Arrival of the AI App Builders

ShippedWith.AI is a weekly roundup of interesting things built atop AI, paired with ideas for entrepreneurs and investors in the space. The AI market moves fast. Our goal is to help you understand and find inspiration in it.

This Week’s Links

Web 1.0 was powered by the LAMP Stack (Linux, Apache, MySQL, PHP). Web 2.0 was powered by stacks like Heroku, Vercel, and Netlify. What’s the stack that powers AI-driven software?

It’s definitely not yesterday’s web stacks, that’s for sure. Everything about the lifecycle, infrastructure, and interactions with AI is too different. It’s also definitely not “just OpenAI” — productized APIs will be an important piece, but not the full whole, of the AI App space.

Enter the AI app builders.

Each offers a different perspective about what an “AI app” is. Looking at how they differ is a great way to see how the AI Stack is evolving — very exciting times.

Here are this week’s links:

Spellbook by Scale - “AI Apps as fine-tuned, task-focused LLM”

Scale focuses heavily on the lifecycle of LLMs: selecting a base model, fine-tuning it with your data, and then deploying a task-specific endpoint atop it, such as prompt completion, text generation, embedding, and so on. The app is the LLM, wrapped in an API that exposes a task LLMs can do.

Sandbox by Cohere - “AI Apps as your existing app, plus our APIs”

Cohere focuses on a suite of client libraries powered by their API. Their app “builder” is the suggestion that you don’t need an app — you just continue developing your own app, and call into their APIs for specific AI operations like vector search or prompt completion. This is essentially the model OpenAI has been pursuing.

Packages by Steamship - “AI Apps as importable software packages"

Steamship focuses on the package management experience. Their app builder is a Heroku-like stack that manages models, API endpoints, and data. Apps published to the platform are called “packages,” and developers can use them by importing client-side stubs. This model focuses on wrapping AI-powered use cases over lower-level AI models.

Disclosure: I’m the founder of Steamship. I promise to only add Steamship to list when it genuinely fits the week’s theme!

Dust by Stanislas Polu - “AI Apps as chained LLM invocations”

Dust is a programming language for LLM completions. It contains metaphors for the operations and strategies that tend to pair best with LLM-based workflows: giving examples in prompts, extracting structure from completions, drilling down with follow-up questions, incorporating in search results, and so on. Apps are executed with the LLM as a JRE-style runtime of sorts.

LangChain by Harrison Chase - “AI apps as composable AI invocations”

LangChain is a framework for wrapping LLM and Embedding APIs and composing those wrappings together into an App. These composable units are called Chains. A Chat Bot Chain might classify your intent and then route you to one of many LLM Prompt Chains that best responds to it. That bot is then itself a Chain, which can be further composed.

Do you have something that should be on this list? DM me @edwardbenson on Twitter.