assistants. In under five minutes, anyone can: - Train an assistant on their own data - Configure its behavior - Deploy it directly to their website We have tested it extensively. It works, and it works well. If you are an ecommerce company with an older website and you have been told that adding an AI assistant will take months and thousands of dollars, Fabric changes that equation. | With Fabric Applications | Benefit | | --- | --- | | Setup Time | Minutes, not months | | Cost Savings | Typically 70-90% lower (depending on usage) | | Infrastructure | No Custom infrastructure required | This is not the first chatbot ever built. But it is one of the first to run on a distributed compute network, making it cheaper, more flexible, and easier to deploy. A sales and lead qualification person inside a company can build AI Assistants in Fabric in 5 minutes. Instead of a static "Contact Us" form, they now become 100x more powerful using assistants to: - Answer product questions - Qualify leads before human involvement - Route users to the right plan or offering - Capture structured intent data An ecommerce company can build an AI assistant on Fabric Applications for stores with large catalogs. Assistants help customers: - Compare products - Find items by intent ("gift for X", "under $Y") - Understand differences between SKUs - Navigate size charts, materials, and compatibility In all of these, the Fabric Applications (in this case the AI Assistant) do not live on Fabric. They live on the user's website. Yes, we provide the very simple integration to do this. **AI Assistants are not special.** They are just one example of what happens when: - Atomic jobs exist - Orchestration is abstracted - Persistence and UI are layered on top ## Data Science Machine Learning Models Another Fabric Application is Data Science Machine Learning Models that are fully trained, hosted, and inferenced on Fabric. If you have the data, you can train the model using any of our 20+ algorithms, without writing any lines of code. ## Image Machine Learning Models The third Fabric Application is Image Machine Learning Models. With any set of images, you can do full training, labeling, and inference on Fabric. Fabric gives you a direct link to your ML Model and you can go ahead and commercialize it. ## Opening Fabric to Everyone Jobs are powerful, but they are not always approachable. Applications are how Fabric opens itself to: - Operators - Analysts - Founders - Teams without engineering resources All without writing code or understanding distributed compute. If jobs are the atoms, applications are the elements they form together. ## What Comes Next Jobs remain the foundation. They are used today to offload specific, high-leverage tasks: training, inference, simulation, scraping, processing, without changing how systems work. **Applications are how Fabric compounds that power.** By packaging multiple jobs into persistent, usable abstractions, Fabric makes distributed compute accessible to anyone, not just engineers. The infrastructure stays the same; the surface area expands. What changes is not what Fabric can do but who can use it and how quickly they can turn it into power for their dreams. **Fabric is not evolving into a single product or a closed ecosystem.** It is becoming an infrastructure layer where products are assembled from atomic units (jobs), and where entire companies can be built without owning servers, clusters, or clouds.