Redefining App Building with AI at the 2025 AWS Summit NYC

Julia Gervasi

At last week’s AWS Summit in New York, we saw a new contingent of companies, both established and brand-new, show us that AI not only augments our ability to crank out code, but it also enables us to build entire systems from ideation to deployment. 

Science fiction has taught us that AI will take over the world and replace humans, but this year’s AWS Summit in NYC flips that narrative on its head. 

Empowering the Masses

Instead of replacing humans, AI is now building tools to empower everyone, from skilled developers down to teenage sci-fi geeks, to create new apps and workflows by using simple prompts, a few configuration files, and an idea.

At the Summit, the core of this shift was clearly AgentCore, AWS’s new modular framework for building autonomous, goal-driven agents. And these aren’t your average chatbots, either. These are complex, flexible systems made with a modular design: components for memory, tool use, reasoning, and observability all designed to be swapped, extended, and reused.

Agents built on AgentCore don’t just process language - they understand context through both short-term memory (via prompts and conversation) and long-term memory (stored in vector databases like S3 Vectors or OpenSearch). They can make real-world decisions using secure API calls through Bedrock’s tool interface, execute Python code in a safe sandbox, browse the web mid-task, and log every action through built-in observability dashboards powered by CloudWatch.

And best of all, they’re designed for real-world use. Developers can customize each layer, from memory storage to reasoning modules. And non-developers can launch and use pre-built agents without ever writing a line of code.

But what we’re seeing now seems to be a wholesale embrace of interoperability. Vendors working together, or at least willing to interoperate with each other, to allow folks to build systems that are generally distributed, modular, and even vendor agnostic. 

To support this, AWS introduced the AI Agent Marketplace, which is a centralized hub where you can find agents built by top partners like Anthropic and IBM, ready to deploy. Each one is modular, configurable, and audit-ready. You can chain agents together through EventBridge or Step Functions, modify their behavior with override prompts, and trace every step they take with full visibility.

Imagine spinning up a document review agent that watches your Box folder, extracts data from contracts, highlights anomalies, and emails your team summaries — all without managing infrastructure or fiddling with machine learning models. Just connect your services, type what you need, and go.

Making Semantic Search Easy

But one of the most transformative launches was S3 Vectors, a vector embedding store baked directly into Amazon S3. With vector embeddings implanted directly into Amazon S3, you can drop in a file like a PDF, a CSV, even a plain text doc and immediately turn it into a searchable, semantic memory layer, embedded using Bedrock models. 

From there, agents can query your data using natural language, retrieve specific insights, and generate outputs using retrieval-augmented generation (RAG), all with zero setup beyond uploading the file.

Though RAG isn’t anything new, the integration of these technologies into a marketplace and drag-and-drop platforms makes agents smarter, more context-aware, and more useful for anyone to use.

While Bedrock provides foundational models like Claude 3, Mistral, and Cohere, AWS has been focusing on Nova, which is their in-house foundation model family, now fine-tunable with just a few clicks through SageMaker.

Developers can personalize how agents speak, behave, and reason and apply instruction tuning, adapter layers via LoRA, and closed-loop evaluations to simulate different use cases or personas.

Combined with AgentCore’s modularity and S3’s native memory, this creates an entirely new development stack: one in which you don’t need GPUs, custom infrastructure, or ML expertise. You just need a clear vision of what you want to build and a good prompt.

A Shift in Building Apps

This year’s NYC Summit wasn’t just a tech showcase. It was a paradigm shift. We’ve moved on from “AI will help you write code”, to “AI will build whatever you imagine.” AI agents are opening the door to a new way of automation.

The intelligence is shared.
The tools are open.
The agent works with you.

You don’t have to be a developer to build anymore. With AWS’s tools, a social worker, a middle schooler, or a sci-fi geek could spin up a fully functional agent that automates tasks, communicates in natural language, and improves over time. You’re no longer just the end user. You’re the architect. And the agent is your co-pilot.