Reviewing AI Platforms: Part 1
There is so much AI tooling available it can be difficult to categorize everything comprehensively. Many have overlapping functions, so it’s not always clear exactly what a service is designed for. In the following segments however, you’ll see a good number of services and how you can use them based on your needs. In the previous lessons, you went through the more hands-on route to understand the internals of modern AI applications. The first category you’ll learn about is code-first AI platforms.
Code-First AI Frameworks
These frameworks offer low-level access to various tools necessary to build AI applications. They can be very involved, requiring lots of learning and expert knowledge to be able to build advanced and robust applications. Developing with these tools certainly takes more time, as does maintaining them. But they offer the most flexibility and the ability to build highly efficient, complex AI apps at scale. You can also build simple apps with these tools.
Visual LLM Orchestrators
These are effectively a visual IDE for LangChain. They’re a step above the code-first tools. They’re simple to work with and much faster than writing code. With these, you simply connect nodes along edges and define logic where necessary. They use much of the same terminology as you’ve seen in the previous lesson. With a basic understanding of how to build based on the graph structure, that is, nodes and edges, you’ll be able to drag-and-drop till you’re done with your app.
Examples of such services are LangFlow, Flowise, and Dify.ai.
AI Workflow & Logic Orchestrators
AI agents are arguably the second most used feature of AI applications besides chats or conversations. Thus it’s no surprise that there are a lot of services designed to build agentic workflows. Agentic workflows boost productivity, and are fun to work with.
These AI workflows are apps used to automate complex business processes, integrating various other apps to handle mundane tasks. Common workflows integrate with apps like Slack, Notion, Google Drive, Salesforce, and Spreadsheets using visual builders, APIs, and no-code interfaces. They can access data, extract data, generate data, and make smart decisions.
They’re much easier to work with than the code-first tools you saw earlier. Their main weakness lies in flexibility, and ability to build advanced applications with high efficiency. Not all services are the same. n8n, for instance, involves a fair bit of code depending on what you’re building. Also, these services can become expensive and complex as you build more advanced workflows with them.
Examples of these workflow builders are Zapier, n8n, Make, Gumloop, Lindy AI, Workato, Activepieces, Pipedream, and Microsoft Power Automate.
AI Agent Builders
These are specialized platforms for creating AI agents and chatbots. They have limited use cases and can be restrictive for complex apps. They allow you to build AI-based apps that automate tasks, perform research, converse with users, and deploy and test AI agents. Some have visual builders and can sometimes be similar in many ways to the AI workflow apps described above.
Examples are Voiceflow, Botpress, Stack AI, and Relevance AI.
Full-Stack AI App Builders
While AI builders and workflow platforms allow you to build custom AI apps, these platforms allow you to build custom native apps, such as web and mobile apps, using AI. They don’t offer visual builders but instead provide text boxes for accepting prompts. You describe the app you want to build in detail, sit back, and watch them build it for you.
These are actually one of the most popular kinds of AI platforms out there besides the AI workflow and agent builders. There are a few important things you need to know about them. The platform you choose, the prompt, the model, and the extra features and refinements available determine the quality of the app that will be built. It’ll take some trial and error to get used to them and some refinement to build a good app.
Since you can use this to build more sophisticated apps than the others, you’ll also need to test the apps thoroughly, to ensure they do exactly what you’d want, before going live with them. You may want to strategize and build them in parts, depending on the kind of app you’re creating. Also note that the more features your app requires, the more tokens you’ll use, and the more refinements you make—thus, the more expensive the service may become.
These platforms offer some level of flexibility in that, while you control the outcome through prompts and parameters, you don’t control every implementation detail. However, you can often make modifications since you’ll have access to the generated code. This means some programming knowledge may be required if you want deeper customization, though it’s not mandatory for basic usage.
They also offer customizability in that some, like rocket.new, offer integration with tools like Figma, which allow you to import any designs in Figma, and then build the app based on the design. Other customizations are in the form of choosing a specific stack for the build. You can choose to build an Android app with Flutter or go native with Java or Kotlin. Or you may decide to build a website with Svelte or Vue, it’s up to you.
Many of these services have little to huge features that make them easy to use and make you more productive. Some have templates to easily get you started, comprehensive guides, generous free tiers with lots of tokens per month, excellent customer support, multiple tool integrations, MCP servers, and so much more.
Examples of these platforms include Replit, Rocket, and Hercules. Here are some example prompts you may use to build an app.
Sample prompt
Act as a Senior Kotlin Multiplatform Developer. Build a fitness tracking app using Compose Multiplatform that targets both iOS and Android.
Project Structure & Tech Stack:
Use the commonMain source set for 100% of the UI and business logic.
UI: Use Compose Multiplatform (CMP) for all screens.
Networking: Use Ktor for API calls.
Local Storage: Use SQLDelight or Room (KMP version) for local persistence.
Architecture: Follow the MVVM pattern with ViewModel (from Jetpack libraries) in the shared module.
App Features:
Dashboard Screen: A central hub showing 'Active Minutes' and 'Calories' using a progress ring component.
Workout History: A list view with smooth scrolling (ensure display-native framerates on iOS).
Platform Integration: In iosMain and androidMain, implement a platform-specific expect/actual function to fetch the device's battery level.
Design Requirements:
Use Material 3 components with a minimalist 'Deep Sea' blue theme.
Ensure native iOS behaviors like swipe-to-back and platform-appropriate text input.
Example prompt for SaaS dashboard:
Act as a Senior Full-Stack Developer. Create a web-based Project Management Dashboard using Next.js and Tailwind CSS. The app should feature:
Kanban Board: A drag-and-drop task management interface with columns for 'To-Do', 'In Progress', and 'Done'.
User Roles: Implement role-based access where Admins can manage tasks and regular Users can only update their own assigned items.
Reporting: A page with monthly task completion charts generated using Recharts.
Integrations: Connect the app to a PostgreSQL database via Prisma for real-time updates.
Auth: Include a secure login page with 'Sign in with Google' and email/password options.
In the next segment, you’ll learn how to build an AI workflow using one of the most popular tools today, n8n.