Week 1 was about getting familiar with the fundamentals no complex agent logic, no database, just understanding how the Vercel AI SDK works and getting Athena's foundation in place.

By the end of the week the chat interface was working, text was streaming, basic tools were wired up, and the documentation system was live. Small wins, but everything that comes after builds on this.

It was also the week I realized how much thinking goes into building an AI product before writing a single line of agent logic.

Concepts Learned

A breakdown of the core concepts I encountered and studied during the week not deep dives, just enough context to understand why each one matters and how it fits into building Athena.

Text Streaming : How the Vercel AI SDK streams responses token by token from the model to the UI instead of waiting for the full response

useChat() hook : How it manages messages, API calls, loading states, error handling and conversation memory out of the box

Tool Calling : How to define tools that the model can decide to call during a conversation

MDX : Writing documentation with Markdown that supports React components inline

The Vercel AI SDK : Understanding the overall structure of the SDK, how the frontend and backend connect, and how requests flow from the UI to the model and back

Features Implemented

Chat Interface : The main conversation UI where users interact with Athena

Text Streaming : Responses stream in real time instead of waiting for the full response

useChat() Integration : Replaced manual message handling with the Vercel AI SDK's built in hook

System Prompting : Defined Athena's personality, behavior and thinking process through a system prompt

Basic Tool Calling : Wired up initial tools that the model can call during a conversation

MDX Documentation System : Set up the docs with MDX support for writing interactive documentation

Errors & Challenges

No major blockers this week. The one issue I ran into was TypeScript errors when trying to access the response from a tool call. The Vercel AI SDK returns tool results in a specific shape and TypeScript will complain until you type it correctly took a bit of digging to figure out the exact structure expected.

What's Next

Week 2 is where Athena starts feeling less like a chatbot and more like an actual agent. Here's what's on the radar:

Structured Outputs : Expanding tool calling to return typed, structured data instead of plain text responses
More Tool Calls : Building out more tools that give Athena a more defined reasoning flow
Agentic UI : Updating the chat interface to reflect the structured outputs and make the experience feel more purposeful

Resources

Vercel AI SDK Docs : The main reference for everything covered this week. If you're following along, start here.